[Samba] Performance is limited by design on larger servers?
I have a fairly large server with 16 CPU, 8 Gb ram and just four HDD's. While the disks are pretty untouched most of the time, three samba processes are working for dear life. Sometimes going up to 100. These processes are running LDAP, SMB and Kerberos where LDAP seem to be the one mostly utilized. If i could spread the load onto more processes this server would be able to handle a lot more requests. Is there any way to do this? //danileh -- With best regards, Daniel Hedblom Sysadmin Department Barn och Skolförvaltningen Municipality of Sollefteå Phone: +46 (0) 620-68 22 02 Mobile: + 46 (0) 70 383 72 44 -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] low samba performance with glusterfs backend
Hello folks, We test samba performance with local ext4 and glusterfs backends, it shows performance is very different. The samba server has 4 1Gbps NICs and bond with mode 6, backend storage is raid0 with 12 SAS disks. A LUN is created over all disks, make as EXT4 file system, and used as glusterfs brick. On the samba server, use dd test local ext4 and glusterfs, write bandwidth are 477MB/s and 357MB/s, details as follow. When mount cifs on Centos 6.0, write bandwidth downgrade to 184MB/s and 117MB/s, very different. Using 4 win7 clients to test with SANergy/Iometer, cifs over ext4 aggregate throughput can get about 391 MB/s. However, cifs over glusterfs aggregate throughput only can get about 180 MB/s. Why CIFS performance is so slow over glusterfs backend? Anybody meet such issues and how to resolve(samba and glusterfs parameters)? thanks a lot in advance. #test local ext4 [root@pana53 5f4554cf-3b56-43e5-847e-cb409b0edc30]# dd if=/dev/zero of=dd.dat bs=1MB count=10k 10240+0 records in 10240+0 records out 1024000 bytes (10 GB) copied, 21.4764 s, 477 MB/s #test local glusterfs [root@pana53 61659a17-69bd-4704-81fe-0853f1c891cd]# dd if=/dev/zero of=dd.dat bs=1MB count=10k 10240+0 records in 10240+0 records out 1024000 bytes (10 GB) copied, 28.7148 s, 357 MB/s #test cifs over ext4 [root@pana53 ext4-share]# dd if=/dev/zero of=dd.dat bs=1MB count=10k 10240+0 records in 10240+0 records out 1024000 bytes (10 GB) copied, 55.5549 s, 184 MB/s #test cifs over glusterfs [root@pana53 glusterfs-share]# dd if=/dev/zero of=dd.dat bs=1MB count=10k 10240+0 records in 10240+0 records out 1024000 bytes (10 GB) copied, 87.4899 s, 117 MB/s here is smb.conf: [global] server string = samba server map to guest = Bad Password log file = /var/log/samba/log.%m max log size = 50 max protocol = SMB2 socket options = TCP_NODELAY IPTOS_LOWDELAY idmap config * : backend = tdb aio read size = 16384 aio write size = 16384 use sendfile = Yes posix locking = No [ext4-share] comment = None path = /data/5f4554cf-3b56-43e5-847e-cb409b0edc30 read only = No guest ok = Yes [glusterfs-share] comment = None path = /reexport/61659a17-69bd-4704-81fe-0853f1c891cd read only = No guest ok = Yes here is gluster volume info: [root@pana53 glusterfs-share]# gluster volume info Volume Name: vol1 Type: Distribute Status: Started Number of Bricks: 1 Transport-type: tcp Bricks: Brick1: pana53:/data/5f4554cf-3b56-43e5-847e-cb409b0edc30 Options Reconfigured: auth.allow: *,192.168.1.* features.quota: on nfs.disable: on BR, liuben nuaa_liu...@sina.com -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] Performance problem using clustered samba via ctdb
Hello, I recently set up a samba cluster with 4 nodes using ctdb. The systems are virtual Citrix xen machines running SuSE SLES11Sp2 with samba 3.6.3. The shared filesystem needed for ctdb is on a ocfs2 share stored on a ISCSI target. The cluster is running fine and ip takeover etc is working fine as well. To find out how the cluster would performe in real life with many clients accessing samba shares I compiled smbtorture (from samba4) to run the nbenchmark test using the loadfile client.txt from the dbench4.0 distribution. What I found out is really strange: I first tried to simulate 50 clients on one of the cluster nodes: $ bin/smbtorture //host1/smbtest1 -UUNIKO/smbtest1%password bench.nbench --loadfile=dbench-4.0/client.txt --num-progs=100 -t 30 The result is an average throughput rate of 50MByte/sec. Ok do far. Now I distributed the 100 clients on all four nodes by starting an smnbtoture with 25 clients on each of the cluster members: $ bin/smbtorture //host[1,2,3,4]/smbtest[1,2,3,4] --num-progs=25 -t 30 The throughput results for the four hosts are now: 4.4 MBytes/sec, 4.6 MBytes/sec, 5.2 MBytes/sec and 2.8 MBytes/sec If I add more clients by increasing the --num-progs-parameter rates drop further down. On one node probably the master I see that all three (virtual) CPU core have a system load of 60% (from top). The other three nodes do not show any high CPU load. I also ran the ping_pong test (ping_pong /shared/cluster/test.dat 5) on the shared filesystem. On one node I get a value of about 36000. If I run the very same ping_pong-command on all four nodes I get a value of 1000 on each node. On our old samba servers we have a total of about 400 connects distributed on two servers. However if I try to put such a load (4x100) on the four new samba cluster nodes via smbtorture the test won't even start. If i put 400 clients on one of the servers it works just fine. Now I ask myself two questions: 1. Is the nbenchmark kind of realistic test? 2. Why do throughput rates drop as much as I found out and is this a known behavior of ctdb or is my configuration somehow bad? Any ideas? Thanks Rainer -- Rainer Krienke, Uni Koblenz, Rechenzentrum, A22, Universitaetsstrasse 1 56070 Koblenz, http://userpages.uni-koblenz.de/~krienke, Tel: +49261287 1312 PGP: http://userpages.uni-koblenz.de/~krienke/mypgp.html,Fax: +49261287 1001312 -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Very slow samba performance on Centos 6
Hi, Jeff! Something for you to reply to ... :-) Volker On Thu, Aug 04, 2011 at 02:11:35PM -0400, vg_ us wrote: -- From: Volker Lendecke volker.lende...@sernet.de Sent: Thursday, August 04, 2011 11:01 AM To: vg_ us vg...@hotmail.com Cc: samba@lists.samba.org Subject: Re: [Samba] Very slow samba performance on Centos 6 On Thu, Aug 04, 2011 at 10:49:50AM -0400, vg_ us wrote: I have 2 identical Dell r510 servers with 10gig card, running centos 6 with samba-3.5.4-68.el6_0.2.x86_64. I setup 16G ramdisk samba share on both and ran cp from local ramdisk to samba ramdisk mount. If I cp 12 1-gig files, I get combined 100MB/s transfer rate. Single file cp maxes out at about 15MB/s. Ftp transfer give me over 300MB/s. Running with 9000 MTU. Most smb.conf is default. I even disabled atime and tried ext2 and xfs on ramdisk. Any help will be greatly appreciated. What client application are you using? If it is a cifsfs kernel mount, you might see such artifacts. Please retry with the smbclient(1) application. If that is also slow, we need to investigate further. I re-ran some of the tests with following result: Ftp ramdisk-to-ramdisk: 13572 MB, 32.8 secs - 413.8 MB/s Ftp ramdisk-to-hardisk: 13572 MB, 62.8 secs - 222.4 MB/s Smbclient ramdisk-to-ramdisk: 13572 MB 40 secs - 339 MB/s Smbclient ramdisk-to-harddisk: 13572 MB 64 secs - 212 MB/s cifsfs mount ramdisk-to-ramdisk: 13572 MB 289.8 - 47MB/s cifsfs mounts are really slow, so what happens when linux, windows and mac clients map/mount the share? Are they gonna be this slow? Any way to speed it up? Thanks - Vadim -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Very slow samba performance on Centos 6
Wouldn't it be better to rerun these tests, not from the Ramdisk, but from a network connection to more closely resemble what the results will be when in a production environment? Doing such tests years back did show that FTP will typically be faster than Samba, due to the difference in overhead costs. Samba isn't a service like FTP, it has to negotiate SMB packets, interpret the requests/commands and then communicate that to the system it is running on. I haven't played with CIFS, but I imagine that it to would have a similar or potentially greater overhead than Samba itself. -- Regards, Robert Adkins II -Original Message- From: samba-boun...@lists.samba.org [mailto:samba-boun...@lists.samba.org] On Behalf Of vg_ us Sent: Thursday, August 04, 2011 2:12 PM To: volker.lende...@sernet.de Cc: samba@lists.samba.org Subject: Re: [Samba] Very slow samba performance on Centos 6 -- From: Volker Lendecke volker.lende...@sernet.de Sent: Thursday, August 04, 2011 11:01 AM To: vg_ us vg...@hotmail.com Cc: samba@lists.samba.org Subject: Re: [Samba] Very slow samba performance on Centos 6 On Thu, Aug 04, 2011 at 10:49:50AM -0400, vg_ us wrote: I have 2 identical Dell r510 servers with 10gig card, running centos 6 with samba-3.5.4-68.el6_0.2.x86_64. I setup 16G ramdisk samba share on both and ran cp from local ramdisk to samba ramdisk mount. If I cp 12 1-gig files, I get combined 100MB/s transfer rate. Single file cp maxes out at about 15MB/s. Ftp transfer give me over 300MB/s. Running with 9000 MTU. Most smb.conf is default. I even disabled atime and tried ext2 and xfs on ramdisk. Any help will be greatly appreciated. What client application are you using? If it is a cifsfs kernel mount, you might see such artifacts. Please retry with the smbclient(1) application. If that is also slow, we need to investigate further. I re-ran some of the tests with following result: Ftp ramdisk-to-ramdisk: 13572 MB, 32.8 secs - 413.8 MB/s Ftp ramdisk-to-hardisk: 13572 MB, 62.8 secs - 222.4 MB/s Smbclient ramdisk-to-ramdisk: 13572 MB 40 secs - 339 MB/s Smbclient ramdisk-to-harddisk: 13572 MB 64 secs - 212 MB/s cifsfs mount ramdisk-to-ramdisk: 13572 MB 289.8 - 47MB/s cifsfs mounts are really slow, so what happens when linux, windows and mac clients map/mount the share? Are they gonna be this slow? Any way to speed it up? Thanks - Vadim -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Very slow samba performance on Centos 6
- From: Robert Adkins II radk...@impelind.com Sent: Friday, August 05, 2011 10:10 AM To: 'vg_ us' vg...@hotmail.com; volker.lende...@sernet.de Cc: samba@lists.samba.org Subject: RE: [Samba] Very slow samba performance on Centos 6 Wouldn't it be better to rerun these tests, not from the Ramdisk, but from a network connection to more closely resemble what the results will be when in a production environment? Doing such tests years back did show that FTP will typically be faster than Samba, due to the difference in overhead costs. Samba isn't a service like FTP, it has to negotiate SMB packets, interpret the requests/commands and then communicate that to the system it is running on. I haven't played with CIFS, but I imagine that it to would have a similar or potentially greater overhead than Samba itself. -- Regards, Robert Adkins II -Original Message- From: samba-boun...@lists.samba.org [mailto:samba-boun...@lists.samba.org] On Behalf Of vg_ us Sent: Thursday, August 04, 2011 2:12 PM To: volker.lende...@sernet.de Cc: samba@lists.samba.org Subject: Re: [Samba] Very slow samba performance on Centos 6 -- From: Volker Lendecke volker.lende...@sernet.de Sent: Thursday, August 04, 2011 11:01 AM To: vg_ us vg...@hotmail.com Cc: samba@lists.samba.org Subject: Re: [Samba] Very slow samba performance on Centos 6 On Thu, Aug 04, 2011 at 10:49:50AM -0400, vg_ us wrote: I have 2 identical Dell r510 servers with 10gig card, running centos 6 with samba-3.5.4-68.el6_0.2.x86_64. I setup 16G ramdisk samba share on both and ran cp from local ramdisk to samba ramdisk mount. If I cp 12 1-gig files, I get combined 100MB/s transfer rate. Single file cp maxes out at about 15MB/s. Ftp transfer give me over 300MB/s. Running with 9000 MTU. Most smb.conf is default. I even disabled atime and tried ext2 and xfs on ramdisk. Any help will be greatly appreciated. What client application are you using? If it is a cifsfs kernel mount, you might see such artifacts. Please retry with the smbclient(1) application. If that is also slow, we need to investigate further. I re-ran some of the tests with following result: Ftp ramdisk-to-ramdisk: 13572 MB, 32.8 secs - 413.8 MB/s Ftp ramdisk-to-hardisk: 13572 MB, 62.8 secs - 222.4 MB/s Smbclient ramdisk-to-ramdisk: 13572 MB 40 secs - 339 MB/s Smbclient ramdisk-to-harddisk: 13572 MB 64 secs - 212 MB/s cifsfs mount ramdisk-to-ramdisk: 13572 MB 289.8 - 47MB/s cifsfs mounts are really slow, so what happens when linux, windows and mac clients map/mount the share? Are they gonna be this slow? Any way to speed it up? Thanks I did include Smbclient ramdisk-to-harddisk (that's reading from local ramdisk filesystem and copying to samba share sitting atop of real hard drive) test. Cifsfs mount with real disk gives the same performance as included cifsfs mount ramdisk-to-ramdisk: - Vadim -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] Very slow samba performance on Centos 6
Hello all, I have 2 identical Dell r510 servers with 10gig card, running centos 6 with samba-3.5.4-68.el6_0.2.x86_64. I setup 16G ramdisk samba share on both and ran cp from local ramdisk to samba ramdisk mount. If I cp 12 1-gig files, I get combined 100MB/s transfer rate. Single file cp maxes out at about 15MB/s. Ftp transfer give me over 300MB/s. Running with 9000 MTU. Most smb.conf is default. I even disabled atime and tried ext2 and xfs on ramdisk. Any help will be greatly appreciated. - Vadim -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Very slow samba performance on Centos 6
On Thu, Aug 04, 2011 at 10:49:50AM -0400, vg_ us wrote: I have 2 identical Dell r510 servers with 10gig card, running centos 6 with samba-3.5.4-68.el6_0.2.x86_64. I setup 16G ramdisk samba share on both and ran cp from local ramdisk to samba ramdisk mount. If I cp 12 1-gig files, I get combined 100MB/s transfer rate. Single file cp maxes out at about 15MB/s. Ftp transfer give me over 300MB/s. Running with 9000 MTU. Most smb.conf is default. I even disabled atime and tried ext2 and xfs on ramdisk. Any help will be greatly appreciated. What client application are you using? If it is a cifsfs kernel mount, you might see such artifacts. Please retry with the smbclient(1) application. If that is also slow, we need to investigate further. With best regards, Volker Lendecke -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Very slow samba performance on Centos 6
-- From: Volker Lendecke volker.lende...@sernet.de Sent: Thursday, August 04, 2011 11:01 AM To: vg_ us vg...@hotmail.com Cc: samba@lists.samba.org Subject: Re: [Samba] Very slow samba performance on Centos 6 On Thu, Aug 04, 2011 at 10:49:50AM -0400, vg_ us wrote: I have 2 identical Dell r510 servers with 10gig card, running centos 6 with samba-3.5.4-68.el6_0.2.x86_64. I setup 16G ramdisk samba share on both and ran cp from local ramdisk to samba ramdisk mount. If I cp 12 1-gig files, I get combined 100MB/s transfer rate. Single file cp maxes out at about 15MB/s. Ftp transfer give me over 300MB/s. Running with 9000 MTU. Most smb.conf is default. I even disabled atime and tried ext2 and xfs on ramdisk. Any help will be greatly appreciated. What client application are you using? If it is a cifsfs kernel mount, you might see such artifacts. Please retry with the smbclient(1) application. If that is also slow, we need to investigate further. I re-ran some of the tests with following result: Ftp ramdisk-to-ramdisk: 13572 MB, 32.8 secs - 413.8 MB/s Ftp ramdisk-to-hardisk: 13572 MB, 62.8 secs - 222.4 MB/s Smbclient ramdisk-to-ramdisk: 13572 MB 40 secs - 339 MB/s Smbclient ramdisk-to-harddisk: 13572 MB 64 secs - 212 MB/s cifsfs mount ramdisk-to-ramdisk: 13572 MB 289.8 - 47MB/s cifsfs mounts are really slow, so what happens when linux, windows and mac clients map/mount the share? Are they gonna be this slow? Any way to speed it up? Thanks - Vadim -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Very slow samba performance on Centos 6
On 8/4/2011 1:11 PM, vg_ us wrote: cifsfs mounts are really slow, so what happens when linux, windows and mac clients map/mount the share? Are they gonna be this slow? Any way to speed it up? Unfortunately I don't have an answer to the slow mounts issue. However, you're showing a peak performance of only about half line speed with FTP, which tends to demonstrate your system is in need of overall performance tuning for 10 GbE. Reading, digesting, and using the information in the following article may get you much closer to the ~1GB/s mark of which 10GbE is capable. http://www.redhat.com/promo/summit/2008/downloads/pdf/Thursday/Mark_Wagner.pdf If tweaking these things can double your raw network and FTP throughput, it should do similar for Samba, which would mean ~94 MB/s for cifsfs mount ramdisk-to-ramdisk or to disk. -- Stan -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] Performance issue on Samba with cups printer
