Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On Thu, 20 Jul 2017 00:49:29 +0300 Mikhailwrote: > > I heard about original OmniOS, by OmniTI, is being discountinued and > that OmniOSCE takes care of it. > But I could not find installation ISO images of the OmniOSCE. Which > procedure to follow to get OmniOSCE installed? I guess it is to get > latest available OmniOS installation ISO from > https://omnios.omniti.com/wiki.php/Installation and then follow > procedure described on the http://www.omniosce.org/ page to convert it > into OmniOSCE? > This is correct. Pay attention to upgrade to latest kernel release r151022i if you have HBA's based on LSI SAS >= 2300 (using mr_sas driver) -- Hilsen/Regards Michael Rasmussen Get my public GnuPG keys: michael rasmussen cc http://pgp.mit.edu:11371/pks/lookup?op=get=0xD3C9A00E mir datanom net http://pgp.mit.edu:11371/pks/lookup?op=get=0xE501F51C mir miras org http://pgp.mit.edu:11371/pks/lookup?op=get=0xE3E80917 -- /usr/games/fortune -es says: "I've finally learned what `upward compatible' means. It means we get to keep all our old mistakes." -- Dennie van Tassel pgpv_8edO5xYq.pgp Description: OpenPGP digital signature ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On 07/20/2017 12:33 AM, Michael Rasmussen wrote: >> The time to give OmniOS a try has come, I also noticed that X550 NICs >> are supported by OmniOS since late autumn 2016. Luckily Proxmox supports >> online storage migration (Move disk) without bringing vms down - this >> simplifies storage migration a lot in live environment! >> > Remember to get it here: http://www.omniosce.org/ I heard about original OmniOS, by OmniTI, is being discountinued and that OmniOSCE takes care of it. But I could not find installation ISO images of the OmniOSCE. Which procedure to follow to get OmniOSCE installed? I guess it is to get latest available OmniOS installation ISO from https://omnios.omniti.com/wiki.php/Installation and then follow procedure described on the http://www.omniosce.org/ page to convert it into OmniOSCE? thanks, Mikhail. ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On Thu, 20 Jul 2017 00:07:44 +0300 Mikhailwrote: > > The time to give OmniOS a try has come, I also noticed that X550 NICs > are supported by OmniOS since late autumn 2016. Luckily Proxmox supports > online storage migration (Move disk) without bringing vms down - this > simplifies storage migration a lot in live environment! > Remember to get it here: http://www.omniosce.org/ -- Hilsen/Regards Michael Rasmussen Get my public GnuPG keys: michael rasmussen cc http://pgp.mit.edu:11371/pks/lookup?op=get=0xD3C9A00E mir datanom net http://pgp.mit.edu:11371/pks/lookup?op=get=0xE501F51C mir miras org http://pgp.mit.edu:11371/pks/lookup?op=get=0xE3E80917 -- /usr/games/fortune -es says: Don't stop at one bug. - The Elements of Programming Style (Kernighan & Plaugher) pgp2heoVsl0rP.pgp Description: OpenPGP digital signature ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On 07/19/2017 06:38 PM, Michael Rasmussen wrote: >> I guess my only way to fix this is to migrate everything off that server >> and reinstall it from scratch, throwing away things like MDADM and LVM >> this time and replacing them with ZFS for storage purposes. >> > You mentioned before that you hoped to use Omnios. Latest stable > now supports your nics. Yes, that was more than a year ago when I deployed this storage server - you have good memory, Michael! =) The time to give OmniOS a try has come, I also noticed that X550 NICs are supported by OmniOS since late autumn 2016. Luckily Proxmox supports online storage migration (Move disk) without bringing vms down - this simplifies storage migration a lot in live environment! cheers, Mikhail. ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On Wed, 19 Jul 2017 18:30:01 +0300 Mikhailwrote: > > Basically, it looks like MDADM array, and LVM on top (and possibly FS > inside the VMs) need to be created with manual calculations for > alignment and these calculations need to be specified on the command > line at the time of creation. It is pity to find out this now, when > server is in active use - many manuals mention that MDADM, LVM, etc are > smart enough these days to make these calculations automatically at the > time of creation, but this does not appear to be true and that's where > problems come from later on. > Your problem is that your disks is native 4k which advertises 512b