NFS Performance: Weirder And Weirder
This is really weird. A FreeBSD 9.1 system mounts the following: /dev/ad4s1a989M625M285M69%/ devfs 1.0k1.0k 0B 100%/dev /dev/ad4s1d7.8G 1G6.1G14%/var /dev/ad4s1e 48G9.4G 35G21%/usr /dev/ad4s1f390G127G231G35%/usr1 /dev/ad6s1d902G710G120G86%/usr1/BKU /usr1/something (under ad4s1f) and /usr1/BKU (all of ad6s1d) are exported for NFS mounting on the LAN. I have tested the speeds of these two drives locally doing a 'dd if=/dev/zero '. Their speeds are quite comparable - around 55-60 MB/s so the problem below is not an artifact of a slow drive. The two mounts are imported like this on a Linux Mint 12 machine: machine:/usr1/BKU /BKU nfs rw,soft,intr 0 0 machine:/usr1/shared /shared nfs rw,soft,intr 0 0 Problem: When I write files from the LM12 machines to /BKU the writes are 1/10 the speed of when writing to /shared. Reads are fine in both cases, at near native disk speeds being reported. Someone here suggested I get rid of any symlinks in the mount and I did that to no avail. Incidentally, the only reason I just noticed this is that I upgraded the NIC on the FreeBSD machine and the switch into which it connects to 1000Base because the LM12 machine had a built in 1000Base NIC. I also changed the cables on both machines to ensure they were not the problem. Prior to this, I was bandwidth constrained by the 100Base so I never saw NFS performance as an issue. When I upgraded, I expected faster transfers and when I didn't get them, I started this whole investigation. So ... I'm stumped: - It's not the drive or SATA ports because both drives show comparable performance. - It's not the cables because I can get great throughput on one of the NFS mountpoints. - It's neither NIC for the same reason. Does anyone: A) Have a clue what might be doing this B) Have a suggestion how to track down the problem Thanks, -- Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS Performance: Weirder And Weirder
On Sat, Mar 16, 2013 at 11:49 AM, Tim Daneliuk tun...@tundraware.comwrote: This is really weird. A FreeBSD 9.1 system mounts the following: /dev/ad4s1a989M625M285M69%/ devfs 1.0k1.0k 0B 100%/dev /dev/ad4s1d7.8G 1G6.1G14%/var /dev/ad4s1e 48G9.4G 35G21%/usr /dev/ad4s1f390G127G231G35%/usr1 /dev/ad6s1d902G710G120G86%/usr1/BKU /usr1/something (under ad4s1f) and /usr1/BKU (all of ad6s1d) are exported for NFS mounting on the LAN. I have tested the speeds of these two drives locally doing a 'dd if=/dev/zero '. Their speeds are quite comparable - around 55-60 MB/s so the problem below is not an artifact of a slow drive. The two mounts are imported like this on a Linux Mint 12 machine: machine:/usr1/BKU /BKU nfs rw,soft,intr 0 0 machine:/usr1/shared /shared nfs rw,soft,intr 0 0 Problem: When I write files from the LM12 machines to /BKU the writes are 1/10 the speed of when writing to /shared. Reads are fine in both cases, at near native disk speeds being reported. Someone here suggested I get rid of any symlinks in the mount and I did that to no avail. Incidentally, the only reason I just noticed this is that I upgraded the NIC on the FreeBSD machine and the switch into which it connects to 1000Base because the LM12 machine had a built in 1000Base NIC. I also changed the cables on both machines to ensure they were not the problem. Prior to this, I was bandwidth constrained by the 100Base so I never saw NFS performance as an issue. When I upgraded, I expected faster transfers and when I didn't get them, I started this whole investigation. So ... I'm stumped: - It's not the drive or SATA ports because both drives show comparable performance. - It's not the cables because I can get great throughput on one of the NFS mountpoints. - It's neither NIC for the same reason. Does anyone: A) Have a clue what might be doing this B) Have a suggestion how to track down the problem Thanks, -- --**--** Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ With respect to your mount points : /usr1 is spanning TWO different partitions : /dev/ad4s1f390G127G231G35%/usr1 /dev/ad6s1d902G710G120G86%/usr1/BKU because /usr1/BKU is a sub-directory of /usr1 . If you create a new directory , for example /usr2 , and /usr2/BKU , and using this new separate directory for sharing , such as : /dev/ad6s1d902G710G120G86%/usr2/BKU and machine:/usr2/BKU /BKU nfs rw,soft,intr 0 0 will it make difference ? Mehmet Erol Sanliturk ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS Performance: Weirder And Weirder
On 03/16/2013 04:20 PM, Mehmet Erol Sanliturk wrote: With respect to your mount points : /usr1 is spanning TWO different partitions : /dev/ad4s1f390G127G231G35%/usr1 /dev/ad6s1d902G710G120G86%/usr1/BKU because /usr1/BKU is a sub-directory of /usr1 . If you create a new directory , for example /usr2 , and /usr2/BKU , and using this new separate directory for sharing , such as : /dev/ad6s1d902G710G120G86%/usr2/BKU and machine:/usr2/BKU /BKU nfs rw,soft,intr 0 0 will it make difference ? Mehmet Erol Sanliturk I just tried this and it made no difference. The same file copied onto the NFS mount on /usr1/shared takes about 20x as long when coppied on to /usr[1|2]/BKU. -- Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS Performance: Weirder And Weirder
On Sat, Mar 16, 2013 at 3:07 PM, Tim Daneliuk tun...@tundraware.com wrote: On 03/16/2013 04:20 PM, Mehmet Erol Sanliturk wrote: With respect to your mount points : /usr1 is spanning TWO different partitions : /dev/ad4s1f390G127G231G35%/usr1 /dev/ad6s1d902G710G120G86%/usr1/BKU because /usr1/BKU is a sub-directory of /usr1 . If you create a new directory , for example /usr2 , and /usr2/BKU , and using this new separate directory for sharing , such as : /dev/ad6s1d902G710G120G86%/usr2/BKU and machine:/usr2/BKU /BKU nfs rw,soft,intr 0 0 will it make difference ? Mehmet Erol Sanliturk I just tried this and it made no difference. The same file copied onto the NFS mount on /usr1/shared takes about 20x as long when coppied on to /usr[1|2]/BKU. -- --**--** Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ Michael W. Lucas in Absolute FeeBSD , 2nd Edition , ( ISBN : 978-1-59327-151-0 ) , is suggesting the following ( p. 248 ) : In client ( mount , or , fstab ) , use options ( -o tcp , intr , soft , -w=32768 , -r=32768 ) tcp option will request a TCP mount instead of UDP mount , because FreeBSD NFS defaults to running over UDF . This subject may be another check point . Mehmet Erol Sanliturk ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS Performance: Weirder And Weirder