Hi everybody, I am using Samba with cups on CentOS 5.5: 1, set up one network printer by cups, it is created as a raw printer; 2, share this printer by Samba, using the below configuration in smb.conf [global] load printers = yes printing = cups printcap name = cups [printers] comment = All Printers path = /var/spool/samba browseable = no guest ok = yes writable = no printable = yes use client driver = Yes Samba users could print by this share printer. But one strange performance issue happened: The share directories in this Samba server are NFS share exported by one NAS(created by Openfiler). With this share printer,Samba users downloaded files from the share in this samba server very slowly, while uploading speed is normal; If I disabled this share printer(just comment the printer section in smb.conf), the download speed became normal. Any help, comment will be highly appreciated! Thanks in advance! Jian -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Juan Pablo wrote: Thanks a lot for the advice. It will run these tests and try to find meaningfull information from them. I will post back results. Thanks Juan Pablo What type of speeds are you expecting? With a GB network, your limit is 125MB/s. I get that with writes, but max out @around 119MB/s on reads due to the not being able to have 'overlapping reads'...;-)... I found to get max performance, I had to adjust the network params in both linux and windows. If I'm totally missing some point, I don't get it. I notice you are trying to use network bonding. I had problems getting network bonding to work correctly. have you tried sniffing with 'wireshark'? Maybe look for duplicate packets or retries? To get optimal speeds you need '0 dups' and '0 retries'... I've only been able to optimize a single Gb ethernet connection. A bonded pair -- even direct from server to Win7 of matched Intel dual-port G-Pro cards gave lower performance than a single wire. It's odd though, with smbclient -- I'd think that would use 'lo0' (no?) I'd think that would get better. I noticed in the test below use of 8MB files. 70MB/s would be a good speed for reading those over the net. My best raw speeds were using 16-256MB on multi-gig files. But opening single files ... I'd try opening them all first, then sending the data, so you are measuring data perf. My maximum write perf was done to a file (from windows) using: CF=notrunc,nocreat; OF=direct dd if=/dev/zero of='file' bs=16M count=128 oflag=$OF conv=$CF Optimizing the network settings on both the linux server and win7 client gave me another ~20-30%. I wouldn't trust my testing now, though, as I recently upgraded, and can't even get nmbd to run...(sigh)... 1 step forward, 3 steps back! Test typeLocal (dd) Local (smbclient) Window 7 Case1161 101 63 Case2122 119 68 Case1: Read 1000 files 8 MByte each Case2: 4 processes each reading 1000 files of 8 MByte each Any idea how can I debug where the bottleneck is or why I get so low numbers when reading from Windows? strace the smbd process with strace -ttT. Network trace. Look at netstat -nt while the test is running. Send/Recv queues full? Run top, is the CPU fully busy? There's no silver bullet for performance tuning unfortunately, sorry. Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Alan Hodgson wrote: On Wed, May 25, 2011 at 08:02:56PM -0700, Juan Pablo wrote: - 4 Intel Gigagit ethernet NIC ports with 802.3ad bonding connected to a switch configured tu use 802.3ad - 8 2TB 7.2 krpm SATA disks with hardware RAID5 (RAID stripe size 1024 bytes, controller and disk cache enabled, readahead enabled) - XFS filesystem (created with the following parameters: size=64k -d su=1024k,sw=7) - Average file size in the share: 8 MByte - Gigabit network composed by Cat5E certified cabling and DLink DGS-3427 gigabit switch. The way Linux does 803.ad is not really how you might expect. ... It's still not great though. You'd really be better off with a 10Gb/s interface out to your switch if you need to guarantee multiple 1Gb/s connections over a small number of simultaneous connections. Given my experience with bonded ethernet, I'd have to agree. I'm 'just' waiting for the 10Gb prices to come down. Still a bit out of reach for a home network setup. BTW... su=1024k?!? What raid controller are you using? Usually 64K is usually recommended for max performance. But then above you say RAID strip size is 1024bytes? There is a difference, no? Which is it? Either way: a bit off from optimal. You want to set your log size to 32768b (not 64k; note: 32768b=128k). For mount options, I have 'swalloc,largeio,logbsize=256k,nobarrier'. Note, for nobarrier, you *should* have your system on a UPS, and a battery backup on the RAID controller's cache (LSI controllers have this, others may as well). Note, some perf-related options(from my smb.conf) (with host networking tuned as well), I have: aio read size = 65546 aio write size = 65536 max xmit = 66576 min receivefile size = 65536 map acl inherit = Yes server schannel = no socket options = TCP_NODELAY IPTOS_LOWDELAY SO_SNDBUF=4194304 SO_RCVBUF=4194304 use sendfile = yes Note: I'm not sure why my max xmit is 64k, I probably had a reason when I set it up -- not even sure if 64k is legal, it might explain why my read rates are 6MB/s slower than my writes (119MB/s vs. 125MB/s) over Gb lan. Those are MAX rates to a linear file -- NOT random small reads/writes, BTW Though I'll regularly see 50MB in random, with 100MB for large files. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
On Wed, Jun 01, 2011 at 06:46:51PM -0700, Juan Pablo wrote: Hi Volker, I've removed the SO_RCVBUF=65536 SO_SNDBUF=65536 and the 3 other setting, reloaded samba and repeated the tests but still getting the same results for the local tests and also from Windows. I am getting the following results in MBytes/s: Test typeLocal (dd) Local (smbclient) Window 7 Case1161 101 63 Case2122 119 68 Case1: Read 1000 files 8 MByte each Case2: 4 processes each reading 1000 files of 8 MByte each Any idea how can I debug where the bottleneck is or why I get so low numbers when reading from Windows? strace the smbd process with strace -ttT. Network trace. Look at netstat -nt while the test is running. Send/Recv queues full? Run top, is the CPU fully busy? There's no silver bullet for performance tuning unfortunately, sorry. Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
On 5/25/2011 10:02 PM, Juan Pablo wrote: OS access: Simultaneous read (4 processes): 118 MByte/s average Samba local access: Simultaneous read (4 processes): 102 MByte/s average Samba server from Windows 7: Simultaneous read (4 terminals): 70 MByte/s average The first two results above demonstrate a slow disk subsystem not suitable for streaming multiple files to multiple concurrent clients at high data rates. Your spindles are too slow and/or you don't have enough to satisfy your test methodology. Four concurrent dd copies yields 118 MB/s per process, only ~15% disk headroom above wire speed GbE. Your smbd+smbclient local process disk bandwidth overhead appears to be roughly 13 percent. I don't know what the optimal percent here should be but 13% above a dd copy process seems reasonable given the additional data movement through smbd and smbclient buffers. It is clear that you don't have enough head seek performance for 4 or more client streams of 1000 x 8MB files. This doesn't necessarily address the 30% drop in over the wire to Win7 client performance, but we'll get to that later. To confirm the disk deficiency issue, I recommend the following test: Make a 2GB tmpfs ramdisk on the server and run your tests against it, albeit with 200 instead of 1000 8MB files. Instructions: http://prefetch.net/blog/index.php/2006/11/30/creating-a-ramdisk-with-linux/ This will tell you if your server block storage subsystem is part of the problem, and will give you a maximum throughput per Samba process baseline. You should get something like 5GB/s+ local smbclient throughput from a tmpfs ramdisk on that Xeon platform with its raw 25GB/s memory bandwidth. Run a single Win7 workstation SMB test copy to a freshly booted machine so most of the memory is free for buffering the inbound files. This will mostly eliminate the slow local disk as a bottleneck. Now run your 4 concurrent Win7 client test and compare to the single client test results. This should tell you if you have a bonding problem or not, either in the server NICs or the switch. You didn't mention jumbo frames. Enable jumbo if not already. It may help. Something else to consider is that the kernel shipped with CentOS 5.6, 2.6.18, the Pirate kernel, is now 4.5 years old, released in Sept of 2006 (http://kerneltrap.org/node/7144). There have been just a few performance enhancements between 2.6.18 and 3.0, specifically to the network stack. ;) The CentOS packages are older than dirt as well. If you're not wed to CentOS you should look at more recent distros. -- Stan -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Thanks a lot for the advice. It will run these tests and try to find meaningfull information from them. I will post back results. Thanks Juan Pablo From: Volker Lendecke volker.lende...@sernet.de To: Juan Pablo jhur...@yahoo.com Cc: Jeremy Allison j...@samba.org; samba@lists.samba.org Sent: Thu, June 2, 2011 3:49:17 AM Subject: Re: [Samba] Samba performance On Wed, Jun 01, 2011 at 06:46:51PM -0700, Juan Pablo wrote: Hi Volker, I've removed the SO_RCVBUF=65536 SO_SNDBUF=65536 and the 3 other setting, reloaded samba and repeated the tests but still getting the same results for the local tests and also from Windows. I am getting the following results in MBytes/s: Test typeLocal (dd) Local (smbclient) Window 7 Case1161 101 63 Case2122 119 68 Case1: Read 1000 files 8 MByte each Case2: 4 processes each reading 1000 files of 8 MByte each Any idea how can I debug where the bottleneck is or why I get so low numbers when reading from Windows? strace the smbd process with strace -ttT. Network trace. Look at netstat -nt while the test is running. Send/Recv queues full? Run top, is the CPU fully busy? There's no silver bullet for performance tuning unfortunately, sorry. Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Hi Stan, Thanks for your feedback and suggestions! The disk subsystem is composed by: - 8 WD2002FAEX SATA 2TB hard drives (7200 RPM, 64MB cache, 4.2 ms avg latency) - 1 Intel RAID controller RS2BL080 with 512 MB configured with 1 virtual drive 12.7 TB (hardware RAID 5 with 1 MB stripe size, caches enabled, read-ahead enabled) In your experience, should I expect higher performance from this hardware? Will try the ramdisk test you are suggesting and post back the results. Thanks for the suggestion! I have jumbo frames enabled in the switches but windows drivers for the Intel network cards don't have the option to enable jumbo frames. I also tried raising the MTU in the linux server but performance was even worse (I thought this was related to the windows NIC driver not supporting MTUs larger than 1500). I also modified windows registry to manually enable smb2 protocol because it was not negotiating smb2. Do you think of any other optimization that can be done on the windows terminals? Thanks Juan Pablo From: Stan Hoeppner s...@hardwarefreak.com To: Juan Pablo jhur...@yahoo.com Cc: samba@lists.samba.org Sent: Thu, June 2, 2011 8:50:21 AM Subject: Re: [Samba] Samba performance On 5/25/2011 10:02 PM, Juan Pablo wrote: OS access: Simultaneous read (4 processes): 118 MByte/s average Samba local access: Simultaneous read (4 processes): 102 MByte/s average Samba server from Windows 7: Simultaneous read (4 terminals): 70 MByte/s average The first two results above demonstrate a slow disk subsystem not suitable for streaming multiple files to multiple concurrent clients at high data rates. Your spindles are too slow and/or you don't have enough to satisfy your test methodology. Four concurrent dd copies yields 118 MB/s per process, only ~15% disk headroom above wire speed GbE. Your smbd+smbclient local process disk bandwidth overhead appears to be roughly 13 percent. I don't know what the optimal percent here should be but 13% above a dd copy process seems reasonable given the additional data movement through smbd and smbclient buffers. It is clear that you don't have enough head seek performance for 4 or more client streams of 1000 x 8MB files. This doesn't necessarily address the 30% drop in over the wire to Win7 client performance, but we'll get to that later. To confirm the disk deficiency issue, I recommend the following test: Make a 2GB tmpfs ramdisk on the server and run your tests against it, albeit with 200 instead of 1000 8MB files. Instructions: http://prefetch.net/blog/index.php/2006/11/30/creating-a-ramdisk-with-linux/ This will tell you if your server block storage subsystem is part of the problem, and will give you a maximum throughput per Samba process baseline. You should get something like 5GB/s+ local smbclient throughput from a tmpfs ramdisk on that Xeon platform with its raw 25GB/s memory bandwidth. Run a single Win7 workstation SMB test copy to a freshly booted machine so most of the memory is free for buffering the inbound files. This will mostly eliminate the slow local disk as a bottleneck. Now run your 4 concurrent Win7 client test and compare to the single client test results. This should tell you if you have a bonding problem or not, either in the server NICs or the switch. You didn't mention jumbo frames. Enable jumbo if not already. It may help. Something else to consider is that the kernel shipped with CentOS 5.6, 2.6.18, the Pirate kernel, is now 4.5 years old, released in Sept of 2006 (http://kerneltrap.org/node/7144). There have been just a few performance enhancements between 2.6.18 and 3.0, specifically to the network stack. ;) The CentOS packages are older than dirt as well. If you're not wed to CentOS you should look at more recent distros. -- Stan -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
On 6/2/2011 2:24 PM, Juan Pablo wrote: Hi Stan, Thanks for your feedback and suggestions! You're welcome. Let's hope they're beneficial. The disk subsystem is composed by: - 8 WD2002FAEX SATA 2TB hard drives (7200 RPM, 64MB cache, 4.2 ms avg latency) - 1 Intel RAID controller RS2BL080 with 512 MB configured with 1 virtual drive 12.7 TB (hardware RAID 5 with 1 MB stripe size, caches enabled, read-ahead enabled) In your experience, should I expect higher performance from this hardware? That depends on your target workload(s). You're currently achieving single stream read performance of 780 MB/s, over 110MB/s per drive. That's a really good streaming read, close to peak drive read performance. The problem I see is when you have 4 readers (Win7 clients) reading 4,000 files each. If these are 16,000 unique files, not each Win7 machine reading the same 4,000 files, i.e. no cache benefit, then I don't think your disk heads are going be able to seek fast enough to service all the read requests and hit wire speed SMB. If your production load will be significantly less than this artificial test load, you may be fine. Will try the ramdisk test you are suggesting and post back the results. Thanks for the suggestion! The results should be informative, one way or the other. I have jumbo frames enabled in the switches but windows drivers for the Intel network cards don't have the option to enable jumbo frames. I also tried raising the MTU in the linux server but performance was even worse (I thought this was related to the windows NIC driver not supporting MTUs larger than 1500). Lack of jumbo frames is probably hurting your wire performance due to increased interrupt processing and other factors. I'm surprised some Intel NICs don't support jumbo frames. Must be desktop adapters. Can you post the model# of the NICs in the Win7 PCs and those in the server so I can do some research? I also modified windows registry to manually enable smb2 protocol because it was not negotiating smb2. Do you think of any other optimization that can be done on the windows terminals? I have no experience yet with SMB2 or Win7 so I can't really say. You should be able to tune that server and the clients to hit near wire speed with regular SMB. I suggest solving that problem first, then worry about SMB2. -- Stan -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Hi Volker, I've removed the SO_RCVBUF=65536 SO_SNDBUF=65536 and the 3 other setting, reloaded samba and repeated the tests but still getting the same results for the local tests and also from Windows. I am getting the following results in MBytes/s: Test typeLocal (dd) Local (smbclient) Window 7 Case1161 101 63 Case2122 119 68 Case1: Read 1000 files 8 MByte each Case2: 4 processes each reading 1000 files of 8 MByte each Any idea how can I debug where the bottleneck is or why I get so low numbers when reading from Windows? Thanks Juan Pablo From: Volker Lendecke volker.lende...@sernet.de To: Juan Pablo jhur...@yahoo.com Cc: Jeremy Allison j...@samba.org; samba@lists.samba.org Sent: Fri, May 27, 2011 11:25:31 AM Subject: Re: [Samba] Samba performance On Fri, May 27, 2011 at 06:34:50AM -0700, Juan Pablo wrote: Hi Volker, I am using the following socket options: socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=65536 SO_SNDBUF=65536 Just remove the SO_RCVBUF=65536 SO_SNDBUF=65536 settings. Unless you're on a very old Linux or other Unix the kernel is far better off figuring out that itself. read raw = yes write raw = yes max xmit = 65535 Just remove these 3 settings. If it's still slow after that, we need to do more analysis. Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Hi Volker, I am using the following socket options: socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=65536 SO_SNDBUF=65536 I've been testing with lower rcvbuf and sndbuf values but this seems to be the best. Any advice on how to calculate this value besides trial and error? Following is my smb.conf: [global] workgroup = Workgrouplasp encrypt passwords = yes security = user smb passwd file = /etc/samba/smbpasswd netbios name = LASPFS server string = LASP Linux File Server dns proxy = no log level = 1 oplocks = yes level2 oplocks = yes socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=65536 SO_SNDBUF=65536 guest account = samba map to guest = Bad User map to guest = Bad Password read raw = yes write raw = yes max xmit = 65535 dead time = 15 getwd cache = yes interfaces = 192.168.1.5/255.255.255.0 bind interfaces only = yes socket address = 192.168.1.5 max protocol = SMB2 [laspfiles] comment = Public File Storage browseable = yes force user = samba force group = samba path = /srv/samba/filestore/laspfiles read only = no guest ok = yes My focus was to tune the fileserver for several 8MB file transfer. Any idea on how to improve this? Thanks for your help! Juan Pablo From: Volker Lendecke volker.lende...@sernet.de To: Juan Pablo jhur...@yahoo.com Cc: Jeremy Allison j...@samba.org; samba@lists.samba.org Sent: Thu, May 26, 2011 2:27:20 PM Subject: Re: [Samba] Samba performance On Thu, May 26, 2011 at 10:14:31AM -0700, Juan Pablo wrote: Hi Jeremy, Thanks for your reply! The tests we did with the Windows 7 terminals was using smb2. When we enabled smb2 in samba we saw in samba logs that it was not being used. We modified Windows 7 registry as described in http://www.techemperor.com/2009/09/21/manual-patch-for-windows-vistaserver-2008-smb2-flaw/ / to enable smb2 to start using smb2. Once this was done smb2 was negotiated but there was no speed difference. The OS read test is done iterating from 0 to 999 a dd if=testFile-xxx of=/dev/null bs=1k. The samba local access test is done with smbclient from the same machine sending the output to /dev/null. Is the speed decrease (from 158 MB/s to 71 MB/s) from what I get when I test from the OS to what I get with samba normal? With smbclient you should get near wire speed. Do you have socket options set in your smb.conf? Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Hi Daniel, Thanks for your advice. I thought that XFS was a good alternative as I have plenty of unused CPU. I've been googling and found that XFS had good performance but used more CPU than other alternatives. I will like to get as much as I can from this fileserver. I will try EXT4 to see if I can make any improvement with it but I am currently seeing a very big difference on the average speed I get from reading directly (using the dd command) to what I get with smbclient without passing through the switch. Thanks again! Juan Pablo From: Daniel Deptuła daniel.dept...@gmail.com To: samba@lists.samba.org Cc: jhur...@yahoo.com Sent: Thu, May 26, 2011 1:19:03 PM Subject: Re: [Samba] Samba performance W dniu 2011-05-26 05:02, Juan Pablo pisze: Hi everyone, I'm trying to use samba in a small video post production house but we are not getting the performance we expected. Our setup: - CenOS 5.6 x86-64 - samba.x86_64 (3.0.33-3.29.el5_6.2 and 3.6.0rc1) - Intel based server (One 4 core Xeon E5620 @ 2.40GHz, 8 GB RAM) - 4 Intel Gigagit ethernet NIC ports with 802.3ad bonding connected to a switch configured tu use 802.3ad - 8 2TB 7.2 krpm SATA disks with hardware RAID5 (RAID stripe size 1024 bytes, controller and disk cache enabled, readahead enabled) - XFS filesystem (created with the following parameters: size=64k -d su=1024k,sw=7) - Average file size in the share: 8 MByte - Gigabit network composed by Cat5E certified cabling and DLink DGS-3427 gigabit switch. - Intel I7 based terminals with Intel gigabit NIC, running Windows 7 Test results: OS access: Sequential write (1 x 31 GByte file): 500 MByte/s Sequential read (1 x 31 GByte file): 780 MByte/s Write (1000 files 8 MByte each): 249 MByte/s average Read (1000 files 8 MByte each): 158 MByte/s average Simultaneous write (4 processes each writing 1000 files of 8 MByte each ): 188 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 118 MByte/s average Samba local access (stock CentOS samba 3.0.33 connecting from the same server with smbclient): Sequential read (1 x 31 GByte file): 267 MByte/s Read (1000 files 8 MByte each): 71 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 102 MByte/s average Samba local access (Samba 3.6.0rc1 compiled from GIT repo. Connecting from the same server with smbclient): Read (1000 files 8 MByte each): 95 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 103 MByte/s average Samba server accessed from Windows 7 terminals (samba 3.6.0rc1): Read (1 terminal copying from samba fileserver to local disk 1000 files 8 MByte each): 60 MByte/s average Simultaneous read (4 terminals each copying from samba fileserver to local disk 1000 files of 8 MByte each): 70 MByte/s average Note: Simultaneos read speed is measured adding the size of all transfered files and dividing it by the time taken to transfer these files. I will appreciate any feedback about the results we are getting and advice on how to improve this. Thanks in advance Juan Pablo Maybe try the ext4 filesystem? With a new kernel - with stable support for it. Many tests have shown that ext4 is faster than XFS, but also remember to tune the parameters when creating the filesystem. You can try several different configurations and compare their performance (performance for the same parameters can be different on different hardware and RAID configurations, so options recommended by other people are not always the best for you). Filesystem mount options are also important! The second thing is network - some switches do not do port trunking well - for example they use always use one wire even if there are 2 or more connected in a trunk - so it does not improve performance - only the reliability. Usually also one data stream does not go through more than one wire, so the only possibility to get 4 Gbit speed from your server is to connect 4 simultaneously downloading stations to the switch. You can check the bandwidth usage on each interface of the server with the iftop command. For measuring the network performance I recommend also the iperf tool. Also google about network and tcp tuning in linux (parameters like txqueuelen, buffer sizes etc). About tuning samba performance you can read for example here: http://www.samba.org/samba/docs/man/Samba-HOWTO-Collection/speed.html But also in many other places on the Internet. Best regards, Daniel -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Hi Alan, My setup is similar to what you are suggesting: options bond0 miimon=100 mode=4 lacp_rate=1 I am still not happy with how bonding is working. It seems that the first 3 connections are getting assigned to different links but the 4th is using one of the previously used ones. I will add xmit_hash_policy=layer2+3 to see if I get any improvement. I thought the algorithm was smart enough to get the first n connections (being n the number of available links) into different links. Thanks for your advice! I will try your suggestion. Regards Juan Pablo From: Alan Hodgson ahodg...@simkin.ca To: samba@lists.samba.org Sent: Thu, May 26, 2011 2:45:12 PM Subject: Re: [Samba] Samba performance On Wed, May 25, 2011 at 08:02:56PM -0700, Juan Pablo wrote: - 4 Intel Gigagit ethernet NIC ports with 802.3ad bonding connected to a switch configured tu use 802.3ad - 8 2TB 7.2 krpm SATA disks with hardware RAID5 (RAID stripe size 1024 bytes, controller and disk cache enabled, readahead enabled) - XFS filesystem (created with the following parameters: size=64k -d su=1024k,sw=7) - Average file size in the share: 8 MByte - Gigabit network composed by Cat5E certified cabling and DLink DGS-3427 gigabit switch. The way Linux does 803.ad is not really how you might expect. Basically, it isn't really smart about how it assigns individual network streams to ports. It hashes destination MAC addresses and then randomly assigns them to ports. It doesn't pay attention to current traffic levels. It is quite likely that all 4 connections ended up using the same port - which you could see by monitoring the total packets transmitted on each interface during your test. You can make it somewhat more likely to use more ports by using a different hashing algorithm. I've had the best results with something like this in modprobe.conf: options bonding mode=802.3ad miimon=100 lacp_rate=1 xmit_hash_policy=layer2+3 It's still not great though. You'd really be better off with a 10Gb/s interface out to your switch if you need to guarantee multiple 1Gb/s connections over a small number of simultaneous connections. -- The whole universe is change and life itself is but what you deem it. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
On Fri, May 27, 2011 at 06:34:50AM -0700, Juan Pablo wrote: Hi Volker, I am using the following socket options: socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=65536 SO_SNDBUF=65536 Just remove the SO_RCVBUF=65536 SO_SNDBUF=65536 settings. Unless you're on a very old Linux or other Unix the kernel is far better off figuring out that itself. read raw = yes write raw = yes max xmit = 65535 Just remove these 3 settings. If it's still slow after that, we need to do more analysis. Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
On Wed, May 25, 2011 at 08:02:56PM -0700, Juan Pablo wrote: Hi everyone, I'm trying to use samba in a small video post production house but we are not getting the performance we expected. Our setup: - CenOS 5.6 x86-64 - samba.x86_64 (3.0.33-3.29.el5_6.2 and 3.6.0rc1) - Intel based server (One 4 core Xeon E5620 @ 2.40GHz, 8 GB RAM) - 4 Intel Gigagit ethernet NIC ports with 802.3ad bonding connected to a switch configured tu use 802.3ad - 8 2TB 7.2 krpm SATA disks with hardware RAID5 (RAID stripe size 1024 bytes, controller and disk cache enabled, readahead enabled) - XFS filesystem (created with the following parameters: size=64k -d su=1024k,sw=7) - Average file size in the share: 8 MByte - Gigabit network composed by Cat5E certified cabling and DLink DGS-3427 gigabit switch. - Intel I7 based terminals with Intel gigabit NIC, running Windows 7 Test results: OS access: Sequential write (1 x 31 GByte file): 500 MByte/s Sequential read (1 x 31 GByte file): 780 MByte/s Write (1000 files 8 MByte each): 249 MByte/s average Read (1000 files 8 MByte each): 158 MByte/s average Simultaneous write (4 processes each writing 1000 files of 8 MByte each ): 188 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 118 MByte/s average Samba local access (stock CentOS samba 3.0.33 connecting from the same server with smbclient): Sequential read (1 x 31 GByte file): 267 MByte/s Read (1000 files 8 MByte each): 71 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 102 MByte/s average Samba local access (Samba 3.6.0rc1 compiled from GIT repo. Connecting from the same server with smbclient): Read (1000 files 8 MByte each): 95 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 103 MByte/s average Samba server accessed from Windows 7 terminals (samba 3.6.0rc1): Read (1 terminal copying from samba fileserver to local disk 1000 files 8 MByte each): 60 MByte/s average Simultaneous read (4 terminals each copying from samba fileserver to local disk 1000 files of 8 MByte each): 70 MByte/s average Note: Simultaneos read speed is measured adding the size of all transfered files and dividing it by the time taken to transfer these files. I will appreciate any feedback about the results we are getting and advice on how to improve this. If you're using 3.6.0 and Windows 7 clients try turning on SMB2 support by setting max protocol = smb2 in the [global] section of your smb.conf. Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
W dniu 2011-05-26 05:02, Juan Pablo pisze: Hi everyone, I'm trying to use samba in a small video post production house but we are not getting the performance we expected. Our setup: - CenOS 5.6 x86-64 - samba.x86_64 (3.0.33-3.29.el5_6.2 and 3.6.0rc1) - Intel based server (One 4 core Xeon E5620 @ 2.40GHz, 8 GB RAM) - 4 Intel Gigagit ethernet NIC ports with 802.3ad bonding connected to a switch configured tu use 802.3ad - 8 2TB 7.2 krpm SATA disks with hardware RAID5 (RAID stripe size 1024 bytes, controller and disk cache enabled, readahead enabled) - XFS filesystem (created with the following parameters: size=64k -d su=1024k,sw=7) - Average file size in the share: 8 MByte - Gigabit network composed by Cat5E certified cabling and DLink DGS-3427 gigabit switch. - Intel I7 based terminals with Intel gigabit NIC, running Windows 7 Test results: OS access: Sequential write (1 x 31 GByte file): 500 MByte/s Sequential read (1 x 31 GByte file): 780 MByte/s Write (1000 files 8 MByte each): 249 MByte/s average Read (1000 files 8 MByte each): 158 MByte/s average Simultaneous write (4 processes each writing 1000 files of 8 MByte each ): 188 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 118 MByte/s average Samba local access (stock CentOS samba 3.0.33 connecting from the same server with smbclient): Sequential read (1 x 31 GByte file): 267 MByte/s Read (1000 files 8 MByte each): 71 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 102 MByte/s average Samba local access (Samba 3.6.0rc1 compiled from GIT repo. Connecting from the same server with smbclient): Read (1000 files 8 MByte each): 95 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 103 MByte/s average Samba server accessed from Windows 7 terminals (samba 3.6.0rc1): Read (1 terminal copying from samba fileserver to local disk 1000 files 8 MByte each): 60 MByte/s average Simultaneous read (4 terminals each copying from samba fileserver to local disk 1000 files of 8 MByte each): 70 MByte/s average Note: Simultaneos read speed is measured adding the size of all transfered files and dividing it by the time taken to transfer these files. I will appreciate any feedback about the results we are getting and advice on how to improve this. Thanks in advance Juan Pablo Maybe try the ext4 filesystem? With a new kernel - with stable support for it. Many tests have shown that ext4 is faster than XFS, but also remember to tune the parameters when creating the filesystem. You can try several different configurations and compare their performance (performance for the same parameters can be different on different hardware and RAID configurations, so options recommended by other people are not always the best for you). Filesystem mount options are also important! The second thing is network - some switches do not do port trunking well - for example they use always use one wire even if there are 2 or more connected in a trunk - so it does not improve performance - only the reliability. Usually also one data stream does not go through more than one wire, so the only possibility to get 4 Gbit speed from your server is to connect 4 simultaneously downloading stations to the switch. You can check the bandwidth usage on each interface of the server with the iftop command. For measuring the network performance I recommend also the iperf tool. Also google about network and tcp tuning in linux (parameters like txqueuelen, buffer sizes etc). About tuning samba performance you can read for example here: http://www.samba.org/samba/docs/man/Samba-HOWTO-Collection/speed.html But also in many other places on the Internet. Best regards, Daniel -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
On Thu, May 26, 2011 at 09:16:02AM -0700, Jeremy Allison wrote: If you're using 3.6.0 and Windows 7 clients try turning on SMB2 support by setting max protocol = smb2 in the [global] section of your smb.conf. Well, using smbclient should definitely get better performance. Something is wrong here... Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
Hi Jeremy, Thanks for your reply! The tests we did with the Windows 7 terminals was using smb2. When we enabled smb2 in samba we saw in samba logs that it was not being used. We modified Windows 7 registry as described in http://www.techemperor.com/2009/09/21/manual-patch-for-windows-vistaserver-2008-smb2-flaw/ to enable smb2 to start using smb2. Once this was done smb2 was negotiated but there was no speed difference. The OS read test is done iterating from 0 to 999 a dd if=testFile-xxx of=/dev/null bs=1k. The samba local access test is done with smbclient from the same machine sending the output to /dev/null. Is the speed decrease (from 158 MB/s to 71 MB/s) from what I get when I test from the OS to what I get with samba normal? Juan Pablo From: Jeremy Allison j...@samba.org To: Juan Pablo jhur...@yahoo.com Cc: samba@lists.samba.org Sent: Thu, May 26, 2011 1:16:02 PM Subject: Re: [Samba] Samba performance On Wed, May 25, 2011 at 08:02:56PM -0700, Juan Pablo wrote: Hi everyone, I'm trying to use samba in a small video post production house but we are not getting the performance we expected. Our setup: - CenOS 5.6 x86-64 - samba.x86_64 (3.0.33-3.29.el5_6.2 and 3.6.0rc1) - Intel based server (One 4 core Xeon E5620 @ 2.40GHz, 8 GB RAM) - 4 Intel Gigagit ethernet NIC ports with 802.3ad bonding connected to a switch configured tu use 802.3ad - 8 2TB 7.2 krpm SATA disks with hardware RAID5 (RAID stripe size 1024 bytes, controller and disk cache enabled, readahead enabled) - XFS filesystem (created with the following parameters: size=64k -d su=1024k,sw=7) - Average file size in the share: 8 MByte - Gigabit network composed by Cat5E certified cabling and DLink DGS-3427 gigabit switch. - Intel I7 based terminals with Intel gigabit NIC, running Windows 7 Test results: OS access: Sequential write (1 x 31 GByte file): 500 MByte/s Sequential read (1 x 31 GByte file): 780 MByte/s Write (1000 files 8 MByte each): 249 MByte/s average Read (1000 files 8 MByte each): 158 MByte/s average Simultaneous write (4 processes each writing 1000 files of 8 MByte each ): 188 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 118 MByte/s average Samba local access (stock CentOS samba 3.0.33 connecting from the same server with smbclient): Sequential read (1 x 31 GByte file): 267 MByte/s Read (1000 files 8 MByte each): 71 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 102 MByte/s average Samba local access (Samba 3.6.0rc1 compiled from GIT repo. Connecting from the same server with smbclient): Read (1000 files 8 MByte each): 95 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 103 MByte/s average Samba server accessed from Windows 7 terminals (samba 3.6.0rc1): Read (1 terminal copying from samba fileserver to local disk 1000 files 8 MByte each): 60 MByte/s average Simultaneous read (4 terminals each copying from samba fileserver to local disk 1000 files of 8 MByte each): 70 MByte/s average Note: Simultaneos read speed is measured adding the size of all transfered files and dividing it by the time taken to transfer these files. I will appreciate any feedback about the results we are getting and advice on how to improve this. If you're using 3.6.0 and Windows 7 clients try turning on SMB2 support by setting max protocol = smb2 in the [global] section of your smb.conf. Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