as well. This means lvm and mdadm got confused ;-) > I guess my only way to fix this is to migrate everything off that server > and reinstall it from scratch, throwing away things like MDADM and LVM > this time and replacing them with ZFS for storage purposes. > You mentioned before that you hoped to use Omnios. Latest stable now supports your nics. -- Hilsen/Regards Michael Rasmussen Get my public GnuPG keys: michael rasmussen cc http://pgp.mit.edu:11371/pks/lookup?op=get=0xD3C9A00E mir datanom net http://pgp.mit.edu:11371/pks/lookup?op=get=0xE501F51C mir miras org http://pgp.mit.edu:11371/pks/lookup?op=get=0xE3E80917 -- /usr/games/fortune -es says: I'm telling you that the kernel is stable not because it's a kernel, but because I refuse to listen to arguments like this. -- Linus Torvalds pgpWi__eiZMSh.pgp Description: OpenPGP digital signature ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On 07/19/2017 05:15 PM, Michael Rasmussen wrote: > Try do read here: > http://dennisfleurbaaij.blogspot.dk/2013/01/setting-up-linux-mdadm-raid-array-with.html Hello, Thanks, I also checked that post earlier today. Basically, it looks like MDADM array, and LVM on top (and possibly FS inside the VMs) need to be created with manual calculations for alignment and these calculations need to be specified on the command line at the time of creation. It is pity to find out this now, when server is in active use - many manuals mention that MDADM, LVM, etc are smart enough these days to make these calculations automatically at the time of creation, but this does not appear to be true and that's where problems come from later on. I guess my only way to fix this is to migrate everything off that server and reinstall it from scratch, throwing away things like MDADM and LVM this time and replacing them with ZFS for storage purposes. Thanks all. ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On Wed, 19 Jul 2017 15:10:22 +0300 Mikhailwrote: > Here's what I can see now: > > 1) fdisk output for one of disks in array: > # fdisk -l /dev/sda > > Disk /dev/sda: 3.7 TiB, 4000787030016 bytes, 7814037168 sectors > Units: sectors of 1 * 512 = 512 bytes > Sector size (logical/physical): 512 bytes / 4096 bytes > I/O size (minimum/optimal): 4096 bytes / 4096 bytes > Disklabel type: gpt > Disk identifier: 482FED1A-9CD0-4AEF-ACFC-D981C9916FE2 > > Device StartEndSectors Size Type > /dev/sda1 204819537911951744 953M Linux filesystem > /dev/sda2 1953792 7814035455 7812081664 3.7T Linux RAID > > 2) MDADM array details: > > # mdadm --detail /dev/md0 > /dev/md0: > Version : 1.2 > Creation Time : Fri Mar 18 18:27:06 2016 > Raid Level : raid10 > Array Size : 7811819520 (7449.93 GiB 7999.30 GB) > Used Dev Size : 3905909760 (3724.97 GiB 3999.65 GB) >Raid Devices : 4 > Total Devices : 4 > Persistence : Superblock is persistent > > Intent Bitmap : Internal > > Update Time : Wed Jul 19 14:58:57 2017 > State : active, checking > Active Devices : 4 > Working Devices : 4 > Failed Devices : 0 > Spare Devices : 0 > > Layout : near=2 > Chunk Size : 512K > Try do read here: http://dennisfleurbaaij.blogspot.dk/2013/01/setting-up-linux-mdadm-raid-array-with.html -- Hilsen/Regards Michael Rasmussen Get my public GnuPG keys: michael rasmussen cc http://pgp.mit.edu:11371/pks/lookup?op=get=0xD3C9A00E mir datanom net http://pgp.mit.edu:11371/pks/lookup?op=get=0xE501F51C mir miras org http://pgp.mit.edu:11371/pks/lookup?op=get=0xE3E80917 -- /usr/games/fortune -es says: He hated being thought of as one of those people that wore stupid ornamental armour. It was gilt by association. -- Terry Pratchett, "Night Watch" pgpPYhGalFprj.pgp Description: OpenPGP digital signature ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On 07/19/2017 02:43 PM, Yannis Milios wrote: > Have you checked if these drives are properly aligned, sometimes that can > cause low r/w performance. > Is there any particular reason you use mdadm instead of h/w raid controller? Hello Yannis There's no h/w raid controller because first we wanted to adopt ZFS on that storage server. I wanted to use OmniOS as a base OS, but by the time (about ~15 months ago) OmniOS did not support Intel's X550 10GiGE (no drivers in kernel) NICs we have inside that server, so had to fall back to Linux. As you know ZFS feels better when it has direct access to the drives, without h/w raid level. The MDADM RAID10 array was created without specifying any special alignment options. What's