just slap an netapp 8.x with an avere flash box in front if you want NFS performance... or isilon. On Sat, Mar 16, 2013 at 5:43 PM, Mehmet Erol Sanliturk m.e.sanlit...@gmail.com wrote: On Sat, Mar 16, 2013 at 3:07 PM, Tim Daneliuk tun...@tundraware.com wrote: On 03/16/2013 04:20 PM, Mehmet Erol Sanliturk wrote: With respect to your mount points : /usr1 is spanning TWO different partitions : /dev/ad4s1f390G127G231G35%/usr1 /dev/ad6s1d902G710G120G86%/usr1/BKU because /usr1/BKU is a sub-directory of /usr1 . If you create a new directory , for example /usr2 , and /usr2/BKU , and using this new separate directory for sharing , such as : /dev/ad6s1d902G710G120G86%/usr2/BKU and machine:/usr2/BKU /BKU nfs rw,soft,intr 0 0 will it make difference ? Mehmet Erol Sanliturk I just tried this and it made no difference. The same file copied onto the NFS mount on /usr1/shared takes about 20x as long when coppied on to /usr[1|2]/BKU. -- --**--** Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ Michael W. Lucas in Absolute FeeBSD , 2nd Edition , ( ISBN : 978-1-59327-151-0 ) , is suggesting the following ( p. 248 ) : In client ( mount , or , fstab ) , use options ( -o tcp , intr , soft , -w=32768 , -r=32768 ) tcp option will request a TCP mount instead of UDP mount , because FreeBSD NFS defaults to running over UDF . This subject may be another check point . Mehmet Erol Sanliturk ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS Performance: Weirder And Weirder
On 03/16/2013 05:43 PM, Mehmet Erol Sanliturk wrote: Michael W. Lucas in Absolute FeeBSD , 2nd Edition , ( ISBN : 978-1-59327-151-0 ) , is suggesting the following ( p. 248 ) : In client ( mount , or , fstab ) , use options ( -o tcp , intr , soft , -w=32768 , -r=32768 ) tcp option will request a TCP mount instead of UDP mount , because FreeBSD NFS defaults to running over UDF . This subject may be another check point . Another very good suggestion but ... to no avail. Thanks for pointing this out. -- Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS Performance: Weirder And Weirder
On Sat, Mar 16, 2013 at 6:46 PM, Tim Daneliuk tun...@tundraware.com wrote: On 03/16/2013 05:43 PM, Mehmet Erol Sanliturk wrote: Michael W. Lucas in Absolute FeeBSD , 2nd Edition , ( ISBN : 978-1-59327-151-0 ) , is suggesting the following ( p. 248 ) : In client ( mount , or , fstab ) , use options ( -o tcp , intr , soft , -w=32768 , -r=32768 ) tcp option will request a TCP mount instead of UDP mount , because FreeBSD NFS defaults to running over UDF . This subject may be another check point . Another very good suggestion but ... to no avail. Thanks for pointing this out. -- --**--** Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ I have read messages once more . There is a phrase : Linux Mint 12 machineS ( plural ) . In your descriptions , there is no any information about network setup : Single client , multiple clients , etc . Then , with some assumptions : If there is ONLY ONE client , and all of the tests are performed on this ONLY client , problem may be attributed to FreeBSD server or kind of file(s) in different directories : One of the is encrypted ( requires decryption ) , another is plain file , etc. . If there is MORE than ONE client , problem may be attributed to any one the components of the network ( server , clients , switch , cable , NICs , interfering other software , etc. ) . Assume there is MULTIPLE clients : Take two clients of them : (A) Client 1 : Mount two directories . (B) Client 2 : Mount two directories . Test transmission performance : If they are similar , inspect server settings , directory privileges , etc . , file systems ( one is ZFS , other is UFS2 , etc. ) . All of the hardware may work properly , but if the file reading is not able to feed NIC sufficiently fast , it may show up as degraded performance . Increasing NIC buffer size ( as standard it is around 1000 bytes ) to maximum available , may offset latency of supply of data to NIC . If they are different : Check client specialties : A cable may be CAT5 ( only maximum 100 Mbits transfer . Network cards are adaptive , they try 1 Gbits , if it is not achievable , it reduces to speed to 100 Mbits , even to 10 Mbits ) . In that case either use CAT6 cable or CAT5x ( for 1 Gbit transmission , I do not remember x now ) The cable kind should be written on cable , if it is not written , select a properly labelled cable . Interchange cable tips to clients : If performance interchanges also : Cable or SWITCH port is faulty : Check switch port : It may be a 100 Mbits , be sure that it is also 1 Gbits and working properly . Mehmet Erol Sanliturk ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS Performance: Weirder And Weirder
On 03/16/2013 10:15 PM, Mehmet Erol Sanliturk wrote: On Sat, Mar 16, 2013 at 6:46 PM, Tim Daneliuk tun...@tundraware.com mailto:tun...@tundraware.com wrote: On 03/16/2013 05:43 PM, Mehmet Erol Sanliturk wrote: Michael W. Lucas in Absolute FeeBSD , 2nd Edition , ( ISBN : 978-1-59327-151-0 ) , is suggesting the following ( p. 248 ) : In client ( mount , or , fstab ) , use options ( -o tcp , intr , soft , -w=32768 , -r=32768 ) tcp option will request a TCP mount instead of UDP mount , because FreeBSD NFS defaults to running over UDF . This subject may be another check point . Another very good suggestion but ... to no avail. Thanks for pointing this out. -- --__--__ Tim Daneliuk tun...@tundraware.com mailto:tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ I have read messages once more . There is a phrase : Linux Mint 12 machineS ( plural ) . In your descriptions , there is no any information about network setup : Single client , multiple clients , etc . Then , with some assumptions : If there is ONLY ONE client , and all of the tests are performed on this ONLY client , problem may be attributed to FreeBSD server or kind of file(s) in different directories : One of the is encrypted ( requires decryption ) , another is plain file , etc. . There is one server - FreeBSD, and one client - LM12. Both have had their cables replaced with new CAT6 wiring. Copying the exact same file to each of the NFS mounts exhibits the problem. Reading from the two NFS mount is fast and as expected, so I do not suspect network issues. The two drives used on the server show similar disk performance locally. The server side exports are identical for both mounts as are the client side mounts. The ONLY difference is that the fast NFS mount has server side permissions of 777 whereas the slow NFS mount has server side permissions of 775. Both are owned by root:wheel. The contents of each filesystem are owned by a user in the wheel group. The one other difference is that all the contents of the slow mount are in a particular user group, and all the ones in the fast mount are in the wheel group. Changing the group ownership of all the stuff in the slow mount to wheel makes no difference. The problem appears to be size related on the slow mount. When I copy, say, a 100MB file to it, performance is just fine. When I copy a 1G file, it's 1/20 the throughput (45MB/sec vs 2MB/sec). This feels like some kind of buffer starvation but the fact that I can run at full speed against another mount point leaves me scratching my head as to just where. It's almost like there's some kind of halting going on during the transfer. Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS Performance: Weirder And Weirder