On Thu, May 26, 2011 at 10:14:31AM -0700, Juan Pablo wrote: Hi Jeremy, Thanks for your reply! The tests we did with the Windows 7 terminals was using smb2. When we enabled smb2 in samba we saw in samba logs that it was not being used. We modified Windows 7 registry as described in http://www.techemperor.com/2009/09/21/manual-patch-for-windows-vistaserver-2008-smb2-flaw/ to enable smb2 to start using smb2. Once this was done smb2 was negotiated but there was no speed difference. The OS read test is done iterating from 0 to 999 a dd if=testFile-xxx of=/dev/null bs=1k. The samba local access test is done with smbclient from the same machine sending the output to /dev/null. Is the speed decrease (from 158 MB/s to 71 MB/s) from what I get when I test from the OS to what I get with samba normal? With smbclient you should get near wire speed. Do you have socket options set in your smb.conf? Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance
On Wed, May 25, 2011 at 08:02:56PM -0700, Juan Pablo wrote: - 4 Intel Gigagit ethernet NIC ports with 802.3ad bonding connected to a switch configured tu use 802.3ad - 8 2TB 7.2 krpm SATA disks with hardware RAID5 (RAID stripe size 1024 bytes, controller and disk cache enabled, readahead enabled) - XFS filesystem (created with the following parameters: size=64k -d su=1024k,sw=7) - Average file size in the share: 8 MByte - Gigabit network composed by Cat5E certified cabling and DLink DGS-3427 gigabit switch. The way Linux does 803.ad is not really how you might expect. Basically, it isn't really smart about how it assigns individual network streams to ports. It hashes destination MAC addresses and then randomly assigns them to ports. It doesn't pay attention to current traffic levels. It is quite likely that all 4 connections ended up using the same port - which you could see by monitoring the total packets transmitted on each interface during your test. You can make it somewhat more likely to use more ports by using a different hashing algorithm. I've had the best results with something like this in modprobe.conf: options bonding mode=802.3ad miimon=100 lacp_rate=1 xmit_hash_policy=layer2+3 It's still not great though. You'd really be better off with a 10Gb/s interface out to your switch if you need to guarantee multiple 1Gb/s connections over a small number of simultaneous connections. -- The whole universe is change and life itself is but what you deem it. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] Samba performance
Hi everyone, I'm trying to use samba in a small video post production house but we are not getting the performance we expected. Our setup: - CenOS 5.6 x86-64 - samba.x86_64 (3.0.33-3.29.el5_6.2 and 3.6.0rc1) - Intel based server (One 4 core Xeon E5620 @ 2.40GHz, 8 GB RAM) - 4 Intel Gigagit ethernet NIC ports with 802.3ad bonding connected to a switch configured tu use 802.3ad - 8 2TB 7.2 krpm SATA disks with hardware RAID5 (RAID stripe size 1024 bytes, controller and disk cache enabled, readahead enabled) - XFS filesystem (created with the following parameters: size=64k -d su=1024k,sw=7) - Average file size in the share: 8 MByte - Gigabit network composed by Cat5E certified cabling and DLink DGS-3427 gigabit switch. - Intel I7 based terminals with Intel gigabit NIC, running Windows 7 Test results: OS access: Sequential write (1 x 31 GByte file): 500 MByte/s Sequential read (1 x 31 GByte file): 780 MByte/s Write (1000 files 8 MByte each): 249 MByte/s average Read (1000 files 8 MByte each): 158 MByte/s average Simultaneous write (4 processes each writing 1000 files of 8 MByte each ): 188 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 118 MByte/s average Samba local access (stock CentOS samba 3.0.33 connecting from the same server with smbclient): Sequential read (1 x 31 GByte file): 267 MByte/s Read (1000 files 8 MByte each): 71 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 102 MByte/s average Samba local access (Samba 3.6.0rc1 compiled from GIT repo. Connecting from the same server with smbclient): Read (1000 files 8 MByte each): 95 MByte/s average Simultaneous read (4 processes each reading 1000 files of 8 MByte each): 103 MByte/s average Samba server accessed from Windows 7 terminals (samba 3.6.0rc1): Read (1 terminal copying from samba fileserver to local disk 1000 files 8 MByte each): 60 MByte/s average Simultaneous read (4 terminals each copying from samba fileserver to local disk 1000 files of 8 MByte each): 70 MByte/s average Note: Simultaneos read speed is measured adding the size of all transfered files and dividing it by the time taken to transfer these files. I will appreciate any feedback about the results we are getting and advice on how to improve this. Thanks in advance Juan Pablo -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
tcpdump reading a 256MB file (from tmpfs) on XP64: http://zewt.org/~glenn/samba.tcpdump.gz, 56M/sec Server (10.0.0.1): Ubuntu (karmic), 2.6.28-13, Samba 3.4.0, Intel 82572EI Client (10.0.0.2): XP64 SP2, Intel PRO/1000 GT (same card as the server, I think) wget http://10.0.0.1/ramdisk -O nul: 97M/sec (slower than it should be; seen around 120M/sec from onboard NICs...) The only thing between the systems is a Procurve 1400-8G. smb.conf attached. -- Glenn Maynard -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
On Tue, Mar 15, 2011 at 02:28:37PM -0400, Glenn Maynard wrote: tcpdump reading a 256MB file (from tmpfs) on XP64: http://zewt.org/~glenn/samba.tcpdump.gz, 56M/sec That's pretty much what you would expect. The trace shows that your program reads sequentially in 61k chunks. Given that your network latency is roughly 600 microseconds, ftp://pserver.samba.org/pub/unpacked/junkcode/expected_throughput.pl gives 55MB/sec. You might want to try smbclient, that does larger parallel reads, or a copy program that uses multiple threads simultaneously. Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
On Tue, Mar 15, 2011 at 3:36 PM, Volker Lendecke volker.lende...@sernet.dewrote: On Tue, Mar 15, 2011 at 02:28:37PM -0400, Glenn Maynard wrote: tcpdump reading a 256MB file (from tmpfs) on XP64: http://zewt.org/~glenn/samba.tcpdump.gz, 56M/sec That's pretty much what you would expect. The trace shows that your program reads sequentially in 61k chunks. Given My program reads in 1M chunks; it's the OS splitting apart the reads. that your network latency is roughly 600 microseconds, ftp://pserver.samba.org/pub/unpacked/junkcode/expected_throughput.pl gives 55MB/sec. You might want to try smbclient, that does larger parallel reads, or a copy program that uses multiple threads simultaneously. That's what I'd expect if things are as broken as they look from that dump: the client (XP) isn't buffering requests at all, so the window caps out at 64k. The mail from earlier[1] suggests that some people get much better speed from other servers (though without enough information to really be useful). I don't know the protocol, so I can't tell if there's something preventing the client from buffering multiple read requests, for example, which I'd expect of any network FS. Multiple threads isn't a good solution (at these speeds, it's liable to severely hurt streaming performance), and in any case it doesn't help actual use, with real applications. [1] http://lists.samba.org/archive/samba/2010-June/156708.html -- Glenn Maynard -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
On Tue, Mar 15, 2011 at 04:35:32PM -0400, Glenn Maynard wrote: On Tue, Mar 15, 2011 at 3:36 PM, Volker Lendecke volker.lende...@sernet.dewrote: On Tue, Mar 15, 2011 at 02:28:37PM -0400, Glenn Maynard wrote: tcpdump reading a 256MB file (from tmpfs) on XP64: http://zewt.org/~glenn/samba.tcpdump.gz, 56M/sec That's pretty much what you would expect. The trace shows that your program reads sequentially in 61k chunks. Given My program reads in 1M chunks; it's the OS splitting apart the reads. that your network latency is roughly 600 microseconds, ftp://pserver.samba.org/pub/unpacked/junkcode/expected_throughput.pl gives 55MB/sec. You might want to try smbclient, that does larger parallel reads, or a copy program that uses multiple threads simultaneously. That's what I'd expect if things are as broken as they look from that dump: the client (XP) isn't buffering requests at all, so the window caps out at 64k. The mail from earlier[1] suggests that some people get much better speed from other servers (though without enough information to really be useful). I don't know the protocol, so I can't tell if there's something preventing the client from buffering multiple read requests, for example, which I'd expect of any network FS. Can you get us a network trace of the fast reads from a Windows server? Volker -- SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen phone: +49-551-37-0, fax: +49-551-37-9 AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
On Tue, Mar 15, 2011 at 04:35:32PM -0400, Glenn Maynard wrote: On Tue, Mar 15, 2011 at 3:36 PM, Volker Lendecke volker.lende...@sernet.dewrote: On Tue, Mar 15, 2011 at 02:28:37PM -0400, Glenn Maynard wrote: tcpdump reading a 256MB file (from tmpfs) on XP64: http://zewt.org/~glenn/samba.tcpdump.gz, 56M/sec That's pretty much what you would expect. The trace shows that your program reads sequentially in 61k chunks. Given My program reads in 1M chunks; it's the OS splitting apart the reads. that your network latency is roughly 600 microseconds, ftp://pserver.samba.org/pub/unpacked/junkcode/expected_throughput.pl gives 55MB/sec. You might want to try smbclient, that does larger parallel reads, or a copy program that uses multiple threads simultaneously. That's what I'd expect if things are as broken as they look from that dump: the client (XP) isn't buffering requests at all, so the window caps out at 64k. The mail from earlier[1] suggests that some people get much better speed from other servers (though without enough information to really be useful). I don't know the protocol, so I can't tell if there's something preventing the client from buffering multiple read requests, for example, which I'd expect of any network FS. Multiple threads isn't a good solution (at these speeds, it's liable to severely hurt streaming performance), and in any case it doesn't help actual use, with real applications. Well having multiple threads on the client side seems to allow multiple simultaneous outstanding requests, which is what I think the Intel NAS test does. Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
On Tue, Mar 15, 2011 at 4:43 PM, Volker Lendecke volker.lende...@sernet.dewrote: That's what I'd expect if things are as broken as they look from that dump: the client (XP) isn't buffering requests at all, so the window caps out at 64k. The mail from earlier[1] suggests that some people get much better speed from other servers (though without enough information to really be useful). I don't know the protocol, so I can't tell if there's something preventing the client from buffering multiple read requests, for example, which I'd expect of any network FS. Can you get us a network trace of the fast reads from a Windows server? I don't have any faster servers to compare to. I've assumed for a long time that 50-60MB/sec was simply the best you can do with SMB1. It's the mail I saw earlier (as well as the one below) that made me want to double-check this, to make sure that, for example, there's nothing wrong with my system preventing request buffering. (It's frustrating not being able to move my bulk storage out of my work area because of this--halving the streaming speed of my drives from 100 to 50MB/sec is just too much of a penalty.) Here's another one (with much more information, thankfully), claiming 70MB/sec reads and 100MB/sec writes, specifically with a single stream. That's at least notably better than 55 (which I see for both reads and writes, by the way). It's Win7 on the client, not XP64, but still would be interesting to compare. I'll send a mail off-list asking for a trace, unless someone happens to already know the difference so I don't have to bother someone about an old thread. http://lists.samba.org/archive/samba/2010-June/156653.html -- Glenn Maynard -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
On Tue, Mar 15, 2011 at 04:35:32PM -0400, Glenn Maynard wrote: On Tue, Mar 15, 2011 at 3:36 PM, Volker Lendecke volker.lende...@sernet.dewrote: On Tue, Mar 15, 2011 at 02:28:37PM -0400, Glenn Maynard wrote: tcpdump reading a 256MB file (from tmpfs) on XP64: http://zewt.org/~glenn/samba.tcpdump.gz, 56M/sec That's pretty much what you would expect. The trace shows that your program reads sequentially in 61k chunks. Given My program reads in 1M chunks; it's the OS splitting apart the reads. that your network latency is roughly 600 microseconds, ftp://pserver.samba.org/pub/unpacked/junkcode/expected_throughput.pl gives 55MB/sec. You might want to try smbclient, that does larger parallel reads, or a copy program that uses multiple threads simultaneously. That's what I'd expect if things are as broken as they look from that dump: the client (XP) isn't buffering requests at all, so the window caps out at 64k. The mail from earlier[1] suggests that some people get much better speed from other servers (though without enough information to really be useful). I don't know the protocol, so I can't tell if there's something preventing the client from buffering multiple read requests, for example, which I'd expect of any network FS. It's the client redirector that is limiting this. The Windows SMB1 client will only issue one outstanding read/write per thread on an open file. Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] Performance with XP64
What sort of performance is expected over GigE, with Samba 3 as the server and XP64 as the client? I havn't been able to find any current benchmarks at all. http://lists.samba.org/archive/samba/2010-June/156708.html talks about the same configuration, and suggests that the protocol should be able to break 100MB/sec. (Unfortunately, the poster disappeared without following up.) Focusing on read throughput (Samba to XP64), I see about the same performance as he did, around 50 MB/sec; with HTTP 97.7M/sec. Modern HDDs can regularly hit 100MB/sec, so halving throughput for large file copies is painful. I've always assumed this was just a limitation of the protocol on a link much faster than it was designed for, but that post suggests otherwise. He didn't mention if the Server 2008 system he was comparing against had large frames enabled, though, which I've never had much luck with. (I'll wait until asked before digging out hardware/configuration info, traces and so on; if the performance I'm seeing is already optimal then I won't bother. I'm on 3.4.0, which is what's packaged with my version of Ubuntu; I'll also try upgrading to 3.5.x if that's likely to help.) -- Glenn Maynard -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
On Mon, Mar 14, 2011 at 09:24:21PM -0400, Glenn Maynard wrote: What sort of performance is expected over GigE, with Samba 3 as the server and XP64 as the client? I havn't been able to find any current benchmarks at all. http://lists.samba.org/archive/samba/2010-June/156708.html talks about the same configuration, and suggests that the protocol should be able to break 100MB/sec. (Unfortunately, the poster disappeared without following up.) Focusing on read throughput (Samba to XP64), I see about the same performance as he did, around 50 MB/sec; with HTTP 97.7M/sec. Modern HDDs can regularly hit 100MB/sec, so halving throughput for large file copies is painful. I've worked with OEM's to get more than 100MB/sec on their NAS devices. Using the Intel NAS benchmark. I've always assumed this was just a limitation of the protocol on a link much faster than it was designed for, but that post suggests otherwise. He didn't mention if the Server 2008 system he was comparing against had large frames enabled, though, which I've never had much luck with. (I'll wait until asked before digging out hardware/configuration info, traces and so on; if the performance I'm seeing is already optimal then I won't bother. I'm on 3.4.0, which is what's packaged with my version of Ubuntu; I'll also try upgrading to 3.5.x if that's likely to help.) Have you enabled sendfile ? Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance with XP64
