[ovirt-users] Gluster Performance

Denis Chaplygin dchaplyg at redhat.com
Thu Oct 5 12:07:18 UTC 2017


Hello!

On Thu, Oct 5, 2017 at 12:21 AM, Bryan Sockel <Bryan.Sockel at altn.com> wrote:

> Is there any performance loss from your  if your gluster replica 3 servers
> are not all the same configuration?
>
>
> Server 1 - Primary
> 16 X 1.2 TB 10k Drives - Raid 10 - Stripe 256k (2.5 Drives)
> 1 CPU - Intel(R) Xeon(R) CPU E5-2609 v3 @ 1.90GHz
> 64 GB Memory
>
> Server 2 - Replica
> 8 X 6 TB 7.5 RPM Drives - Raid 10 - Trip 512 (3.5)
> 1 CPU - Intel(R) Xeon(R) CPU E5-2609 v3 @ 1.90GHz
> 16 GB Memory
>
>

As gluster needs to copy each 'write' operation twice and it waits for
actual completion of every op, your write iops will be limited by the
slowest drive.



> Server 3 - Arbiter (Also Virtual machine node)
> 2 X 1.2 TB 10k Drives Raid 10
> 2 CPU - Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
> 192 GB Memory
>
>

But it will be relaxed for the arbiter node, as arbiter keeps no data,
therefore no writes* happens on it, so it doesn't impacts overall
performance


* There are actually some metadata updates on the arbiter node, which are,
obviously, write operations, but we can neglect them.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20171005/888eb76e/attachment.html>


More information about the Users mailing list