There is one more point to check : From your mount information , in the server , directories are on DIFFERENT drives . Assume one of the drives is very INTELLIGENT to save power . During local reading , due to reading speed , it may not go to SLEEP , but during network access , it may go to sleep due to its exceeded waiting time . If this is the case , please stay away from INTELLIGENT drives in a server : These are designed and produced by very IGNORANT entities . For simple , personal applications , their latency may not be noticed very much , but in a server , they can not be used . Another point may be file sizes . To check effect of file size , into the two different directories copy a large ( for example , 5 GB , or a 4.n GB .iso file ) and transmit these same files from their directories to a single client . If directory structure makes a difference , assuming hardware parts and client does not behave differently to these files , performance difference may be attributed to server side . Mehmet Erol Sanliturk ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Weird NFS Performance Problem
I have a FreeBSD 9.1-STABLE exhibiting weird NFS performance issues and I'd appreciate any suggestions. I have several different directories exported from the same filesystem. The machine that mounts them (a Linux Mint 12 desktop) writes nice and fast to one of them, but writes to the other one are dreadfully slow. Both are mounted on the LM machine using 'rw,soft,intr' in that machine's fstab file. Any ideas on what might be the culprit here? -- Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Weird NFS Performance Problem
On Fri, Mar 15, 2013 at 5:09 PM, Tim Daneliuk tun...@tundraware.com wrote: I have a FreeBSD 9.1-STABLE exhibiting weird NFS performance issues and I'd appreciate any suggestions. I have several different directories exported from the same filesystem. The machine that mounts them (a Linux Mint 12 desktop) writes nice and fast to one of them, but writes to the other one are dreadfully slow. Both are mounted on the LM machine using 'rw,soft,intr' in that machine's fstab file. Any ideas on what might be the culprit here? -- --**--** Tim Daneliuk tun...@tundraware.com PGP Key: http://www.tundraware.com/PGP/ Is the slow directory a LINK ? If it is a LINK , then , try with the real directory name . Thank you very much . Mehmet Erol Sanliturk ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Asymmetric NFS Performance
In the last episode (Feb 02), Tim Daneliuk said: Server:FBSD 8.2-STABLE / MTU set to 15000 Client:Linux Mint 12 / MTU set to 8192 NFS Mount Options: rw,soft,intr Problem: Throughput copying from Server to Client is about 2x that when copying a file from client to server. The client does have a SSD whereas the server has conventional SATA drives but ... This problem is evident with either 100- or 1000- speed ethernet so I don't think it is a drive thing since you'd expect to saturate 100-BASE with either type of drive. Things I've Tried So Far: - Increasing the MTUs - This helped speed things up, but the up/down ratio stayed about the same. - Fiddling with rsize and wsize on the client - No real difference If iostat -zx 1 on the server shows the disks at 100% busy, you're probably getting hit by the fact that NFS has to commit writes to stable storage before acking the client, so writes over NFS can be many times slower than local write speed. Setting the vfs.nfsrv.async sysctl to 1 will speed things up, but if the server reboots while a client is writing, you will probably end up with missing data even though the client thought everything was written. If you are serving ZFS filesystems, stick an SSD in the server and point the ZFS intent log at it: zpool add mypool log da3. 8GB of ZIL is more than enough, but it needs to be fast, so no sticking a $10 thumb drive in and expecting any improvement :) -- Dan Nelson dnel...@allantgroup.com ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Asymmetric NFS Performance
Server:FBSD 8.2-STABLE / MTU set to 15000 Client:Linux Mint 12 / MTU set to 8192 NFS Mount Options: rw,soft,intr Problem: Throughput copying from Server to Client is about 2x that when copying a file from client to server. The client does have a SSD whereas the server has conventional SATA drives but ... This problem is evident with either 100- or 1000- speed ethernet so I don't think it is a drive thing since you'd expect to saturate 100-BASE with either type of drive. Things I've Tried So Far: - Increasing the MTUs - This helped speed things up, but the up/down ratio stayed about the same. - Fiddling with rsize and wsize on the client - No real difference Ideas anyone? --- Tim Daneliuk ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
NFS performance-tuning FreeBSD - NetApp
Hi, I've got a FreeBSD 7.2 box (HP C-class Blade - AMD dual core Opteron (x64), 4GB RAM, Broadcom NetXtreme II BCM5706) that should be connected to a NetApp 3170 filer via NFS. Out of the box, with nothing tuned (no special parameters for mount_nfs, no kernel tuning), performance is very sluggish: I've got ~250Mbit/sec performance with peaks around 400Mbit/sec. Sure enough, neither CPU (server and NetApp) nor network performance is the problem here - it must be something NFS-related. Any ideas on how to increas my NFS-performance? (Special mount parameters, kernel tuning,...) Thanks in advance for any clue, -ewald ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS performance-tuning FreeBSD - NetApp
Merhaba Ewald, You can read http://communities.netapp.com/thread/39 thread. There is special mount options for Linux and also for FreeBSD give it a try. Regards. Monday, August 3, 2009, 11:55:16 AM, you wrote: Hi, I've got a FreeBSD 7.2 box (HP C-class Blade - AMD dual core Opteron (x64), 4GB RAM, Broadcom NetXtreme II BCM5706) that should be connected to a NetApp 3170 filer via NFS. Out of the box, with nothing tuned (no special parameters for mount_nfs, no kernel tuning), performance is very sluggish: I've got ~250Mbit/sec performance with peaks around 400Mbit/sec. Sure enough, neither CPU (server and NetApp) nor network performance is the problem here - it must be something NFS-related. Any ideas on how to increas my NFS-performance? (Special mount parameters, kernel tuning,...) Thanks in advance for any clue, -ewald ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org -- Best regards, Omermailto:of...@enderunix.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: NFS performance-tuning FreeBSD - NetApp