On Mon, Mar 14, 2011 at 9:28 PM, Jeremy Allison j...@samba.org wrote: Have you enabled sendfile ? I hadn't; is there a reason it's not enabled by default? It's an old, reliable API in my experience. It made a small but measurable difference: about 2-3MB/sec. FYI, I'm testing with tmpfs on the Linux side to keep drive speeds out of the picture. -- Glenn Maynard -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance transfer (samba VS ftp)
Volker Lendecke put forth on 9/18/2010 12:44 AM: On Sat, Sep 18, 2010 at 12:22:53AM -0500, Stan Hoeppner wrote: Pol Hallen put forth on 9/15/2010 9:36 AM: debian stable (samba version 2:3.2.5-4lenny9) from clients by ftp the transfer of huge file is about 10/11Mb/s (with an ethernet 10/100) by samba came 5/6Mb/s is it correct? Good luck. It appears that tuning smbd and clients, both Windows and smbclient, to get anywhere close to wire speed is somewhat of a black art. I asked the same question many months ago, and dropped the subject after Jeremy said it had to be a problem with the W2K redirector. Funny thing is, that same W2K redirector can pull at almost wire speed from a WinXP box. The most I've ever been able to get out of smbd is ~8MB/s. I'm running 3.2.5-4lenny12. To get anything better than that I'll have to go to GigE. I probably won't get anywhere close to wire speed, but I should get at least 30-40MB/s, which is 4-5 times what I get now, and would thus be a huge improvement for relatively little cost--a few NICs and a decent desktop GigE switch can be had for around $100 USD. Even without using jumbo frames this would be a huge improvement over 100FDX. As always: What about get/put of large files with smbclient, = 3.2? Hi Volker. I don't have a Linux client machine to test smbclient against my Debian/Samba server. However, running smbclient (3.2.5) on the server and connecting to shares on a WinXP machine and W2K Pro machine hits 11 MB/s (near wire speed of 12.5) all day long with GETing moderate to large files (30MB+). PUTing the same files maxes out at ~6MB/s--very lopsided. I've tried various smbclient -O socket options with no effect on PUT performance. Copying from an smbd share to the Windows machines maxes at 9MB/s. Copying from the Windows machines to an smbd share yields 8MB/s--much more consistent than smbclient. It sure would be nice to have smbclient's 11MB/s GET speed in both directions with all OSes involved. I've tried every option and optimization in smb.conf and the registries of both Windows machines and can't get over 9MB/s. It's sure better than the 5MB/s come people report, so I'm not complaining. It's kinda academic anyway, because the bulk of our transfers to/from smbd are large quantities of small files ( 1MB). Such transfers can crawl at less than 1MB/s. Like I said, I think the best solution for me would be to move to GigE. -- Stan -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] performance transfer (samba VS ftp)
On Sat, Sep 18, 2010 at 1:36 AM, Stan Hoeppner s...@hardwarefreak.comwrote: Volker Lendecke put forth on 9/18/2010 12:44 AM: On Sat, Sep 18, 2010 at 12:22:53AM -0500, Stan Hoeppner wrote: Pol Hallen put forth on 9/15/2010 9:36 AM: debian stable (samba version 2:3.2.5-4lenny9) from clients by ftp the transfer of huge file is about 10/11Mb/s (with an ethernet 10/100) by samba came 5/6Mb/s is it correct? Good luck. It appears that tuning smbd and clients, both Windows and smbclient, to get anywhere close to wire speed is somewhat of a black art. I asked the same question many months ago, and dropped the subject after Jeremy said it had to be a problem with the W2K redirector. Funny thing is, that same W2K redirector can pull at almost wire speed from a WinXP box. The most I've ever been able to get out of smbd is ~8MB/s. I'm running 3.2.5-4lenny12. To get anything better than that I'll have to go to GigE. I probably won't get anywhere close to wire speed, but I should get at least 30-40MB/s, which is 4-5 times what I get now, and would thus be a huge improvement for relatively little cost--a few NICs and a decent desktop GigE switch can be had for around $100 USD. Even without using jumbo frames this would be a huge improvement over 100FDX. As always: What about get/put of large files with smbclient, = 3.2? Hi Volker. I don't have a Linux client machine to test smbclient against my Debian/Samba server. However, running smbclient (3.2.5) on the server and connecting to shares on a WinXP machine and W2K Pro machine hits 11 MB/s (near wire speed of 12.5) all day long with GETing moderate to large files (30MB+). PUTing the same files maxes out at ~6MB/s--very lopsided. I've tried various smbclient -O socket options with no effect on PUT performance. Copying from an smbd share to the Windows machines maxes at 9MB/s. Copying from the Windows machines to an smbd share yields 8MB/s--much more consistent than smbclient. It sure would be nice to have smbclient's 11MB/s GET speed in both directions with all OSes involved. I've tried every option and optimization in smb.conf and the registries of both Windows machines and can't get over 9MB/s. It's sure better than the 5MB/s come people report, so I'm not complaining. It's kinda academic anyway, because the bulk of our transfers to/from smbd are large quantities of small files ( 1MB). Such transfers can crawl at less than 1MB/s. Like I said, I think the best solution for me would be to move to GigE. -- I'm using gigerbit ethernet with samaba 3.4.7 default network settings under ubuntu 10.04 LTS server and last eveing I moved 35 Gigbytes from an iMac to the samba server over gigabit ethernet and it took around 15 minutes which works out to around 300 mega bits per second which is about a third of wire speed which on tests comes out at about 920 Mbits/sec. I have seen mention on this list a while back that the smb protocol is the bottleneck. OTOH I can live with 300Mbits/sec -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance transfer (samba VS ftp)
On Sat, Sep 18, 2010 at 10:27:55AM -0700, grant little wrote: I'm using gigerbit ethernet with samaba 3.4.7 default network settings under ubuntu 10.04 LTS server and last eveing I moved 35 Gigbytes from an iMac to the samba server over gigabit ethernet and it took around 15 minutes which works out to around 300 mega bits per second which is about a third of wire speed which on tests comes out at about 920 Mbits/sec. I have seen mention on this list a while back that the smb protocol is the bottleneck. OTOH I can live with 300Mbits/sec The SMB protocol as such is not the bottleneck. It's the typical client use of it that accounts for many inefficiencies. I'm not saying that smbd could not be improved, but the clients using SMB inefficiently is what makes it slow in the vast majority of the cases I have seen so far. Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] performance transfer (samba VS ftp)
Hi folks :-) situation: debian stable (samba version 2:3.2.5-4lenny9) from clients by ftp the transfer of huge file is about 10/11Mb/s (with an ethernet 10/100) by samba came 5/6Mb/s is it correct? In smb.conf I don't have any strangeoptions: thanks Pol domain master = yes preferred master = yes os level = 65 workgroup = WORKGROUP netbios name = name Server String = name #wins support = yes wins server = x.x.x.x name resolve order = wins #host dns bcast hosts allow = x.x.x.x/x loglevel=2 log file = /var/log/samba/%m.log security=USER encrypt passwords = yes null passwords = no unix extensions = yes ### SHARE ### [...] -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance transfer (samba VS ftp)
Pol Hallen put forth on 9/15/2010 9:36 AM: debian stable (samba version 2:3.2.5-4lenny9) from clients by ftp the transfer of huge file is about 10/11Mb/s (with an ethernet 10/100) by samba came 5/6Mb/s is it correct? Good luck. It appears that tuning smbd and clients, both Windows and smbclient, to get anywhere close to wire speed is somewhat of a black art. I asked the same question many months ago, and dropped the subject after Jeremy said it had to be a problem with the W2K redirector. Funny thing is, that same W2K redirector can pull at almost wire speed from a WinXP box. The most I've ever been able to get out of smbd is ~8MB/s. I'm running 3.2.5-4lenny12. To get anything better than that I'll have to go to GigE. I probably won't get anywhere close to wire speed, but I should get at least 30-40MB/s, which is 4-5 times what I get now, and would thus be a huge improvement for relatively little cost--a few NICs and a decent desktop GigE switch can be had for around $100 USD. Even without using jumbo frames this would be a huge improvement over 100FDX. -- Stan -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance transfer (samba VS ftp)
On Sat, Sep 18, 2010 at 12:22:53AM -0500, Stan Hoeppner wrote: Pol Hallen put forth on 9/15/2010 9:36 AM: debian stable (samba version 2:3.2.5-4lenny9) from clients by ftp the transfer of huge file is about 10/11Mb/s (with an ethernet 10/100) by samba came 5/6Mb/s is it correct? Good luck. It appears that tuning smbd and clients, both Windows and smbclient, to get anywhere close to wire speed is somewhat of a black art. I asked the same question many months ago, and dropped the subject after Jeremy said it had to be a problem with the W2K redirector. Funny thing is, that same W2K redirector can pull at almost wire speed from a WinXP box. The most I've ever been able to get out of smbd is ~8MB/s. I'm running 3.2.5-4lenny12. To get anything better than that I'll have to go to GigE. I probably won't get anywhere close to wire speed, but I should get at least 30-40MB/s, which is 4-5 times what I get now, and would thus be a huge improvement for relatively little cost--a few NICs and a decent desktop GigE switch can be had for around $100 USD. Even without using jumbo frames this would be a huge improvement over 100FDX. As always: What about get/put of large files with smbclient, = 3.2? Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] Testing Samba performance
Hi All -- I have setup a samba share on ubuntu linuxhttp://www.phoronix.com/forums/showthread.php?p=137771# and the intention is to setup a fileserver. I need to test the performancehttp://www.phoronix.com/forums/showthread.php?p=137771# of my samba share i.e. the response time to the user when he browses the share, adds, updates and deletes files and folders. Any help regarding what tests I need to execute and how to interpret the response will be highly appreciated. Thanks Hass. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Performance issues: have eliminated disk and network as cause
Just been told the config file didn't appear in the email as it went out (even though it certainly appears in the copy I've got), so I'm attaching inline this time. Oh, BTW: it's version 3.4.7 on Debian Lenny, installed from backports. [global] workgroup = U4EATECH netbios name = tiamat enable privileges = yes server string = Primary Domain Controller %v security = user local master = no os level = 33 domain master = no preferred master = no encrypt passwords = true null passwords = no hide unreadable = yes hide dot files = yes obey pam restrictions = Yes unix password sync = Yes remote browse sync = 172.30.20.109 172.30.20.130 172.27.0.6 enhanced browsing = yes passwd program = /usr/sbin/smbldap-passwd %u passwd chat = Changing UNIX and samba passwords for*\nNew password* %n\n *Retype new password* %n\n ldap passwd sync = Yes log level = 0 syslog = 1 log file = /var/log/samba/log.%m max log size = 1000 read raw = yes write raw = yes kernel oplocks = yes max xmit = 65535 dead time = 15 use sendfile = yes socket options = TCP_NODELAY SO_KEEPALIVE IPTOS_LOWDELAY getwd cache = yes mangling method = hash2 Dos charset = 850 Unix charset = ISO8859-1 logon script = logon.bat logon path = logon home = \\atlas\%U logon drive = H: domain logons = Yes wins server = 172.30.20.109 #name resolve order = hosts bcast name resolve order = wins lmhosts hosts bcast dns proxy = yes time server = yes passdb backend = ldapsam:ldap://ldap.u4eatech.com/ ldap:// ldap-slave.u4eatech.com ldap admin dn = cn=smbadmin,dc=u4eatech,dc=com ldap suffix = dc=u4eatech,dc=com ldap group suffix = ou=Group ldap user suffix = ou=People ldap machine suffix = ou=Hosts ldap idmap suffix = ou=People ldap ssl = no add user script = /usr/sbin/smbldap-useradd -m %u ldap delete dn = Yes delete user script = /usr/sbin/smbldap-userdel %u add machine script = /usr/sbin/smbldap-useradd -w %u add group script = /usr/sbin/smbldap-groupadd -p %g delete group script = /usr/sbin/smbldap-groupdel %g add user to group script = /usr/sbin/smbldap-groupmod -m %u %g delete user from group script = /usr/sbin/smbldap-groupmod -x %u %g set primary group script = /usr/sbin/smbldap-usermod -g %g %u load printers = no create mask = 0640 directory mask = 0750 nt acl support = Yes guest account = nobody dont descend = /proc,/dev,/etc,/lib,/lost+found,/initrd #show add printer wizard = yes ; to maintain capital letters in shortcuts in any of the profile folders: preserve case = yes short preserve case = yes case sensitive = no [netlogon] path = /home/samba/netlogon guest ok = yes browseable = No read only = no [wpkg] path = /home/samba/wpkg read only = yes guest ok = yes browseable = no [homes] comment = Home Directories browseable = yes writable = yes oplocks = yes GOS Networks Limited, 1 Friary, Temple Quay, Bristol, BS1 6EA, UK. Registered company number: 6917663 The information in this email is confidential and may be legally privileged. It is intended solely for the addressee. Access to this email by anyone else is unauthorised. If you are not the intended recipient, any disclosure, copying, distribution or any action taken or omitted to be taken in reliance on it, is prohibited and may be unlawful. When addressed to our clients any opinions or advice contained in this email are subject to the terms and conditions expressed in the governing GOS Networks agreement. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] Performance issues: have eliminated disk and network as cause
Hi, I'm not entirely happy with the performance I'm seeing using Samba, and I wonder if anyone can shine any light. The server is a Dell PowerEdge 2950 with hardware RAID10, 4GB RAM and a quad-core Intel Xeon processor. It's not live yet, so there's no load from other tasks. I've already eliminated the RAID (able to sustain 130-140MB/s for reads/writes) and the network (GigE, tar | nc to this server and untar'd at the other end sustains 8-900Mbps) as bottlenecks, which leaves me dealing with Samba. Samba is peaking at around 280Mbps (reading and writing a single 500MB file) and normal performance (which I have benchmarked with a 350MB directory containing about 1,000 files of various sizes up to 2MB) is closer to 90-100Mbps (write), 117Mbps (read). This is with a Windows XP client, using smbmount from a Linux client is not appreciably faster. Obviously there's going to be a much larger overhead associated with SMB versus netcat, but 3.5-8 times slower? I have attached my smb.conf (though I have removed most of the shares for brevity's sake), in the hope that someone can help. James. GOS Networks Limited, 1 Friary, Temple Quay, Bristol, BS1 6EA, UK. Registered company number: 6917663 The information in this email is confidential and may be legally privileged. It is intended solely for the addressee. Access to this email by anyone else is unauthorised. If you are not the intended recipient, any disclosure, copying, distribution or any action taken or omitted to be taken in reliance on it, is prohibited and may be unlawful. When addressed to our clients any opinions or advice contained in this email are subject to the terms and conditions expressed in the governing GOS Networks agreement. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance tweaks??
On Tue, 2010-01-19 at 01:15 +0100, Volker Lendecke wrote: On Tue, Jan 19, 2010 at 01:10:41PM +1300, steve wrote: I tried this, and ended up woth 2 pc's that could only see the top level of the shaare. So I reverted it, and they *still* can only see one level down. Ok, well... Maybe you have upper/lower case directories in your shares? Sorry, I had thought that the meaning of those parameters is pretty obvious: They assume that all file names on disk are a specific case. You should add default case = lower if you have everything lower-case. My apologies for not telling you in the first place, Volker As these are microsofties, there is mixed case and white space all over the place. Will this have the desired effect? case sensitive = no preserve case = no short preserve case = no I'm seeing noticeable degraded performance when swapping from the old server, with samba 3.0.22, to a new server with double the memory, 50% more horsepower and 30% faster disks, running 3.2.5. Any suggestions will be gratefully received. Steve -- Steve Holdoway st...@greengecko.co.nz http://www.greengecko.co.nz MSN: st...@greengecko.co.nz GPG Fingerprint = B337 828D 03E1 4F11 CB90 853C C8AB AF04 EF68 52E0 signature.asc Description: This is a digitally signed message part -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance tweaks??
