[ovirt-users] Questions about converged infrastructure setup and glusterFS sizing/performance

Jayme jaymef at gmail.com
Fri Jan 19 14:14:44 UTC 2018


I am attempting to narrow down choices for storage in a new oVirt build
that will eventually be used for a mix of dev and production servers.

My current space usage excluding backups sits at about only 1TB so I figure
3-5 TB would be more than enough for VM storage only + some room to grow.
There will be around 24 linux VMs total but 80% of them are VERY low usage
and low spec servers.

I've been considering a 3 host hyperconverged oVirt setup, replica 3
arbiter 1 setup with a disaster recovery plan to replicate the gluster
volume to a separate server.  I would of course do additional incremental
backups to an alternate server as well probably with rsync or some other
method.

Some questions:

1. Is it recommended to use SSDs for glusterFS or can the performance of
regular server/sas drives be sufficient enough performance.  If using SSDs
is it recommended to use enterprise SSDs are consumer SSDs good enough due
to the redundancy of glusterFS?   I would love to hear of any use cases
from any of you regarding hardware specs you used in hyperconverged setups
and what level of performance you are seeing.

2. Is it recommended to RAID the drives that form the gluster bricks?  If
so what raid level?

3. How do I calculate how much space will be usable in a replicate 3
arbiter 1 configuration?  Will it be 75% of total drive capacity minus what
I lose from raid (if I raid the drives)?

4. For replication of the gluster volume, is it possible for me to
replicate the entire volume to a single drive/raid array in an alternate
server or does the replicated volume need to match the configuration of the
main glusterFS volume (i.e. same amount of drives/configuration etc).

5. Has the meltdown bug caused or expected to cause major issues with oVirt
hyperconverged setup due to performance loss from the patches.  I've been
reading articles suggesting up to 30% performance loss on some
converged/storage setups due to how CPU intensive converged setups are.

Thanks in advance!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20180119/b99daec7/attachment.html>


More information about the Users mailing list