On Aug 3, 2009, at 4:55 AM, Ewald Jenisch wrote: Hi, I've got a FreeBSD 7.2 box (HP C-class Blade - AMD dual core Opteron (x64), 4GB RAM, Broadcom NetXtreme II BCM5706) that should be connected to a NetApp 3170 filer via NFS. Out of the box, with nothing tuned (no special parameters for mount_nfs, no kernel tuning), performance is very sluggish: I've got ~250Mbit/sec performance with peaks around 400Mbit/sec. Sure enough, neither CPU (server and NetApp) nor network performance is the problem here - it must be something NFS-related. Any ideas on how to increas my NFS-performance? (Special mount parameters, kernel tuning,...) I would suggest bumping the read and write sizes to 32K and using tcp instead of udp If you have very large directories, you can also see an increase in responsiveness by enabling readdirplus as well, but that wont help with raw throughput. Try passing the following parameters to mount and see if performance is any better -r=32768 -w=32768 -l -T -- Steven Kreuzer http://www.exit2shell.com/~skreuzer ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: very poor NFS performance from a beta3
Jonathan Horne wrote: i updated my workstatino to beta3, and then got on a 6.2-p8 machine and mounted /usr/src and /usr/obj from the beta3. tried to installkernel, but it moved as painful pace. would get to the point where it moves kernel to kernel.old, and would just pause for a long time. file transfer showed about 104k. i took this same 6.2-p8 box, and mounted src and obj from my main 6.2 build server, and reinstalled the 6.2-p8 kernel, and speed was as expected. is there anywhere i can being looking to troubleshoot this problem (as to why the 7.0b3 would serve NFS so slowly)? This is usually because your NIC has mis-negotiated and is unable to pass packets properly. Kris ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: very poor NFS performance from a beta3
On Sunday 18 November 2007 05:59:12 am Kris Kennaway wrote: Jonathan Horne wrote: i updated my workstatino to beta3, and then got on a 6.2-p8 machine and mounted /usr/src and /usr/obj from the beta3. tried to installkernel, but it moved as painful pace. would get to the point where it moves kernel to kernel.old, and would just pause for a long time. file transfer showed about 104k. i took this same 6.2-p8 box, and mounted src and obj from my main 6.2 build server, and reinstalled the 6.2-p8 kernel, and speed was as expected. is there anywhere i can being looking to troubleshoot this problem (as to why the 7.0b3 would serve NFS so slowly)? This is usually because your NIC has mis-negotiated and is unable to pass packets properly. Kris and another reason why this is so peculiar, is that when my desktop was at 7.0b2, i installed b2 to another laptop (not the same system as the 6.2p8), and it was perfectly normal. cheers, -- Jonathan Horne http://dfwlpiki.dfwlp.org [EMAIL PROTECTED] ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: very poor NFS performance from a beta3
On Sunday 18 November 2007 05:59:12 am Kris Kennaway wrote: Jonathan Horne wrote: i updated my workstatino to beta3, and then got on a 6.2-p8 machine and mounted /usr/src and /usr/obj from the beta3. tried to installkernel, but it moved as painful pace. would get to the point where it moves kernel to kernel.old, and would just pause for a long time. file transfer showed about 104k. i took this same 6.2-p8 box, and mounted src and obj from my main 6.2 build server, and reinstalled the 6.2-p8 kernel, and speed was as expected. is there anywhere i can being looking to troubleshoot this problem (as to why the 7.0b3 would serve NFS so slowly)? This is usually because your NIC has mis-negotiated and is unable to pass packets properly. Kris kris, thanks for your reply. whats the best way to tell if i have mis-negotiated? is just 'ifconfig' sufficient? i was wondering about this earlier, and reboots of the hosts as well as the switches yielded no different results. also, i did a few other tests as well. i can scp to/from the 7.0b3 and 6.2p8 from/to each other as well as other hosts on the network at what i would call normal speeds (6-10 megabytes/sec). also, i can 'cp -vpnRP' directory trees from/to the same with normal results. only installing kernel from the 7.0 to the 6.2 stalls out. -- Jonathan Horne http://dfwlpiki.dfwlp.org [EMAIL PROTECTED] ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
very poor NFS performance from a beta3
i updated my workstatino to beta3, and then got on a 6.2-p8 machine and mounted /usr/src and /usr/obj from the beta3. tried to installkernel, but it moved as painful pace. would get to the point where it moves kernel to kernel.old, and would just pause for a long time. file transfer showed about 104k. i took this same 6.2-p8 box, and mounted src and obj from my main 6.2 build server, and reinstalled the 6.2-p8 kernel, and speed was as expected. is there anywhere i can being looking to troubleshoot this problem (as to why the 7.0b3 would serve NFS so slowly)? -- Jonathan Horne http://dfwlpiki.dfwlp.org [EMAIL PROTECTED] ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Question on NFS performance
Hi all we have a FreeBSD 5.4 exporting some NFS filesystem to a cluster of gentoo boxes (kernel 2.6.12). Our exported storage disk is an Apple XRaid. We have Gigabit Ethernet both on the client and the server. We would like to improve our read performance. This is our performance: about 10Mb reading a file of 1Gb with dd and iozone confirms this result. We already use the normal optimization flags (we use rpc.lockd and rpc.statd, on the client we have read size 65536 and a read ahead of 4 blocks, the async options). Is our performance the best we can get? Can we improve it? Thanks for your help Valerio Daelli ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Question on NFS performance
On Wed, May 10, 2006 at 02:54:39PM +0200, Valerio daelli wrote: Hi all we have a FreeBSD 5.4 exporting some NFS filesystem to a cluster of gentoo boxes (kernel 2.6.12). Our exported storage disk is an Apple XRaid. We have Gigabit Ethernet both on the client and the server. We would like to improve our read performance. This is our performance: about 10Mb reading a file of 1Gb with dd and iozone confirms this result. We already use the normal optimization flags (we use rpc.lockd and rpc.statd, on the client we have read size 65536 and a read ahead of 4 blocks, the async options). Is our performance the best we can get? Can we improve it? 5.4 and filesystem performance cannot be said together in the same sentence. Upgrade to 6.1. Kris pgpndSMng4P0q.pgp Description: PGP signature