On Wed, Jan 20, 2010 at 3:13 PM, steve st...@greengecko.co.nz wrote: On Tue, 2010-01-19 at 01:15 +0100, Volker Lendecke wrote: On Tue, Jan 19, 2010 at 01:10:41PM +1300, steve wrote: I tried this, and ended up woth 2 pc's that could only see the top level of the shaare. So I reverted it, and they *still* can only see one level down. Ok, well... Maybe you have upper/lower case directories in your shares? Sorry, I had thought that the meaning of those parameters is pretty obvious: They assume that all file names on disk are a specific case. You should add default case = lower if you have everything lower-case. My apologies for not telling you in the first place, Volker As these are microsofties, there is mixed case and white space all over the place. I have a script I use to force the case. Here is a link: http://github.com/drescherjm/jmdgentoooverlay/blob/master/Other/shell-scripts/mvcase.sh John -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance tweaks??
On 2010-01-20 at 15:19 -0500 John Drescher sent off: I have a script I use to force the case. Here is a link: http://github.com/drescherjm/jmdgentoooverlay/blob/master/Other/shell-scripts/mvcase.sh and in case you also have umlauts and other non-ascii characters in file names, you can use convmv, which can also lowercase your files and take care of the encoding, whatever encoding you tell it your files have. Cheers Björn pgpb5W7zzxZFU.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance tweaks??
On Mon, 2010-01-18 at 08:19 +0100, Volker Lendecke wrote: On Mon, Jan 18, 2010 at 11:12:19AM +1300, steve wrote: Has anyone any tips on improving samba performance with debian lenny? I've set up a raid 0 partition on a couple of new WD 1TB disks, formatting the majority of it as a single reiserfs partition. I must admit I should probably have set the blocksize to something other than the default at the time, but now it's well populated, changing stuff like that is going to be a big problem. Performance is pretty poor to put it politely - primary use is a single samba share with a very large directory structure, millions of files. Can anyone suggest any improvements that can stop it being that slow? I regularly see 50+% in waitio. Do you have individual directories with huge numbers of files? This is a known performance problem with a workaround: case sensitive = yes preserve case = no short preserve case = no Volker I tried this, and ended up woth 2 pc's that could only see the top level of the shaare. So I reverted it, and they *still* can only see one level down. Is this some known xp weirdness? Most of the domain are fine. Cheers, Steve -- Steve Holdoway st...@greengecko.co.nz http://www.greengecko.co.nz MSN: st...@greengecko.co.nz GPG Fingerprint = B337 828D 03E1 4F11 CB90 853C C8AB AF04 EF68 52E0 signature.asc Description: This is a digitally signed message part -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance tweaks??
On Tue, Jan 19, 2010 at 01:10:41PM +1300, steve wrote: I tried this, and ended up woth 2 pc's that could only see the top level of the shaare. So I reverted it, and they *still* can only see one level down. Ok, well... Maybe you have upper/lower case directories in your shares? Sorry, I had thought that the meaning of those parameters is pretty obvious: They assume that all file names on disk are a specific case. You should add default case = lower if you have everything lower-case. My apologies for not telling you in the first place, Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] performance tweaks??
Has anyone any tips on improving samba performance with debian lenny? I've set up a raid 0 partition on a couple of new WD 1TB disks, formatting the majority of it as a single reiserfs partition. I must admit I should probably have set the blocksize to something other than the default at the time, but now it's well populated, changing stuff like that is going to be a big problem. Performance is pretty poor to put it politely - primary use is a single samba share with a very large directory structure, millions of files. Can anyone suggest any improvements that can stop it being that slow? I regularly see 50+% in waitio. Version in use samba 3.2.5-4lenny7, and I'd rather not change off the formal release path unless it'll make a big improvement. Cheers, Steve -- Steve Holdoway st...@greengecko.co.nz http://www.greengecko.co.nz MSN: st...@greengecko.co.nz GPG Fingerprint = B337 828D 03E1 4F11 CB90 853C C8AB AF04 EF68 52E0 signature.asc Description: This is a digitally signed message part -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance tweaks??
On Mon, Jan 18, 2010 at 11:12:19AM +1300, steve wrote: Has anyone any tips on improving samba performance with debian lenny? I've set up a raid 0 partition on a couple of new WD 1TB disks, formatting the majority of it as a single reiserfs partition. I must admit I should probably have set the blocksize to something other than the default at the time, but now it's well populated, changing stuff like that is going to be a big problem. Performance is pretty poor to put it politely - primary use is a single samba share with a very large directory structure, millions of files. Can anyone suggest any improvements that can stop it being that slow? I regularly see 50+% in waitio. Do you have individual directories with huge numbers of files? This is a known performance problem with a workaround: case sensitive = yes preserve case = no short preserve case = no Volker signature.asc Description: Digital signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] samba performance multi-thread and multi core
Hi fellows I have question about samba performance with multi-thread and multi core cpu. What can we do for samba performance with multi-thread and multi core ? -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] samba performance multi-thread and multi core
I have question about samba performance with multi-thread and multi core cpu. What can we do for samba performance with multi-thread and multi core ? Each connected user gets their own process and thus threads. The system will balance the threads over the cpus. John -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
[Samba] performance problem with 3.2.8: unbuffered reads for some users
Hi!
I'm experiencing strange performance problems after upgrading to samba
3.2.8 from 3.0.30.
For all users except smbadmin (who has administrative rights), read
performance is _very_ bad. Looking at the read-requests using filemon
and wireshark, I found out that for those users, every read is handled
transparently (unbuffered) over the net. (I.e. a 2 byte read-request of
the application leads to a 2 byte Read And X Request over the net.)
If the user is smbadmin, reads are block buffered. (A 2 byte
read-request of the same application as above leads to a 4096 byte Read
And X Request over the net.)
Clients are WinXP SP3.
For details, see my test below..
When are those buffering parameters negotiated? Do you have any idea why
the behavior depends on the connected user?
Any hints how I could further track down this problem?
Cheers,
--leo
The test was done using 2 byte reads on the windows box:
perl -le 'sysopen(F, R:/firefox/LICENSE, O_RDONLY);
do { $n= sysread(F, $buf, 2) } while ($n)'
The result can be found here:
smbadmin (buffered reads):
http://leo.kloburg.at/tmp/samba/smbadmin-tshark.txt
abergolth (unbuffered reads, same box):
http://leo.kloburg.at/tmp/samba/abergolth-tshark.txt
smb.conf
http://leo.kloburg.at/tmp/samba/smb.conf
--
e-mail ::: Leo.Bergolth (at) wu-wien.ac.at
fax ::: +43-1-31336-906050
location ::: IT-Services | Vienna University of Economics | Austria
--
To unsubscribe from this list go to the following URL and read the
instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On Wed, Mar 04, 2009 at 01:56:26PM +0100, Alexander 'Leo' Bergolth wrote: I'm experiencing strange performance problems after upgrading to samba 3.2.8 from 3.0.30. For all users except smbadmin (who has administrative rights), read performance is _very_ bad. Looking at the read-requests using filemon and wireshark, I found out that for those users, every read is handled transparently (unbuffered) over the net. (I.e. a 2 byte read-request of the application leads to a 2 byte Read And X Request over the net.) If the user is smbadmin, reads are block buffered. (A 2 byte read-request of the same application as above leads to a 4096 byte Read And X Request over the net.) Clients are WinXP SP3. For details, see my test below.. Unfortunately, the log files do not show enough information about what is happening. Simple tshark output is not sufficient, see http://wiki.samba.org/index.php/Capture_Packets for more information on creating useful sniffs. When are those buffering parameters negotiated? Do you have any idea why the behavior depends on the connected user? If it really depends on the connected user, then we need a debug level 10 log of smbd doing it. I would however suspect that this depends on the fact if a file is shared between two users or two applications on the same client box or not. If this is the case, then it is intended behaviour, because the clients lost their oplocks. You might need to restructure your application to not do 2 byte read requests. Volker pgp0XlE9ij9cQ.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On 03/04/2009 02:16 PM, Volker Lendecke wrote:
On Wed, Mar 04, 2009 at 01:56:26PM +0100, Alexander 'Leo' Bergolth wrote:
I'm experiencing strange performance problems after upgrading to samba
3.2.8 from 3.0.30.
For all users except smbadmin (who has administrative rights), read
performance is _very_ bad. Looking at the read-requests using filemon
and wireshark, I found out that for those users, every read is handled
transparently (unbuffered) over the net. (I.e. a 2 byte read-request of
the application leads to a 2 byte Read And X Request over the net.)
If the user is smbadmin, reads are block buffered. (A 2 byte
read-request of the same application as above leads to a 4096 byte Read
And X Request over the net.)
Clients are WinXP SP3.
For details, see my test below..
Unfortunately, the log files do not show enough information
about what is happening. Simple tshark output is not
sufficient, see
http://wiki.samba.org/index.php/Capture_Packets for more
information on creating useful sniffs.
OK, here are more details:
http://leo.kloburg.at/tmp/samba/abergolth-unbuffered.pcap
http://leo.kloburg.at/tmp/samba/smbadmin-buffered.pcap
Both files are produced with
perl -le sysopen(F, \R:/firefox/LICENSE\, O_RDONLY); do { $n=
sysread(F, $buf, 2) } while ($n)
Unfortunately I cannot put the server in debug 10 mode now because there
are some clients connected...
When are those buffering parameters negotiated? Do you have any idea why
the behavior depends on the connected user?
If it really depends on the connected user, then we need a
debug level 10 log of smbd doing it. I would however suspect
that this depends on the fact if a file is shared between
two users or two applications on the same client box or not.
My test case was just reading the firefox LICENSE file, which isn't in
use by any other user. I can reproduce this behavior with arbitrary
other files.
Cheers,
--leo
--
e-mail ::: Leo.Bergolth (at) wu-wien.ac.at
fax ::: +43-1-31336-906050
location ::: IT-Services | Vienna University of Economics | Austria
--
To unsubscribe from this list go to the following URL and read the
instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On Wed, Mar 04, 2009 at 02:50:59PM +0100, Alexander 'Leo' Bergolth wrote: http://leo.kloburg.at/tmp/samba/abergolth-unbuffered.pcap http://leo.kloburg.at/tmp/samba/smbadmin-buffered.pcap The key pieces are frame 704 in the buffered and frame 14 in the unbuffered case. Assuming in both cases the path \firefox\LICENSE refers to the same physical file on disk (the traces don't show enough information to see whether both clients connected to the same share), there is indeed a difference how the server behaves. In the buffered case it grants an oplock, in the unbuffered case it does not, thus the difference in behaviour. \firefox\LICENSE indeed does refer the same physical file, we'd need to see a debug level 10 log. If however the \firefox\LICENSE does not refer to the same physical file on disk, I would highly suspect that in the unbuffered case some other client or another application on the client has this file open. You can see this in the output of smbstatus. Volker pgp5Zv8ZPpErB.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On Wed, Mar 04, 2009 at 02:50:59PM +0100, Alexander 'Leo' Bergolth wrote: My test case was just reading the firefox LICENSE file, which isn't in use by any other user. I can reproduce this behavior with arbitrary other files. Ah, sorry, missed that part. Please send your smb.conf file and a debug level 10 log of the whole unbuffered session. Thanks, Volker P.S: You don't happen to have oplocks = no on some share definition? pgpKv6gaHNPzc.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On 03/04/2009 03:10 PM, Volker Lendecke wrote: On Wed, Mar 04, 2009 at 02:50:59PM +0100, Alexander 'Leo' Bergolth wrote: My test case was just reading the firefox LICENSE file, which isn't in use by any other user. I can reproduce this behavior with arbitrary other files. Ah, sorry, missed that part. Please send your smb.conf file and a debug level 10 log of the whole unbuffered session. When does the session start? Is it sufficient to first establish the connection and then put the corresponding smbd process in debug level 10? The client does a domain logon so capturing the whole login process will be quite huge... P.S: You don't happen to have oplocks = no on some share definition? No. Cheers, --leo -- e-mail ::: Leo.Bergolth (at) wu-wien.ac.at fax ::: +43-1-31336-906050 location ::: IT-Services | Vienna University of Economics | Austria -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
Unfortunately I cannot put the server in debug 10 mode now because there are some clients connected... There is a handly trick for dealing with this: use include files. At the end of the globals section put a: include = /etc/samba/smb.conf.%m Then create a file like: /etc/samba/smb.conf.PC02004 [globals] debug level = 10 - where PC02004 is the client you are testing from. You''ll just get enhanced logs for that client. Samba will pick up the change almost immediately and you can create and remove smb.conf.%m files whenever you want. No down time involved. Very sweet. -- OpenGroupware developer: awill...@whitemice.org http://whitemiceconsulting.blogspot.com/ -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On 03/04/2009 03:10 PM, Volker Lendecke wrote: On Wed, Mar 04, 2009 at 02:50:59PM +0100, Alexander 'Leo' Bergolth wrote: My test case was just reading the firefox LICENSE file, which isn't in use by any other user. I can reproduce this behavior with arbitrary other files. Ah, sorry, missed that part. Please send your smb.conf file and a debug level 10 log of the whole unbuffered session. Here's the log for the unbuffered session: http://leo.kloburg.at/tmp/samba/log.gf2.gz Cheers, --leo -- e-mail ::: Leo.Bergolth (at) wu-wien.ac.at fax ::: +43-1-31336-906050 location ::: IT-Services | Vienna University of Economics | Austria -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On Wed, Mar 04, 2009 at 03:58:20PM +0100, Alexander 'Leo' Bergolth wrote: Here's the log for the unbuffered session: http://leo.kloburg.at/tmp/samba/log.gf2.gz That's the key: [2009/03/04 15:51:48, 3] smbd/oplock_linux.c:linux_set_kernel_oplock(138) linux_set_kernel_oplock: Refused oplock on file Firefox/LICENSE, fd = 28, file_id = fd03:157b181. (Permission denied) Do you have something like SELinux or so? Or do you share the files via NFS and some NFS client has the files open? To work around this problem, you might want to try setting kernel oplocks = no in the global section of your smb.conf file. Volker pgpXLmDQW412Y.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On 03/04/2009 04:16 PM, Volker Lendecke wrote:
On Wed, Mar 04, 2009 at 03:58:20PM +0100, Alexander 'Leo' Bergolth wrote:
Here's the log for the unbuffered session:
http://leo.kloburg.at/tmp/samba/log.gf2.gz
That's the key:
[2009/03/04 15:51:48, 3] smbd/oplock_linux.c:linux_set_kernel_oplock(138)
linux_set_kernel_oplock: Refused oplock on file Firefox/LICENSE, fd = 28,
file_id = fd03:157b181. (Permission denied)
Do you have something like SELinux or so? Or do you share
the files via NFS and some NFS client has the files open?
Hmm. Thanks for tracking this down.
Please help me uderstand why this fails...
Are there any corresponding recent samba or kernel changes?
It fails on Fedora 10 with kernel-PAE-2.6.27.15-170.2.24.fc10.i686 and
samba-3.2.8-0.26.fc10.i386 while it did work with kernel 2.6.22.9
samba-3.0.30.
SELinux is disabled, NFS is not in use.
# selinuxenabled echo yes || echo no
no
# /etc/init.d/nfs status
rpc.mountd is stopped
nfsd is stopped
rpc.rquotad is stopped
According to the source (oplock_linux.c), linux_setlease() does a
F_SETLEASE fcntl call. If that fails, it calls
set_effective_capability(LEASE_CAPABILITY) and tries the same call again.
The strace output of the corresponding part is:
19115 open(Firefox/LICENSE, O_RDONLY|O_LARGEFILE) = 28
19115 fcntl64(28, F_SETSIG, 0x23) = 0
19115 fcntl64(28, 0x400 /* F_??? */, 0x1) = -1 EACCES (Permission denied)
19115 fcntl64(28, 0x400 /* F_??? */, 0x1) = -1 EACCES (Permission denied)
19115 fcntl64(12, F_SETLKW64, {type=F_UNLCK, whence=SEEK_SET,
start=22032, len=1}, 0xbfedbce4) = 0
According to the fcntl man-page, only privileged processes or processes
with the CAP_LEASE capability may do F_SETLEASE:
8
Leases may only be taken out on regular files. An unprivileged process
may only take out a lease on a file whose UID (owner) matches the file
system UID of the process. A process with the CAP_LEASE
capability may take out leases on arbitrary files.
8
The file isn't owned by the user that accesses it, so I guess the
CAP_LEASE capability should be necessary. But shouldn't strace show a
call to capset(2) between those two F_SETLEASE fcntl calls (0x400)?
