Hi oVirt gurus,
Thank s to Tony, who pointed me into discovery process, the performance of the IO seems greatly dependent on the flags.
[root@lgu215 ~]# dd if=/dev/zero of=/tmp/test1.img bs=512 count=100000
100000+0 records in
100000+0 records out
51200000 bytes (51 MB) copied, 0.108962 s, 470 MB/s
[root@lgu215 ~]# dd if=/dev/zero of=/tmp/test1.img bs=512 count=100000 oflag=dsync
100000+0 records in
100000+0 records out
51200000 bytes (51 MB) copied, 322.314 s, 159 kB/s
Dsync flag tells dd to ignore all buffers, cache except certain kernel buffers and write data physically to the disc, before writing further. According to number of sites I looked at, this is the way to test Server Latency in regards to IO operations. Difference in performance is huge, as you can see (below I have added results from tests with 4k and 8k block)
Still, certain software component we run tests with writes data in this/similar way, which is why I got this complaint in the first place.
Here is my current NFS mount settings:
rw,relatime,vers=4.0,rsize=65536,wsize=65536,namlen=255,soft,nosharecache,proto=tcp,timeo=600,retrans=6,sec=sys,clientaddr=172.17.28.11,local_lock=none,addr=172.17.28.5
If you have any suggestions on possible NFS tuning options, to try to increase performance, I would highly appreciate it.
Can someone tell me how to change NFS mount options in oVirt for already existing/used storage?
Test results with 4096 and 8192 byte size.
[root@lgu215 ~]# dd if=/dev/zero of=/tmp/test1.img bs=4096 count=100000
100000+0 records in
100000+0 records out
409600000 bytes (410 MB) copied, 1.49831 s, 273 MB/s
[root@lgu215-admin ~]# dd if=/dev/zero of=/tmp/test1.img bs=4096 count=100000 oflag=dsync
100000+0 records in
100000+0 records out
409600000 bytes (410 MB) copied, 349.041 s, 1.2 MB/s
[root@lgu215 ~]# dd if=/dev/zero of=/tmp/test1.img bs=8192 count=100000
100000+0 records in
100000+0 records out
819200000 bytes (819 MB) copied, 11.6553 s, 70.3 MB/s
[root@lgu215-admin ~]# dd if=/dev/zero of=/tmp/test1.img bs=8192 count=100000 oflag=dsync
100000+0 records in
100000+0 records out
819200000 bytes (819 MB) copied, 393.035 s, 2.1 MB/s
From: "Vrgotic, Marko" <M.Vrgotic@activevideo.com>
Date: Thursday, 26 September 2019 at 09:51
To: Amit Bawer <abawer@redhat.com>
Cc: Tony Brian Albers <tba@kb.dk>, "hunter86_bg@yahoo.com" <hunter86_bg@yahoo.com>, "users@ovirt.org" <users@ovirt.org>
Subject: Re: [ovirt-users] Re: Super Low VM disk IO via Shared Storage
Dear all,
I very much appreciate all help and suggestions so far.
Today I will send the test results and current mount settings for NFS4. Our production setup is using Netapp based NFS server.
I am surprised with results from Tony’s test.
We also have one setup with Gluster based NFS, and I will run tests on those as well.
Sent from my iPhone
On 25 Sep 2019, at 14:18, Amit Bawer <abawer@redhat.com> wrote:
On Wed, Sep 25, 2019 at 2:44 PM Tony Brian Albers <tba@kb.dk> wrote:
Guys,
Just for info, this is what I'm getting on a VM that is on shared
storage via NFSv3:
--------------------------snip----------------------
[root@proj-000 ~]# time dd if=/dev/zero of=testfile bs=4096
count=1000000
1000000+0 records in
1000000+0 records out
4096000000 bytes (4.1 GB) copied, 18.0984 s, 226 MB/s
real 0m18.171s
user 0m1.077s
sys 0m4.303s
[root@proj-000 ~]#
--------------------------snip----------------------
my /etc/exports:
/data/ovirt
*(rw,sync,no_subtree_check,all_squash,anonuid=36,anongid=36)
and output from 'mount' on one of the hosts:
sto-001.kac.lokalnet:/data/ovirt on /rhev/data-center/mnt/sto-
001.kac.lokalnet:_data_ovirt type nfs
(rw,relatime,vers=3,rsize=1048576,wsize=1048576,namlen=255,soft,nolock,
nosharecache,proto=tcp,timeo=600,retrans=6,sec=sys,mountaddr=172.16.216
.41,mountvers=3,mountport=20048,mountproto=udp,local_lock=all,addr=172.
16.216.41)
Worth to compare mount options with the slow shared NFSv4 mount.