poor NFS performance in 5.x
Hi :) I upgraded two boxes to FreeBSD-5.2-BETA a week ago and I noticed that NFS performance is very slow compared to 4.x-RELEASE. Before, NFS transfers were between 10 and 12 MB/s and now I don't go past 7 MB/s. My exports/mount settings did not change and the hardware is obviously the same. Any idea where I should start looking for resolving this ? It is important since I'm planning on upgrading some servers in the futur and homedir are mounted via NFS. Thanks in advance. Antoine ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: poor NFS performance in 5.x
On Thu, Nov 27, 2003 at 10:12:21AM +0100, Antoine Jacoutot wrote: Hi :) I upgraded two boxes to FreeBSD-5.2-BETA a week ago and I noticed that NFS performance is very slow compared to 4.x-RELEASE. Before, NFS transfers were between 10 and 12 MB/s and now I don't go past 7 MB/s. My exports/mount settings did not change and the hardware is obviously the same. Any idea where I should start looking for resolving this ? Try reading the basic documentation that comes with 5.2-BETA, for example the /usr/src/UPDATING file, which tells you clearly that performance is not expected to be good unless you disable the standard debugging options. Also make sure you read the 5.x Early Adopter's Guide, available on the website. Kris pgp0.pgp Description: PGP signature
Re: poor NFS performance in 5.x
On Thursday 27 November 2003 10:28, Kris Kennaway wrote: Try reading the basic documentation that comes with 5.2-BETA, for example the /usr/src/UPDATING file, which tells you clearly that performance is not expected to be good unless you disable the standard I've been running CURRENT on test boxes for months, I know that and this is not the problem, I have my own compiled kernel without debugging options set. ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Improving FreeBSD NFS performance (esp. directory updates)
On Thu, May 29, 2003 at 04:05:04PM -0500, Marc Wiz wrote: On Thu, May 29, 2003 at 04:54:00PM -0400, Tom Limoncelli wrote: I have a NFS server with (so far) a single NFS client. Things work fine, however if (on the client) I do an rm -rf foo on a large (deep and wide) directory tree the tty receives NFS server not responding/NFS server ok messages. I don't think the network is at fault, nor is the server really going away. I think the client is just impatient. Is there a way to speed up a large rm -rf? I have soft-writes enabled but alas Tom, please reproduce the problem but before doing it run the following commands and save the output: On the client: nfsstat -c netstat -m netstat -s On the server: nfsstat -s netstat -m netstat -s Run the rm -rf /foo Rerun the above commands on both the client and server and of course save the output again :-) RTFM-ing for nfsstat I am disappointed that nfsstat does not have -z option for zeroing out the counters. Time to look at the source :-) It turns out the issue isn't only the rm -rf, so I've eliminated that. The process is creating many subdirectories and files. Here's the output that you requested. I've generated the output at a baseline, 1 minute into the process, and then every 4 hours. Sorry for the flood of information but I figure more is better. You can look at the first and last items if you prefer. I really appreciate the help! --tal -- Tom Limoncelli -- [EMAIL PROTECTED] -- www.lumeta.Com # CLIENT BASELINE: Thu Jun 5 01:41:32 UTC 2003 Client Info: Rpc Counts: Getattr SetattrLookup Readlink Read WriteCreateRemove 16222151 2470 9976708 91232 23111364 23342730544746 51291 Rename Link Symlink Mkdir Rmdir Readdir RdirPlusAccess 385 0 11004760080 13602 16651 0 69815088 MknodFsstatFsinfo PathConfCommitGLeaseVacate Evict 030962158 0 21980734 0 0 0 Rpc Info: TimedOut Invalid X Replies Retries Requests 0 0 0 239 166249915 Cache Info: Attr HitsMisses Lkup HitsMisses BioR HitsMisses BioW HitsMisses 359087567 84614481 137095640 9976703 201904089 23084158 15151506 23342730 BioRLHitsMisses BioD HitsMisses DirE HitsMisses 5296031 91232817844 16635 53658 0 257/608/18240 mbufs in use (current/peak/max): 257 mbufs allocated to data 256/312/4560 mbuf clusters in use (current/peak/max) 776 Kbytes allocated to network (5% of mb_map in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines tcp: 205722821 packets sent 155645171 data packets (2158194817 bytes) 109 data packets (89632 bytes) retransmitted 0 resends initiated by MTU discovery 34720623 ack-only packets (365546 delayed) 0 URG only packets 0 window probe packets 15355469 window update packets 1449 control packets 213168802 packets received 152983420 acks (for 2158194689 bytes) 1424 duplicate acks 0 acks for unsent data 208828275 packets (1042433032 bytes) received in-sequence 3 completely duplicate packets (192 bytes) 0 old duplicate packets 0 packets with some dup. data (0 bytes duped) 62 out-of-order packets (5136 bytes) 0 packets (0 bytes) of data after window 0 window probes 62774 window update packets 0 packets received after close 0 discarded for bad checksums 0 discarded for bad header offset fields 0 discarded because packet too short 671 connection requests 119 connection accepts 1 bad connection attempt 0 listen queue overflows 790 connections established (including accepts) 777 connections closed (including 2 drops) 202 connections updated cached RTT on close 202 connections updated cached RTT variance on close 3 connections updated cached ssthresh on close 0 embryonic connections dropped 152983420 segments updated rtt (of 149305673 attempts) 74 retransmit timeouts 1 connection dropped by rexmit timeout 0 persist timeouts 0 connections dropped by persist timeout 31 keepalive timeouts 31 keepalive probes sent 0 connections dropped by keepalive 3834814 correct ACK header predictions 60088755 correct data packet header predictions 119 syncache entries added
Improving FreeBSD NFS performance (esp. directory updates)