Cheers,
--leo
--
e-mail ::: Leo.Bergolth (at) wu-wien.ac.at
fax ::: +43-1-31336-906050
location ::: IT-Services | Vienna University of Economics | Austria
--
To unsubscribe from this list go to the following URL and read the
instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On Wed, Mar 04, 2009 at 06:21:27PM +0100, Alexander 'Leo' Bergolth wrote: The file isn't owned by the user that accesses it, so I guess the CAP_LEASE capability should be necessary. But shouldn't strace show a call to capset(2) between those two F_SETLEASE fcntl calls (0x400)? There is code to acquire CAP_LEASE, but this only is enabled if at compile HAVE_POSIX_CAPABILITIES is found. You might want to look at your config.log why this is not detected. Volker pgpOJzEcFakLX.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] performance problem with 3.2.8: unbuffered reads for some users
On 03/04/2009 06:35 PM, Volker Lendecke wrote: On Wed, Mar 04, 2009 at 06:21:27PM +0100, Alexander 'Leo' Bergolth wrote: The file isn't owned by the user that accesses it, so I guess the CAP_LEASE capability should be necessary. But shouldn't strace show a call to capset(2) between those two F_SETLEASE fcntl calls (0x400)? There is code to acquire CAP_LEASE, but this only is enabled if at compile HAVE_POSIX_CAPABILITIES is found. You might want to look at your config.log why this is not detected. Got it! Fedora's RPM spec file is missing a dependency on libcap-devel: http://kojipkgs.fedoraproject.org/packages/samba/3.2.8/0.26.fc10/data/logs/i386/build.log The previously used package was rebuilt by myself with libcap-devel so it did (accidentally) include capabilities support! I've filed a bugreport at redhats bugzilla since this seems to dramatically affect performance. Many thanks for your help! Cheers, --leo P.S.: After having rebuilt the samba package with capabilities, everything works at normal speed again! -- e-mail ::: Leo.Bergolth (at) wu-wien.ac.at fax ::: +43-1-31336-906050 location ::: IT-Services | Vienna University of Economics | Austria -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/options/samba
Re: [Samba] Samba performance issue
Hi Fabien, I haven't tried FUSE, and to be honest I will not. I doubt that the performace will even match cifs or smbfs implementation. I would say that file systems in user space are good for development or special applications, not for general purpose. Anyway I you are to try it out please let me know what you find. Thank you Ales Původní zpráva Od: Fabien azertyz...@free.fr Předmět: Re: [Samba] Samba performance issue Datum: 11.1.2009 00:29:18 Hello, as you say, I also think it would be nice to mention the issue in the documentation to avoid people turning crazy ! If you need the filesystem to be mounted, did you try the fuse implementations ? I found two : SMB for Fuse and usmb. I haven't had time to try this out yet. Regards, Fabien ales...@seznam.cz a écrit : Hello, I can confirm that. I've tried smbclient from Samba 3.3.0rc2. It is considerably faster than cifs kernel module (version 1.52 from Linux 2.6.25). The file system is exported via Samba 3.0.33. Reading 700MB file (residing in the buffer cache) from the server I get: cifs.ko ~ 30MB/s smbclient ~ 74MB/s I needed to remove the NIC IRQ affinity to only one CPU as it got completely saturated at about 66MB/s (1GHz Pentium III). Still think that the machine would do more than 74MB/s, but it is much better than 30MB/s anyway. Unfortunately I need the share mounted as a regular file system, so I'm gonna stick with the kernel implementation. One day I will try using jumbo frames, but for the moment I'm stuck with 1500b frames. In regards to the question Fabien has raised - I think Steven French has the performance optimization in TODO section for the cifs module, but he only knows when he gets there. Haven't asked him though. Anyway thanks for the help. Regards Aleš P.S.: The performance problems of various SMB/CIFS clients should probably be mentioned in Samba docs, that would prevent people from asking the same questions over and over again. Might even force Steven to fix the cifs kernel module ;-) Původní zpráva Od: Volker Lendecke volker.lende...@sernet.de Předmět: Re: [Samba] Samba performance issue Datum: 05.1.2009 21:42:42 On Mon, Jan 05, 2009 at 08:25:34PM +0100, Fabien wrote: I've seen I'm not the only one impacted with this issue these times on the mailing list :) I did the following test (Debian packages) : Server Client : samba 3.2.5 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~80Mo/s Server Client : samba 3.0.24 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~60Mo/s This is the first time I try smbclient. There is a real big difference between mount and smbclient ! And it seems to be better to use the 3.2.5 version which is ~ 20Mo/s better than the 3.0.24 version. Again, all of this was tested without using the disks (buffer cache). Do you know where does this difference comes from ? It's the latencies that kill performance. Given the request-response nature of the protocol with a limited request size (no matter how large you make them), you can only get a certain number of round trips per second. smbclient 3.2 and even more in upcoming 3.3 hides those latencies by issuing more than one request at the same time using the Multiplex ID field in the SMB header properly. Neither cifs nor smbfs do this. Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
[Samba] Performance problem when tagging mp3 files
Hello you, maybe you can help me with a performance problem i´ve when tagging mp3 files from a windows vista system where the files are located on a samba share hosted by a suse linux 9.3 (samba releases is 3.0.23). these are my performance relevant settings. # These settings are a suggestion for a local network. Cf. section # 'socket options' in the man page of smb.conf and socket(7). time server = yes getwd cache = yes level2 oplocks = yes preserve case = yes case sensitive = no map archive = no socket options = SO_KEEPALIVE IPTOS_LOWDELAY TCP_NODELAY the performance when transferring large files is great in my opinion. I´ve hand stopped 72 seconds for transferring about 800MB to a share(in a 100MBit network). But tagging mp3 files is incredibly slow. For example: If i just change the album tag of 10 mp3 files and save the result, saving lasts about 30 seconds or more. And my samba log (with log level 2) shows something like: [2008/08/24 10:01:04, 2] smbd/open.c:open_file(352) js opened file mp3/Rock/U2/U218 Singles/U2 And Green Day - U218 Singles - 17 - The Saints Are Coming.mp3 read=Yes write=No (numopen=1) [2008/08/24 10:01:04, 2] smbd/close.c:close_normal_file(344) ftp closed file mp3/Rock/U2/U218 Singles/U2 And Green Day - U218 Singles - 17 - The Saints Are Coming.mp3 (numopen=0) [2008/08/24 10:01:04, 2] smbd/open.c:open_file(352) js opened file mp3/Rock/U2/U218 Singles/U2 And Green Day - U218 Singles - 17 - The Saints Are Coming.mp3 read=Yes write=Yes (numopen=1) [2008/08/24 10:01:07, 2] smbd/close.c:close_normal_file(344) ftp closed file mp3/Rock/U2/U218 Singles/U2 And Green Day - U218 Singles - 17 - The Saints Are Coming.mp3 (numopen=0) [2008/08/24 10:01:07, 2] smbd/open.c:open_file(352) js opened file mp3/Rock/U2/U218 Singles/U2 And Green Day - U218 Singles - 17 - The Saints Are Coming.mp3 read=Yes write=No (numopen=1) there is this strange 3 seconds delay on the write operation? I must say that the share has force user ftp and force group daemon and myself i logged on the windows system with the account js. Can i do anything against this problem? Greetings, Jörg. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
Hello, I can confirm that. I've tried smbclient from Samba 3.3.0rc2. It is considerably faster than cifs kernel module (version 1.52 from Linux 2.6.25). The file system is exported via Samba 3.0.33. Reading 700MB file (residing in the buffer cache) from the server I get: cifs.ko ~ 30MB/s smbclient ~ 74MB/s I needed to remove the NIC IRQ affinity to only one CPU as it got completely saturated at about 66MB/s (1GHz Pentium III). Still think that the machine would do more than 74MB/s, but it is much better than 30MB/s anyway. Unfortunately I need the share mounted as a regular file system, so I'm gonna stick with the kernel implementation. One day I will try using jumbo frames, but for the moment I'm stuck with 1500b frames. In regards to the question Fabien has raised - I think Steven French has the performance optimization in TODO section for the cifs module, but he only knows when he gets there. Haven't asked him though. Anyway thanks for the help. Regards Aleš P.S.: The performance problems of various SMB/CIFS clients should probably be mentioned in Samba docs, that would prevent people from asking the same questions over and over again. Might even force Steven to fix the cifs kernel module ;-) Původní zpráva Od: Volker Lendecke volker.lende...@sernet.de Předmět: Re: [Samba] Samba performance issue Datum: 05.1.2009 21:42:42 On Mon, Jan 05, 2009 at 08:25:34PM +0100, Fabien wrote: I've seen I'm not the only one impacted with this issue these times on the mailing list :) I did the following test (Debian packages) : Server Client : samba 3.2.5 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~80Mo/s Server Client : samba 3.0.24 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~60Mo/s This is the first time I try smbclient. There is a real big difference between mount and smbclient ! And it seems to be better to use the 3.2.5 version which is ~ 20Mo/s better than the 3.0.24 version. Again, all of this was tested without using the disks (buffer cache). Do you know where does this difference comes from ? It's the latencies that kill performance. Given the request-response nature of the protocol with a limited request size (no matter how large you make them), you can only get a certain number of round trips per second. smbclient 3.2 and even more in upcoming 3.3 hides those latencies by issuing more than one request at the same time using the Multiplex ID field in the SMB header properly. Neither cifs nor smbfs do this. Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
Hello, as you say, I also think it would be nice to mention the issue in the documentation to avoid people turning crazy ! If you need the filesystem to be mounted, did you try the fuse implementations ? I found two : SMB for Fuse and usmb. I haven't had time to try this out yet. Regards, Fabien ales...@seznam.cz a écrit : Hello, I can confirm that. I've tried smbclient from Samba 3.3.0rc2. It is considerably faster than cifs kernel module (version 1.52 from Linux 2.6.25). The file system is exported via Samba 3.0.33. Reading 700MB file (residing in the buffer cache) from the server I get: cifs.ko ~ 30MB/s smbclient ~ 74MB/s I needed to remove the NIC IRQ affinity to only one CPU as it got completely saturated at about 66MB/s (1GHz Pentium III). Still think that the machine would do more than 74MB/s, but it is much better than 30MB/s anyway. Unfortunately I need the share mounted as a regular file system, so I'm gonna stick with the kernel implementation. One day I will try using jumbo frames, but for the moment I'm stuck with 1500b frames. In regards to the question Fabien has raised - I think Steven French has the performance optimization in TODO section for the cifs module, but he only knows when he gets there. Haven't asked him though. Anyway thanks for the help. Regards Aleš P.S.: The performance problems of various SMB/CIFS clients should probably be mentioned in Samba docs, that would prevent people from asking the same questions over and over again. Might even force Steven to fix the cifs kernel module ;-) Původní zpráva Od: Volker Lendecke volker.lende...@sernet.de Předmět: Re: [Samba] Samba performance issue Datum: 05.1.2009 21:42:42 On Mon, Jan 05, 2009 at 08:25:34PM +0100, Fabien wrote: I've seen I'm not the only one impacted with this issue these times on the mailing list :) I did the following test (Debian packages) : Server Client : samba 3.2.5 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~80Mo/s Server Client : samba 3.0.24 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~60Mo/s This is the first time I try smbclient. There is a real big difference between mount and smbclient ! And it seems to be better to use the 3.2.5 version which is ~ 20Mo/s better than the 3.0.24 version. Again, all of this was tested without using the disks (buffer cache). Do you know where does this difference comes from ? It's the latencies that kill performance. Given the request-response nature of the protocol with a limited request size (no matter how large you make them), you can only get a certain number of round trips per second. smbclient 3.2 and even more in upcoming 3.3 hides those latencies by issuing more than one request at the same time using the Multiplex ID field in the SMB header properly. Neither cifs nor smbfs do this. Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
Thanks for the information. Do you know why the smbclient, although faster, is not fast enough to go over 80Mo/s ? Is there any plan to do the fiddly work on the smbfs implementation to make it as fast as smbclient ? :) I didn't try the fuse implemtations yet. I found two : SMB for Fuse and usmb. I going to try them when possible. Has anyone already tried them ? Thanks ! Fabien Volker Lendecke a écrit : I did the following test (Debian packages) : Server Client : samba 3.2.5 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~80Mo/s Server Client : samba 3.0.24 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~60Mo/s It's the latencies that kill performance. Given the request-response nature of the protocol with a limited request size (no matter how large you make them), you can only get a certain number of round trips per second. smbclient 3.2 and even more in upcoming 3.3 hides those latencies by issuing more than one request at the same time using the Multiplex ID field in the SMB header properly. Neither cifs nor smbfs do this. Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Thu, Jan 08, 2009 at 03:27:53PM +0100, Fabien wrote: Thanks for the information. Do you know why the smbclient, although faster, is not fast enough to go over 80Mo/s ? No, not from the top of my head. This needs much closer investigation. Volker pgpCbC6MT7n10.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Thu, Jan 08, 2009 at 03:27:53PM +0100, Fabien wrote: Thanks for the information. Do you know why the smbclient, although faster, is not fast enough to go over 80Mo/s ? Is there any plan to do the fiddly work on the smbfs implementation to make it as fast as smbclient ? :) smbfs is dead. CIFSFS is under active development. Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Thu, 2009-01-08 at 08:27 -0800, Jeremy Allison wrote: On Thu, Jan 08, 2009 at 03:27:53PM +0100, Fabien wrote: Thanks for the information. Do you know why the smbclient, although faster, is not fast enough to go over 80Mo/s ? Is there any plan to do the fiddly work on the smbfs implementation to make it as fast as smbclient ? :) smbfs is dead. CIFSFS is under active development. Is the fiddly work being done in CIFSFS? Or planned? -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Thu, Jan 08, 2009 at 09:24:16AM -0800, rhubbell wrote: On Thu, 2009-01-08 at 08:27 -0800, Jeremy Allison wrote: On Thu, Jan 08, 2009 at 03:27:53PM +0100, Fabien wrote: Thanks for the information. Do you know why the smbclient, although faster, is not fast enough to go over 80Mo/s ? Is there any plan to do the fiddly work on the smbfs implementation to make it as fast as smbclient ? :) smbfs is dead. CIFSFS is under active development. Is the fiddly work being done in CIFSFS? Or planned? Talk to Steve French and Jeff Layton, who are the most active developers on CIFSFS. Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Mon, 2009-01-05 at 21:47 +0100, Volker Lendecke wrote: On Mon, Jan 05, 2009 at 08:25:34PM +0100, Fabien wrote: I've seen I'm not the only one impacted with this issue these times on the mailing list :) I did the following test (Debian packages) : Server Client : samba 3.2.5 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~80Mo/s Server Client : samba 3.0.24 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~60Mo/s This is the first time I try smbclient. There is a real big difference between mount and smbclient ! And it seems to be better to use the 3.2.5 version which is ~ 20Mo/s better than the 3.0.24 version. Again, all of this was tested without using the disks (buffer cache). Do you know where does this difference comes from ? It's the latencies that kill performance. Given the request-response nature of the protocol with a limited request size (no matter how large you make them), you can only get a certain number of round trips per second. smbclient 3.2 and even more in upcoming 3.3 hides those latencies by issuing more than one request at the same time using the Multiplex ID field in the SMB header properly. Neither cifs nor smbfs do this. Why do cifs and smbfs not have this capability? Is it too much work? Is it due to differences in the purpose of each? Is there a way to setup smbclient to act like a mount point acts? I'm pretty sure the answer's No. but I ask anyway. Is this definition correct? Multiplex ID: Used by the server to verify the file access permissions of groups where consumer-based file protection is in effect. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Tue, Jan 06, 2009 at 09:35:39AM -0800, rhubbell wrote: Why do cifs and smbfs not have this capability? Is it too much work? Is it due to differences in the purpose of each? It's fiddly work nobody has done yet. Is there a way to setup smbclient to act like a mount point acts? I'm pretty sure the answer's No. but I ask anyway. What do you mean by that? You want to slow down smbclient? Is this definition correct? Multiplex ID: Used by the server to verify the file access permissions of groups where consumer-based file protection is in effect. No, that is not correct. See http://go.microsoft.com/fwlink/?LinkId=89836 for a description of CIFS, alternatively look at http://ubiqx.org/cifs. Volker pgpnhqJxuFGqb.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Tue, 2009-01-06 at 19:20 +0100, Volker Lendecke wrote: On Tue, Jan 06, 2009 at 09:35:39AM -0800, rhubbell wrote: Why do cifs and smbfs not have this capability? Is it too much work? Is it due to differences in the purpose of each? It's fiddly work nobody has done yet. fiddly = not hard work, but tedious and sort of annoying? Is there a way to setup smbclient to act like a mount point acts? I'm pretty sure the answer's No. but I ask anyway. What do you mean by that? You want to slow down smbclient? Ha, lol, no. My question was probably ridiculous beyond comprehension. Was asking if there was a way to make use of smbclient to replace cifs or smbfs. Is this definition correct? Multiplex ID: Used by the server to verify the file access permissions of groups where consumer-based file protection is in effect. No, that is not correct. See http://go.microsoft.com/fwlink/?LinkId=89836 for a description of CIFS, alternatively look at http://ubiqx.org/cifs. Thanks for those links, that definition I found definitely had me scratching my head. Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Tue, Jan 06, 2009 at 10:41:55AM -0800, rhubbell wrote: Why do cifs and smbfs not have this capability? Is it too much work? Is it due to differences in the purpose of each? It's fiddly work nobody has done yet. fiddly = not hard work, but tedious and sort of annoying? Fiddly as in not many of lines of code, but code with complex interactions and data dependencies. So I'd say for me it is hard work, others might find it easier. But as nobody has done it yet, I'd say I'm not completely alone in that assessment. Volker pgpvVWEcp88kz.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Tue, 2009-01-06 at 20:25 +0100, Volker Lendecke wrote: On Tue, Jan 06, 2009 at 10:41:55AM -0800, rhubbell wrote: Why do cifs and smbfs not have this capability? Is it too much work? Is it due to differences in the purpose of each? It's fiddly work nobody has done yet. fiddly = not hard work, but tedious and sort of annoying? Fiddly as in not many of lines of code, but code with complex interactions and data dependencies. So I'd say for me it is hard work, others might find it easier. But as nobody has done it yet, I'd say I'm not completely alone in that assessment. Ah, ok, understood, thanks. So it falls into the class of enhancements under the heading Worthy but tricky. -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