the best way to check if the drives are aligned in a proper way on existent arrwy? Here's what I can see now: 1) fdisk output for one of disks in array: # fdisk -l /dev/sda Disk /dev/sda: 3.7 TiB, 4000787030016 bytes, 7814037168 sectors Units: sectors of 1 * 512 = 512 bytes Sector size (logical/physical): 512 bytes / 4096 bytes I/O size (minimum/optimal): 4096 bytes / 4096 bytes Disklabel type: gpt Disk identifier: 482FED1A-9CD0-4AEF-ACFC-D981C9916FE2 Device StartEndSectors Size Type /dev/sda1 204819537911951744 953M Linux filesystem /dev/sda2 1953792 7814035455 7812081664 3.7T Linux RAID 2) MDADM array details: # mdadm --detail /dev/md0 /dev/md0: Version : 1.2 Creation Time : Fri Mar 18 18:27:06 2016 Raid Level : raid10 Array Size : 7811819520 (7449.93 GiB 7999.30 GB) Used Dev Size : 3905909760 (3724.97 GiB 3999.65 GB) Raid Devices : 4 Total Devices : 4 Persistence : Superblock is persistent Intent Bitmap : Internal Update Time : Wed Jul 19 14:58:57 2017 State : active, checking Active Devices : 4 Working Devices : 4 Failed Devices : 0 Spare Devices : 0 Layout : near=2 Chunk Size : 512K Check Status : 43% complete Name : storage:0 (local to host storage) UUID : 7346ef36:0a6b33f6:37eb29cd:58d04b7c Events : 1010431 Number Major Minor RaidDevice State 0 820 active sync set-A /dev/sda2 1 8 181 active sync set-B /dev/sdb2 2 8 342 active sync set-A /dev/sdc2 3 8 503 active sync set-B /dev/sdd2 3) and LVM information for the PV that resides on md0 arrway: # pvdisplay --- Physical volume --- PV Name /dev/md0 VG Name vg0 PV Size 7.28 TiB / not usable 2.00 MiB Allocatable yes PE Size 4.00 MiB Total PE 1907182 Free PE 182430 Allocated PE 1724752 PV UUID CefFFF-Q6yz-eX2p-Ziev-jdFW-3G6h-vHaesD The mdadm array is running check right now, but the speed is limited to it's defaults: # cat /proc/sys/dev/raid/speed_limit_max 20 # cat /proc/sys/dev/raid/speed_limit_min 1000 # cat /proc/mdstat Personalities : [raid10] md0 : active raid10 sda2[0] sdd2[3] sdc2[2] sdb2[1] 7811819520 blocks super 1.2 512K chunks 2 near-copies [4/4] [] [>] check = 43.6% (3412377088/7811819520) finish=17599.2min speed=4165K/sec bitmap: 16/59 pages [64KB], 65536KB chunk unused devices: Thanks for your help! ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
>> (storage server has 4x4TB SAS >> drives in RAID10 configured with MDADM) Have you checked if these drives are properly aligned, sometimes that can cause low r/w performance. Is there any particular reason you use mdadm instead of h/w raid controller? Yannis ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On 07/19/2017 12:52 PM, Emmanuel Kasper wrote: > do not use dd to benchmark storages, use fio > > with a command line like > > fio --size=9G --bs=64k --rw=write --direct=1 --runtime=60 > --name=64kwrite --group_reporting | grep bw > > inside your mount point > > or use the --filename option to point to a block device > > from this you will get reliable sequential write info Emmanuel, thanks for the hint! Just tried benchmarking with fio using your command line. Results below, looks very slow (avg=24888.52 KB/s): # fio --size=9G --bs=64k --rw=write --direct=1 --runtime=60 --name=64kwrite --group_reporting 64kwrite: (g=0): rw=write, bs=64K-64K/64K-64K/64K-64K, ioengine=sync, iodepth=1 fio-2.1.11 Starting 1 process 64kwrite: Laying out IO file(s) (1 file(s) / 9216MB) Jobs: 1 (f=1): [W(1)] [15.4% done] [0KB/1022KB/0KB /s] [0/15/0 iops] [eta 05m:34s] 64kwrite: (groupid=0, jobs=1): err= 0: pid=7841: Wed Jul 19 12:57:15 2017 write: io=1422.6MB, bw=24231KB/s, iops=378, runt= 60117msec clat (usec): min=87, max=293416, avg=2637.70, stdev=14667.15 lat (usec): min=87, max=293418, avg=2639.85, stdev=14667.17 clat percentiles (usec): | 1.00th=[ 87], 5.00th=[ 88], 10.00th=[ 88], 20.00th=[ 89], | 30.00th=[ 101], 40.00th=[ 135], 50.00th=[ 195], 60.00th=[ 235], | 70.00th=[ 334], 80.00th=[ 414], 90.00th=[ 700], 95.00th=[ 8384], | 99.00th=[81408], 99.50th=[117248], 99.90th=[193536], 99.95th=[211968], | 99.99th=[250880] bw (KB /s): min= 555, max=172928, per=100.00%, avg=24888.52, stdev=34949.10 lat (usec) : 100=29.27%, 250=32.97%, 500=25.85%, 750=2.35%, 1000=1.41% lat (msec) : 2=0.49%, 4=0.37%, 10=3.04%, 20=1.57%, 50=1.22% lat (msec) : 100=0.78%, 250=0.67%, 500=0.01% cpu : usr=0.18%, sys=1.34%, ctx=26211, majf=0, minf=8 IO depths: 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0% submit: 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% issued: total=r=0/w=22761/d=0, short=r=0/w=0/d=0 