I have a NFS server with (so far) a single NFS client. Things work fine, however if (on the client) I do an rm -rf foo on a large (deep and wide) directory tree the tty receives NFS server not responding/NFS server ok messages. I don't think the network is at fault, nor is the server really going away. I think the client is just impatient. Is there a way to speed up a large rm -rf? I have soft-writes enabled but alas Details: The server is an Intel Xeon CPU 2.20GHz running FreeBSD 4.8 (updated to the latest RELENG_4_8 release yesterday). It has a 1000TX interface to a Cisco switch. The partition is part of a SCSI RAID unit (160MB/s transfer rate, Tagged Queueing enabled). The partition that is exported to the client is: # mount | egrep e2 /dev/da0s1f on /e2 (ufs, NFS exported, local, soft-updates) /etc/rc.conf lists: nfs_server_flags=-u -t -n 16 The client is a (accoriding to dmesg) Pentium III/Pentium III Xeon/Celeron (934.99-MHz 686-class CPU) running FreeBSD 4.7-RELEASE. amd is configured to mount the partion with opts:=rw:nfs_proto=udp;nfs_vers=3. /etc/rc.conf lists: nfs_client_flags=-n 4 Any tuning suggestions? --tal ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Improving FreeBSD NFS performance (esp. directory updates)
On Thu, May 29, 2003 at 04:54:00PM -0400, Tom Limoncelli wrote: I have a NFS server with (so far) a single NFS client. Things work fine, however if (on the client) I do an rm -rf foo on a large (deep and wide) directory tree the tty receives NFS server not responding/NFS server ok messages. I don't think the network is at fault, nor is the server really going away. I think the client is just impatient. Is there a way to speed up a large rm -rf? I have soft-writes enabled but alas Tom, please reproduce the problem but before doing it run the following commands and save the output: On the client: nfsstat -c netstat -m netstat -s On the server: nfsstat -s netstat -m netstat -s Run the rm -rf /foo Rerun the above commands on both the client and server and of course save the output again :-) RTFM-ing for nfsstat I am disappointed that nfsstat does not have -z option for zeroing out the counters. Time to look at the source :-) Marc (who in a former life and now current life is doing NFS support) -- Marc Wiz [EMAIL PROTECTED] Yes, that really is my last name. ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: nfs performance
recently I discovered problems with my FreeBSD nfs server. I mount my /home/user from my linux box via automounter/nfs from my server. They are connected with a switch on a 100baseTX Ethernet. Now, whenever I copy large files from a local driver to my home dir or do anything else that involves moving some bigger amounts of data to my home dir the nfs server times out and doesnt respond anymore. Any ideas, suggestions would be appreciated, I could never get NFS to work reliably on Linux. The server here is slow and needs to be upgraded to newer iron and runs OpenBSD, but Linux is the only client OS that exhibited these problems. I observed them always on writing to the server with large files, reads seemed to work fine, no matter what the file size. I spent considerable time fiddling with timeouts, cachesizes and so on, this was several months ago, but I remember thinking this had something to do with the attribute caching. I was advised by a Linux guru friend of mine that the kernel NFS on Linux had multiple problems, and he advised me to use the userland NFS. I didn't follow this advice, choosing to try FreeBSD instead. So far, it has worked with minimal problems. Once every couple of weeks, I start getting NFS Server not responding messages, but switching to TCP transport seems to have cured that, and provided better performance as well. I'm sure this isn't what you wanted to hear, but you might find it helpful... sdb -- [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
Re: nfs performance
On Monday, December 30, 2002, at 01:12 PM, Scott Ballantyne wrote: I could never get NFS to work reliably on Linux I'd like to chime in on this. There are some serious problems with Linux NFS support. At the company where I work, we use Solaris NFS servers on Sun hardware, with a mix of many UNIX clients (including some BSD and Mac OS X). The biggest problems come from Linux clients. The interesting thing to note with Linux is that it hoses itself after a while. If the user reboots their Linux client, the problems go away for a few weeks. We've messed with options in automount and AMD, but no help. -john To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
Re: NFS Performance woes
On Tue, Nov 05, 2002 at 06:44:47PM +0100, Lasse Laursen wrote: How is the optimum number of nfsd processes determined on the server? On our current setup we have 4 nfs daemons running serving 3 clients (webservers) Is the number of daemons to start determined by the number of clients or the number of files that has to be transferred simultaniously? I use something like ps waux | grep nfs and check the CPU time used by the processes. Add processes until you have 1 or 2 that are typically unused in normal loads. -T -- Losing an illusion makes you wiser than finding a truth. Ludwig Borne To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
Re: NFS Performance woes
On Wed, 2002-11-06 at 19:52, BigBrother wrote: Although the man page says this, I *think* that the communication is done like this CLIENT = NFSIOD(CLIENT) = NFSIOD (SERVER) = NFSD which menas that NFSIOD 'speak' with each other and then they pass the requests to NFS. Of course if u dont need to have on the server too many NFSIOD. So in my case I just have 8 nfsiod on server running and most of them are idle, and besides, they only take 1.5MB of memory which I can afford. So I think having *some* NFSIOD also on server is not a bad idea. Of course on the server u should have a lot of NFSD. in other words, running NFSIOD on server is not a bad idea.. NFSIOD is enabled by putting nfs_client_enable=YES into rc.conf. Empirical evidence suggests that NFSIOD is not used server-side. I ran 4 nfsiod daemons on the server and checked their usage time. All were 0:00. If the server does any client NFS stuff it would make a difference, and it may be different under other OS. Certainly it does no harm to have them running. Also monitor the mbufs on all your machines (especially the server). do from time to time a 'netstat -m' and see the peak value of mbuf and mbuf clusters...if it is close to the max limit then you will suffer from mbuff exhaustion will will eventually make the machine unreachable from network. u can change mbuff max value before kernel load ...see tuning (7) kern.nmbclusters=value in /boot/loader.conf, if anyone needs to do this. Have to reboot for this one :-( Also if u have mbuf exhastion try to use smaller block size in NFS mounts. Now this is interesting. I had thought mbuf cluster exhaustion was due to a high number of connections. Although I guess a high number of connections * large buffer size would do it too. Thank you for your response and suggestions vis the other NFS stuff - I managed to get our server talking UDP. The wildcard binding was the problem, and the -h flag to nfsd fixed it. Network usage graphs are showing the differential between incoming and outgoing traffic to be much less now, so I would say there was a lot of overhead in there, as well as retransmissions. I am still playing with buffer sizes but chances are in this case the FreeBSD default is best. Regards -- Duncan Anker Senior Systems Administrator Dark Blue Sea To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