Window size tuning can be found at bottom of [1], although its relating to NFSv3, it could be relevant to v4 as well.
connected via single 10gbit ethernet. Storage on NFS server is 8 x 4TB
SATA disks in RAID10. NFS server is running CentOS 7.6.
Maybe you can get some inspiration from this.
/tony
On Wed, 2019-09-25 at 09:59 +0000, Vrgotic, Marko wrote:
> Dear Strahil, Amit,
>
> Thank you for the suggestion.
> Test result with block size 4096:
> Network storage:
> avshared:
> [root@mpoludctest4udc-1 ~]# dd if=/dev/zero of=/tmp/test2.img bs=4096
> count=100000 oflag=dsync
> 100000+0 records in
> 100000+0 records out
> 409600000 bytes (410 MB) copied, 275.522 s, 1.5 MB/s
>
> Local storage:
>
> avlocal2:
> [root@mpollocalcheck22 ~]# dd if=/dev/zero of=/tmp/test2.img bs=4096
> count=100000 oflag=dsync
> 100000+0 records in
> 100000+0 records out
> 409600000 bytes (410 MB) copied, 53.093 s, 7.7 MB/s
> 10:38
> avlocal3:
> [root@mpollocalcheck3 ~]# dd if=/dev/zero of=/tmp/test2.img bs=4096
> count=100000 oflag=dsync
> 100000+0 records in
> 100000+0 records out
> 409600000 bytes (410 MB) copied, 46.0392 s, 8.9 MB/s
>
> As Amit suggested, I am also going to execute same tests on the
> BareMetals and between BareMetal and NFS to compare results.
>
>
> — — —
> Met vriendelijke groet / Kind regards,
>
> Marko Vrgotic
>
>
>
>
> From: Strahil <hunter86_bg@yahoo.com>
> Date: Tuesday, 24 September 2019 at 19:10
> To: "Vrgotic, Marko" <M.Vrgotic@activevideo.com>, Amit <abawer@redhat
> .com>
> Cc: users <users@ovirt.org>
> Subject: Re: [ovirt-users] Re: Super Low VM disk IO via Shared
> Storage
>
> Why don't you try with 4096 ?
> Most block devices have a blcok size of 4096 and anything bellow is
> slowing them down.
> Best Regards,
> Strahil Nikolov
> On Sep 24, 2019 17:40, Amit Bawer <abawer@redhat.com> wrote:
> have you reproduced performance issue when checking this directly
> with the shared storage mount, outside the VMs?
>
> On Tue, Sep 24, 2019 at 4:53 PM Vrgotic, Marko <M.Vrgotic@activevideo
> .com> wrote:
> Dear oVirt,
>
> I have executed some tests regarding IO disk speed on the VMs,
> running on shared storage and local storage in oVirt.
>
> Results of the tests on local storage domains:
> avlocal2:
> [root@mpollocalcheck22 ~]# dd if=/dev/zero of=/tmp/test2.img bs=512
> count=100000 oflag=dsync
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 45.9756 s, 1.1 MB/s
>
> avlocal3:
> [root@mpollocalcheck3 ~]# dd if=/dev/zero of=/tmp/test2.img bs=512
> count=100000 oflag=dsync
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 43.6179 s, 1.2 MB/s
>
> Results of the test on shared storage domain:
> avshared:
> [root@mpoludctest4udc-1 ~]# dd if=/dev/zero of=/tmp/test2.img bs=512
> count=100000 oflag=dsync
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 283.499 s, 181 kB/s
>
> Why is it so low? Is there anything I can do to tune, configure VDSM
> or other service to speed this up?
> Any advice is appreciated.
>
> Shared storage is based on Netapp with 20Gbps LACP path from
> Hypervisor to Netapp volume, and set to MTU 9000. Used protocol is
> NFS4.0.
> oVirt is 4.3.4.3 SHE.
>
>
> _______________________________________________
> Users mailing list -- users@ovirt.org
> To unsubscribe send an email to users-leave@ovirt.org
> Privacy Statement: <
> _______________________________________________
> Users mailing list -- users@ovirt.org
> To unsubscribe send an email to users-leave@ovirt.org
> Privacy Statement: https://www.ovirt.org/site/privacy-policy/
> oVirt Code of Conduct: https://www.ovirt.org/community/about/communit
> y-guidelines/
> List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/
> message/7XYSFEGAHCWXIY2JILDE24EVAC5ZVKWU/
--
Tony Albers - Systems Architect - IT Development
Royal Danish Library, Victor Albecks Vej 1, 8000 Aarhus C, Denmark
Tel: +45 2566 2383 - CVR/SE: 2898 8842 - EAN: 5798000792142