latency : target=0, window=0, percentile=100.00%, depth=1 Run status group 0 (all jobs): WRITE: io=1422.6MB, aggrb=24231KB/s, minb=24231KB/s, maxb=24231KB/s, mint=60117msec, maxt=60117msec Disk stats (read/write): dm-7: ios=0/22961, merge=0/0, ticks=0/77576, in_queue=77692, util=98.84%, aggrios=2437/28407, aggrmerge=0/0, aggrticks=0/0, aggrin_queue=0, aggrutil=0.00% md0: ios=2437/28407, merge=0/0, ticks=0/0, in_queue=0, util=0.00%, aggrios=1035/14632, aggrmerge=53/259, aggrticks=4785/68958, aggrin_queue=73796, aggrutil=67.74% sda: ios=1782/14834, merge=50/265, ticks=8488/77372, in_queue=85876, util=67.74% sdb: ios=1153/14837, merge=50/264, ticks=4460/71308, in_queue=75792, util=63.19% sdc: ios=737/14428, merge=57/254, ticks=3924/65828, in_queue=69896, util=56.76% sdd: ios=471/14431, merge=55/255, ticks=2268/61324, in_queue=63620, util=54.84% # I have also changed CPU freq. to max 3.40GHz, but looks like this was not an issue. Mikhail. ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
On 07/19/2017 11:32 AM, Mikhail wrote: > Hello, > > Thanks for your responses. > The issue appears to be somewhere beyond iSCSI. > I just tried to do some "dd" tests locally on the storage server and I'm > getting very low write speeds: do not use dd to benchmark storages, use fio with a command line like fio --size=9G --bs=64k --rw=write --direct=1 --runtime=60 --name=64kwrite --group_reporting | grep bw inside your mount point or use the --filename option to point to a block device from this you will get reliable sequential write info ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
Hello, Thanks for your responses. The issue appears to be somewhere beyond iSCSI. I just tried to do some "dd" tests locally on the storage server and I'm getting very low write speeds: root@storage:/root# dd if=/dev/vg0/isoimages of=isoimages.vg0 62914560+0 records in 62914560+0 records out 32212254720 bytes (32 GB) copied, 945.573 s, 34.1 MB/s root@storage:/root# (/dev/vg0/isoimages is local LV to the storage server) So will have to find out what's the problem or bottleneck somewhere else. Load average on storage server is 4.0-5.0 for the following CPU according to lscpu output: Model name:Intel(R) Xeon(R) CPU E3-1230 v5 @ 3.40GHz Stepping: 3 CPU MHz: 800.000 CPU max MHz: 3401. CPU min MHz: 800. Could this be due to low (800.000 MHz) CPU frequency? Thanks! On 07/19/2017 10:03 AM, Eneko Lacunza wrote: > El 19/07/17 a las 08:41, Dietmar Maurer escribió: >>> So I cannot figure out why LVM-over-iSCSI is so slow. >> I guess your benchmark is simply wrong. You are testing the >> local cache, because you do not sync the data back to the storage. > Really, 2.7GB/s for 4x4TB disks in RAID10 seems totally unreasonable (I > guess they're not SSD drives...) > > I think that in the best conditions that could give about 200-250MB/s > max, totally sequential writes, etc. > > Don't know why iSCSI is so slow, have you checked CPU usage on both sides? > > Anyhow your test copy is too small, use a file that at least is double > the available RAM on storage server, or otherwise force sync. > > Cheers > Eneko > ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
El 19/07/17 a las 08:41, Dietmar Maurer escribió: So I cannot figure out why LVM-over-iSCSI is so slow. I guess your benchmark is simply wrong. You are testing the local cache, because you do not sync the data back to the storage. Really, 2.7GB/s for 4x4TB disks in RAID10 seems totally unreasonable (I guess they're not SSD drives...) I think that in the best conditions that could give about 200-250MB/s max, totally sequential writes, etc. Don't know why iSCSI is so slow, have you checked CPU usage on both sides? Anyhow your test copy is too small, use a file that at least is double the available RAM on storage server, or otherwise force sync. Cheers Eneko -- Zuzendari Teknikoa / Director Técnico Binovo IT Human Project, S.L. Telf. 943493611 943324914 Astigarraga bidea 2, planta 6 dcha., ofi. 3-2; 20180 Oiartzun (Gipuzkoa) www.binovo.es ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
Re: [PVE-User] Shared storage on NAS speed - LVM(over iSCSI) vs NFS
> So I cannot figure out why LVM-over-iSCSI is so slow. I guess your benchmark is simply wrong. You are testing the local cache, because you do not sync the data back to the storage. ___ pve-user mailing list pve-user@pve.proxmox.com https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user