Re: NFS Performance woes
I recently did some research into NFS performance tuning and came across the suggestion in an article on onlamp.com by Michael Lucas, that 32768 is a good value for the read and write buffers. His suggestion is these flags: tcp,intr,nfsv3,-r=32768,-w=32768 I used these options (I found tcp was mandatory, as we have multiple IPs and UDP was refusing to play nice), also adding dumbtimer to avoid the log messages about server not responding. According to my experience UDP is much preffered for NFS transport protocols. Also try to have the NFSIOD daemon being executed on every machine by putting in the /etc/rc.conf nfs_client_enable=YES nfs_client_flags=-n 10 [u may put more than 10 instances if u suspect that more than 10 simultaneous transactions will happen] Also use the -w=32768,-r=32768 switch only on the machines that have a fast cpu and a good network card [e.g. nestat -w 1, doesnt show errors under heavy load] On all the other machines dont put any w,r values [which will default to 8k blocks] In some machines of mine I have even used blocks of -r=4096,-w=4096 because they were old machines that could not keep up with the traffic and they were complaining about mbufs [they run out of mbufs and after some time they crashed]..(and because they machines were diskless it was unable to change the value of mbufss, after the kernel loading the value is readonly and cannot be changed). Use good networking hardware...scrappy hardware will certainly put you into great trouble. If you use TCP for NFS on a 1GB network you will sure have problems on your machines and they will not be able to keep up. TCP causes a great overhead. UDP doesnt. So bottom line: a) Use UDP b) Run a lot of NFSIOD - the more the better c) Examine what is the best block size for every host idividually! (dont assume that 32k block is good for every host) Hopes it does your job..I was searching for over 3 months when I once dealt with thisRead also from the 'Sun' site the 'Optimizing and Tunning NFS' guide which is a nice PDF document that you can download for free, and has a lot of interesting things similar with FreeBSD! To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
Re: NFS Performance woes
Howdy! I have done some simulations with NFS servers - Intel SCB2 (4G RAM) serving files from 500G RAID devices. I created a treed directory structure with 300G of 32k files that approximates our homedirectory structure. I had about 6 diskless front ends (tyan 2518 with 2G) that NFS booted and mounted the homedir and ran multiple scripts that walked through the directory structure reading files and writing them to /dev/null. All machines have 3 intel 100 NICs. One interface is used to mount the root etc. and the other is used to mount homedirs. The NFS server serves root from fxp0 and homedir data from both fxp1 fxp2. The diskless front ends mount root from fxp0 and mount homedir from fxp1. When the simulation was running full out I was serving about 1/3-1/2 data from page cache and 2/3-1/2 from disk. I tried numerous configurations and tuning of network parameters after doing research and discovered... for FreeBSD 4.6.2 1) NFS over UDP outperforms NFS over TCP. I was able to average about 70Mbs over *both* (occasionally they would almost max out ie. 95Mbs) interfaces serving data using UDP mounts with 8K rw. (the default). No matter what I tried with TCP I never got more than half that throughput. 2) The optimal number of nfsd's to run on the server was about 100! If I reduced the number of nfsds below 80 it would start to choke off the data moving through the network. I found that at around 100 there was no more increase. You must make a minor change to source as the max allowed now by default is 20. I was running 8 nfsiod's on the clients. TCP mounts under tested conditions always had much higher loads than UDP. Also it was impossible to do an ls on a mounted directory under load with TCP. With UDP there were no such problems. If you are using UDP it is *essential* that you monitor fragments that are being droppend because of timeout. If you have a good network this should not be a problem. For example I have a webserver that has been up 20 days and has moved 1G of fragments but has only dropped about 800. Also TCP mounts will require a remount of the clients if the server should crash/whatever. UDP just keeps on ticking. If you have Gig ether than there is other tuning you *must* do to realize this potential. Personally I think it is better to use multiple 100Mbs NIC's than to use Gig ether if you can get away with it. YMMV. -steve To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
Re: NFS Performance woes
Hi, According to my experience UDP is much preffered for NFS transport protocols. Also try to have the NFSIOD daemon being executed on every machine by putting in the /etc/rc.conf nfs_client_enable=YES nfs_client_flags=-n 10 [u may put more than 10 instances if u suspect that more than 10 simultaneous transactions will happen] How is the optimum number of nfsd processes determined on the server? On our current setup we have 4 nfs daemons running serving 3 clients (webservers) Is the number of daemons to start determined by the number of clients or the number of files that has to be transferred simultaniously? Same question goes for the number of nfsiod processes... Regards -- Lasse Laursen [EMAIL PROTECTED] - Systems Developer NetGroup A/S, St. Kongensgade 40H, DK-1264 København K, Denmark Phone: +45 3370 1526 - Fax: +45 3313 0066 - Web: www.netgroup.dk - We don't surf the net, we make the waves. To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
Re: NFS Performance woes
