[ovirt-users] Some tests results: lustrefs over nfs on VM

[Yaniv Kaul]

On Nov 19, 2017 11:05 PM, "Arman Khalatyan" <arm2arm at gmail.com> wrote:

hi Yaniv,
yes for sure I hit some cache in between but not the vm cache, it has a 4GB
ram, with oflag=direct I get about 120MB/s

for the data analysis the cache is our friend:)

The backend is a lustrefs 2.10.x.
yes we have dedicated 10G on the hosts, where we can limit the vm interface
to 10Gbit?


Define QoS on the NIC.
But I think you wish to limit IO, no?
Y.


Am 19.11.2017 8:33 nachm. schrieb "Yaniv Kaul" <ykaul at redhat.com>:



On Sun, Nov 19, 2017 at 7:08 PM, Arman Khalatyan <arm2arm at gmail.com> wrote:

> Hi, in our environment we got pretty good io performance on VM, with
> following configuration:
> lustrebox: /lust mounted on "GATEWAY" over IB
> GATEWAY: export /lust as nfs4 on 10G interface
> VM(test.vm): import as NFS over 10G interface
>
> [root at test.vm <root at gans.vm> ~]# dd  if=/dev/zero bs=128K count=100000
>

Without oflag=direct, you are hitting (somewhat) the cache.


>
> of=/test/tmp/test.tmp
> 100000+0 records in
> 100000+0 records out
> 13107200000 <(310)%20720-0000> bytes (13 GB) copied, 20.8402 s, 629 MB/s
> looks promising for the future deployments.
>

Very - what's the backend storage?


>
> only one problem remains that on heavy io I get some wornings that the vm
> network is saturated, are there way to configure the bandwidth limits to
> 10G for the VM Interface??
>

Yes, but you really need a dedicated storage interface, no?
Y.


>
>
> thank you beforehand,
> Arman.
>
> _______________________________________________
> Users mailing list
> Users at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20171120/21036afe/attachment.html>