Is there a way to setup smbclient to act like a mount point acts? I'm pretty sure the answer's No. but I ask anyway. What do you mean by that? You want to slow down smbclient? Ha, lol, no. My question was probably ridiculous beyond comprehension. Was asking if there was a way to make use of smbclient to replace cifs or smbfs. I would also like to know the answer to this. Can I use smbclient to create a mount point on the unix filesystem the way I can with mount.cifs? thanks, -wes -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Tue, Jan 06, 2009 at 11:45:33AM -0800, wes wrote: Ha, lol, no. My question was probably ridiculous beyond comprehension. Was asking if there was a way to make use of smbclient to replace cifs or smbfs. I would also like to know the answer to this. Can I use smbclient to create a mount point on the unix filesystem the way I can with mount.cifs? No, you can't. Your options are mount.cifs, mount.smbfs, sharity (nfs2smb converter) and some fuse-based file systems. And then there are the pure user-space VFS implementations of gnome and kde and possibly others. But they only work for the applications using them. Volker pgpubbDt8pwHy.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
Hallo, wes, Du (samba) meintest am 06.01.09: Can I use smbclient to create a mount point on the unix filesystem the way I can with mount.cifs? That's the job of mkdir. No other program. Viele Gruesse! Helmut -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
Hello, I've seen I'm not the only one impacted with this issue these times on the mailing list :) I did the following test (Debian packages) : Server Client : samba 3.2.5 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~80Mo/s Server Client : samba 3.0.24 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~60Mo/s This is the first time I try smbclient. There is a real big difference between mount and smbclient ! And it seems to be better to use the 3.2.5 version which is ~ 20Mo/s better than the 3.0.24 version. Again, all of this was tested without using the disks (buffer cache). Do you know where does this difference comes from ? Fabien Volker Lendecke a écrit : On Sat, Jan 03, 2009 at 02:26:01AM +0100, Fabien wrote: I'm gonna try that and post the results here as soon as possible. Do you think it could really make a difference knowing that I also tried the WindowsXP native client without being able to notice any difference ? Not sure, but for me it makes a difference. Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Mon, Jan 05, 2009 at 08:25:34PM +0100, Fabien wrote: I've seen I'm not the only one impacted with this issue these times on the mailing list :) I did the following test (Debian packages) : Server Client : samba 3.2.5 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~80Mo/s Server Client : samba 3.0.24 mount -t smbfs : ~35Mo/s mount -t cifs : ~35Mo/s smbclient : ~60Mo/s This is the first time I try smbclient. There is a real big difference between mount and smbclient ! And it seems to be better to use the 3.2.5 version which is ~ 20Mo/s better than the 3.0.24 version. Again, all of this was tested without using the disks (buffer cache). Do you know where does this difference comes from ? It's the latencies that kill performance. Given the request-response nature of the protocol with a limited request size (no matter how large you make them), you can only get a certain number of round trips per second. smbclient 3.2 and even more in upcoming 3.3 hides those latencies by issuing more than one request at the same time using the Multiplex ID field in the SMB header properly. Neither cifs nor smbfs do this. Volker pgpPjH7wyRn3x.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
[Samba] Samba performance issue
Hello, smbclient seems to be really better than mount (cifs smbfs). Have a look on my thread :) Fabien -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
I'm gonna try that and post the results here as soon as possible. Do you think it could really make a difference knowing that I also tried the WindowsXP native client without being able to notice any difference ? I must also say that I used cifs for my tests (mount -t cifs). Thanks again, Fabien Volker Lendecke a écrit : On Thu, Jan 01, 2009 at 07:35:06PM +0100, Fabien wrote: * My server config : - AMD X2 4200+ - 2 Go RAM - 4 x 500 Go -- RAID5 - Gigabyte connection - Debian ETCH - debian package : Samba 3.0.24 (I also tried to backport the testing version = 3.2.5 but the results were exactly the same) Can you try the smbclient from 3.2.5 to get a large file and see if that performs better? Thanks, Volker -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
[Samba] Samba performance issue
Hi there, I've got similar problem as Fabien. The configuration is as follows: server: 2x Intel Pentium III @ 1GHz 1GB RAM Compaq SmartArray 431 RAID controller Seagate Medalist 3.2GB - system disk 2x Seagate Cheetah, 18GB, 15k RPM in RAID 0 - Samba share Intel 82540EM GbE NIC client: Compaq NC6220 laptop Intel Pentium M @ 1.73GHz 512MB RAM Fujitsu 40MB ATA Broadcom BCM5751M GbE NIC switch: Dlink DGS-1005D 5-port, unmanaged GbE switch Both computers run Gentoo Linux 2008, kernel 2.6.25-r9, server runs Samba 3.0.33, client mount.cifs 3.0.30. The underlying filesystem for Samba is Ext3 with xattr and acls. I wasn't able to break 32MB/s (250Mbps) transfer speed neither reading nor writing to the server. The disk subsystem of the server is capable of 60MB/s and generaly the hardware is not the bottleneck. Neither is the network - the bw_tcp from LMbench suite shows around 108MB/s with 1500b messages, which is what I would expect from GbE TCP/IP. I've been tinkering with very much all the knobs the linux's TCP/IP stack has, the same goes for smb.conf - to no avail. The only thing I couldn't test are Jumbo frames, because the BCM5751M doesn't support them. Unfortunately I can't post my smb.conf, as I am off today, but I can post it later. However, I have find out that having several reads (or writes) pending increases the transfer speed up to 108MB/s, which proves that the hardware is not the bottleneck: I've created a 500MB file on the server and as it fits into the buffer cache the disk subsystem's limits are eliminated. Then I opened it ten times with 'dd if=/mnt/samba/examplefile of=/dev/null ' and this way I was able to saturate the GbE network. But I would like to read and write to Samba close to the hw speed within one tcp session. I was not able to manage that and after some googling on the Internet I've found out that I'm not the only one suffering from this issue. Does Samba scale to gigabit speed within one opened tcp session? Limitition of SMB protocol? Or is it linux TCP/IP stack issue? Any suggestions? Thank you in advance Ales Blaha -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Fri, Jan 02, 2009 at 05:54:11PM +0100, Aleš Bláha wrote: Both computers run Gentoo Linux 2008, kernel 2.6.25-r9, server runs Samba 3.0.33, client mount.cifs 3.0.30. The underlying filesystem for Samba is Ext3 with xattr and acls. I wasn't able to break 32MB/s (250Mbps) transfer speed neither reading nor writing to the server. The disk subsystem of the server is capable of 60MB/s and generaly the hardware is not the bottleneck. Neither is the network - the bw_tcp from LMbench suite shows around 108MB/s with 1500b messages, which is what I would expect from GbE TCP/IP. I've been tinkering with In a test I did lately it made a huge difference if I just did raw TCP benchmarks, raw disk benchmarks or a combined one. The test I used was netcat -l -p diskfile on the receiving end and netcat server-ip diskfile on the sending end. This made my hardware which would otherwise happily saturate gigE crawl down to something like 50MB/sec. Can you try that? Volker pgpjKUD24NYHS.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
Hi Volker, Thank you for your help. I will try what you propose as soon as I get to the machines. But, to be honest, I don't think, the hardware is the bottleneck. The RAID controller and the NIC in the server sit on a different PCI bus and each one has its interrupt hooked to a different CPU. Appart from that, as I've mentioned in the previous post, I can saturate the network when copying files that are cached on the server - but only as long as there are several pending requests - with only one I get just those 30MBs or so (one tcp session vs. couple of them). I will do some more benchmarks next week, post the smb.conf and tcp/ip stack config. Regards Ales On Fri, 02 Jan 2009 20:48:53 +0100 (CET) Volker Lendecke volker.lende...@sernet.de wrote: On Fri, Jan 02, 2009 at 05:54:11PM +0100, Aleš Bláha wrote: Both computers run Gentoo Linux 2008, kernel 2.6.25-r9, server runs Samba 3.0.33, client mount.cifs 3.0.30. The underlying filesystem for Samba is Ext3 with xattr and acls. I wasn't able to break 32MB/s (250Mbps) transfer speed neither reading nor writing to the server. The disk subsystem of the server is capable of 60MB/s and generaly the hardware is not the bottleneck. Neither is the network - the bw_tcp from LMbench suite shows around 108MB/s with 1500b messages, which is what I would expect from GbE TCP/IP. I've been tinkering with In a test I did lately it made a huge difference if I just did raw TCP benchmarks, raw disk benchmarks or a combined one. The test I used was netcat -l -p diskfile on the receiving end and netcat server-ip diskfile on the sending end. This made my hardware which would otherwise happily saturate gigE crawl down to something like 50MB/sec. Can you try that? Volker -- -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Fri, Jan 02, 2009 at 09:57:34PM +0100, Aleš Bláha wrote: Thank you for your help. I will try what you propose as soon as I get to the machines. But, to be honest, I don't think, the hardware is the bottleneck. The RAID controller and the NIC in the server sit on a different PCI bus and each one has its interrupt hooked to a different CPU. Appart from that, as I've mentioned in the previous post, I can saturate the network when copying files that are cached on the server - but only as long as there are several pending requests - with only one I get just those 30MBs or so (one tcp session vs. couple of them). I will do some more benchmarks next week, post the smb.conf and tcp/ip stack config. One thing you might also want to check is get/put operations with latest smbclient from the git master branch. This does almost optimal streaming smb operations. Volker pgpFcGD0pFnnJ.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Fri, 02 Jan 2009 22:01:49 +0100 (CET) Volker Lendecke volker.lende...@sernet.de wrote: On Fri, Jan 02, 2009 at 09:57:34PM +0100, Aleš Bláha wrote: Thank you for your help. I will try what you propose as soon as I get to the machines. But, to be honest, I don't think, the hardware is the bottleneck. The RAID controller and the NIC in the server sit on a different PCI bus and each one has its interrupt hooked to a different CPU. Appart from that, as I've mentioned in the previous post, I can saturate the network when copying files that are cached on the server - but only as long as there are several pending requests - with only one I get just those 30MBs or so (one tcp session vs. couple of them). I will do some more benchmarks next week, post the smb.conf and tcp/ip stack config. One thing you might also want to check is get/put operations with latest smbclient from the git master branch. This does almost optimal streaming smb operations. Volker Hello Volker, I will try that also, that is after I find out how to build it from the git source. Thanks Ales -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
[Samba] Samba performance issue
Hello, I sent the following message to the Debian folks. They don't think that the Debian packaging could be responsible for the issue described there. Well, I'm not completely convinced that we will have very useful input for you. I don't really see any reason for this to be caused by the Debian packaging. To check this, why not compile samba from sources, install it in /usr/local and run the same tests? I also recommend you bring this problem in the samba users mailing lists (samba@lists.samba.org seems appropriate). I must say I didn't try to compile it from the upstream sources for many reasons. Anyway, I think that there is poor chances for the issue to be related with the Debian package. Thank you in advance for your advices ;) Happy new year ! (yeah, it depends on where you are in the world !!) Fabien -- Hello, sorry to write directly to you, but after long hours spent trying to solve this strange problem, I thought it could be handled better by speaking directly with you. I hope I'm not wrong... While testing samba speed on my file server, using a fully tested gigabyte ethernet connexion, I encontered some very strange performance problems. I had big speed differences when using samba locally, or via the network (gigabyte). Mb stands for Mbytes. * My server config : - AMD X2 4200+ - 2 Go RAM - 4 x 500 Go -- RAID5 - Gigabyte connection - Debian ETCH - debian package : Samba 3.0.24 (I also tried to backport the testing version = 3.2.5 but the results were exactly the same) * My samba share folder is a partition of my RAID 5 (reading : 170 Mb/s writing : 90 Mb/s). * Here are my benchmarks (nearly the same from Linux or Windows) : - Reading via ftp on the server (no samba) from a Gigabyte client : = 120Mb/s - Reading on the samba server from a Gigabyte client = 35Mb/s - Writting via ftp on the server (no samba) from a Gigabyte client : = 90Mb/s - Writing on the samba server from a Gigabyte client : = 40Mb/s I also noticed that if I try to read/write on the samba server from more than just one client, the results is exactly the same : the server bandwith limit is 35(up)/40(down) Mb/s. /!\ When I mount the samba partition localy (on the server) I can read at 115 Mb/s or write at 90 Mb/s, which is the good behavior. I tried MS Windows and Linux (etch) smbfs clients. I did a lot of testing, using a 1Go RAM disk on both sides to avoid disks problems, and got the same test results. Do you think that the problem could come from the Debian package ? I investigated google a lot and found very few people reporting this strange behavior (mainly from ubuntu forums). I am pretty sure that several (every ?) people, at least from debian, are concerned. I think that they just didn't notice the speed problem (35Mo/s is still quite nice to be noticeable). I found some people trying to tweak the samba buffers : they only got a little better speed (+ 5Mo/s). My smb.conf file is attached. Thanks for your help. Cheers, Fabien [global] workgroup = MSHOME server string = zeus interfaces = eth1 bind interfaces only = Yes security = SHARE guest account = sambauser name resolve order = host wins bcast load printers = No os level = 0 preferred master = No local master = No domain master = No invalid users = root force group = sambashare [private_data] comment = private_data path = /data1 read only = No create mask = 0770 directory mask = 0770 [common_data] comment = common_data path = /data2 read only = No create mask = 0775 directory mask = 0775 guest ok = Yes -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] Samba performance issue
On Thu, Jan 01, 2009 at 07:35:06PM +0100, Fabien wrote: * My server config : - AMD X2 4200+ - 2 Go RAM - 4 x 500 Go -- RAID5 - Gigabyte connection - Debian ETCH - debian package : Samba 3.0.24 (I also tried to backport the testing version = 3.2.5 but the results were exactly the same) Can you try the smbclient from 3.2.5 to get a large file and see if that performs better? Thanks, Volker pgptHifpmfoVo.pgp Description: PGP signature -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
[Samba] performance problem with access database
Hello I have problem with a access application, when I try to start the application then I must wait 5 minutes ago before he started. I do this from a WinXp Workstation to a Linux Debian Etch and samba 3.0.24 installation. What can I do for better performance. best regards pat -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
Re: [Samba] performance problem with access database
Scheidegger Patrick wrote: Hello I have problem with a access application, when I try to start the application then I must wait 5 minutes ago before he started. I do this from a WinXp Workstation to a Linux Debian Etch and samba 3.0.24 installation. What can I do for better performance. best regards pat If you've got more than a handful of users at any given moment, you can disable op-locks and reduce locking overhead. You can do this via registry, Samba, or both. Also, a database (and I use that in the loosest sense of the term!) compact and repair never hurt ;) -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