According to my experience UDP is much preffered for NFS transport protocols. Also try to have the NFSIOD daemon being executed on every machine by putting in the /etc/rc.conf nfs_client_enable=YES nfs_client_flags=-n 10 [u may put more than 10 instances if u suspect that more than 10 simultaneous transactions will happen] How is the optimum number of nfsd processes determined on the server? On our current setup we have 4 nfs daemons running serving 3 clients (webservers) Is the number of daemons to start determined by the number of clients or the number of files that has to be transferred simultaniously? Same question goes for the number of nfsiod processes... Well the only rule for selecting the number of nfsiods and nfsd is the maximum number of threads that are going to request an NFS operation on the server. For example assume that your web server has a typical number of httpd dameons of 50, that means that every httpd can access files on the server, and in the worst case both 50 httpd will request simultaneoulsy different NFS operations. This means that you should have at least 50 NFSIOD (on the client+server) and 50 NFSD running (on the server). Remember that NFSIOD must run both on CLIENT and SERVER. So you determine what is the maximum number of NFS operations...for example in your client you dont have only 50 httpd running, but you make from time to time compile with the -j 4 (4 parallel compilation jobs), this means that you should increase the number of 50 by +4... also in your client you usually have some users that login and their home directories are on NFS mounted media...usually 10 people are using NFS mounted home, which means that in the worst case 10 people may request something from their home so you have to increase the number fo 54 by 10 more I know the handbook says taht 8 nfsiod/nfsd is a nice number but I think that is not correct. I have an ftp server that uses NFS mounted directories, and usually 15 people are connected...so I have put a 20 NFS processes running... Having too much NFSIOD is not bad...every NFSIOD eats just 220KB of memory (which means that you should also consider your memory-if you can afford to run a lot of nfsiod) Having too much NFSD also is not bad...every NFS eats just 356Kbyte of memory, which again you have to note it. So with simple words, just add all the things that you can imagine that can happen simultaneously on all the NFS mounted dirs and put that number...let it run for one week and note down how many NFSIOD are idle or NFSD.If you have put 100 NFSIOD and you see that usually there are more than 50 NFSIOD idle (doing nothing) [on your ps axwu or TOP output] then its a safe bet to reduce the number... Of course you cannot optimize the NFS system in one day...it needs a lot of time to take measurements and check from time to time if you have enough NFSIOD or NFSD, because system load distribution tend may change and you may see that more or less NFS processes have to exist.. I hope I make it clear for you!! To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
Re: NFS Performance woes
Hi, Thanks for your reply. I have some additional questions: Well the only rule for selecting the number of nfsiods and nfsd is the maximum number of threads that are going to request an NFS operation on the server. For example assume that your web server has a typical number of httpd dameons of 50, that means that every httpd can access files on the server, and in the worst case both 50 httpd will request simultaneoulsy different NFS operations. This means that you should have at least 50 NFSIOD (on the client+server) and 50 NFSD running (on the server). A read operation (typical operation for all the clients) does not alter any data? So does every read request require a nfsd? Lets assume a worst case scenario where 50 http servers access 50 different files - would I need 50 NFS daemons to run on the server to obtain maximum performance then? Remember that NFSIOD must run both on CLIENT and SERVER. (Taken from the man page pf nfsiod) Nfsiod runs on an NFS client machine to service asynchronous I/O requests to its server. It improves performance but is not required for correct operation. Why should I start the nfsiod daemon on the server? Of course you cannot optimize the NFS system in one day...it needs a lot of time to take measurements and check from time to time if you have enough NFSIOD or NFSD, because system load distribution tend may change and you may see that more or less NFS processes have to exist.. Yep - At the moment one nfsd idles - I will monitor the number of processes and try to change the setup and see how the cluster performs. Regards -- Lasse Laursen [EMAIL PROTECTED] - Systems Developer NetGroup A/S, St. Kongensgade 40H, DK-1264 København K, Denmark Phone: +45 3370 1526 - Fax: +45 3313 0066 - Web: www.netgroup.dk - We don't surf the net, we make the waves. To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message
NFS Performance woes
I recently did some research into NFS performance tuning and came across the suggestion in an article on onlamp.com by Michael Lucas, that 32768 is a good value for the read and write buffers. His suggestion is these flags: tcp,intr,nfsv3,-r=32768,-w=32768 I used these options (I found tcp was mandatory, as we have multiple IPs and UDP was refusing to play nice), also adding dumbtimer to avoid the log messages about server not responding. Shortly thereafter, our project manager complained about a lot of timeouts on his web servers. More research suggested that bigger is not always better for NFS buffers (some articles indicate that r/w buffers are only for UDP, which is clearly rubbish in this case). I have tried fiddling with varying parameters to find optimal values, but I am having difficulty working out what they are. Our web servers all mount via NFS - my expectation therefore is that I should see approximately equal amounts of traffic in and out (working on the principle that the web request and the file request cancel out and the file read from the NFS server is going out unmodified). My observations have borne out this expectation. However, tuning various values from 1k up to 8k has resulted in slight increases in traffic - I don't know whether this is more throughput or transmission retries. Once I get up to 16k and 32k the amount of traffic into the boxes exceeds the amount out quite substantially, which seems to indicate problems (and indeed is causing timeouts). On the NFS server side of things, we saw a lot of watchdog timeouts, with the network choking at around 20Mbs. The card in the server is 1Gb so it really should be able to push through more than that without a sweat (and it does - it hits about 60Mbs during backups). There's something pretty screwy going on with all of this. As mentioned, the server is 1Gb, all the clients are running at 100Mbs. All machines are on the same switch, so there is nothing to degrade the performance. The NFS section of the FreeBSD handbook talked about slow FreeBSD clients with highspeed servers, and I am wondering if this is that type of scenario. I'd appreciate hearing the experience of others in terms of tuning a setup like this, or even if someone could point me in the direction of some more useful analysis tools. Thanks -- Duncan Anker Senior Systems Administrato Dark Blue Sea To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-questions in the body of the message