I can't get the idea. Can you give an example.

let me share some of my setup.
1. Volume -> data_fast is consisting of:
ovirt1:/gluster_bricks/data_fast/data_fast -> 500GB NVMe
ovirt2:/gluster_bricks/data_fast/data_fast -> 500GB NVMe
ovirt3:/gluster_bricks/data_fast/data_fast -> small LV on a slow (QLC-based) SATA SSD

All hosted on thin LVM.
Of course , for the engine I have:
ovirt1:/gluster_bricks/engine/engine -> SATA ssd shared between OS and brick
ovirt2:/gluster_bricks/engine/engine -> SATA ssd shared between OS and brick
ovirt3:/gluster_bricks/engine/engine -> SATA ssd shared between OS and 4 other bricks

Since I have switched from old HDDs to consumer SSD disks - the engine volume is not reported by sanlock.service , despite Gluster v52.XX has higher latency.

Best Regards,
Strahil Nikolov



В сряда, 24 април 2019 г., 21:25:10 ч. Гринуич-4, Leo David <leoalex@gmail.com> написа:


Thank you very much Strahil, very helpful. As always. So I would equip the 3rd server and alocate one small ( 120 - 240gb) consumer grade ssd for each of the gluster volume, and at volume creation, to specify the small ssds as the 3rd brick.
Do it make sense ?
Thank you !

On Wed, Apr 24, 2019, 18:10 Strahil <hunter86_bg@yahoo.com> wrote:

I think 2 small ssds (raid 1 mdadm) can do the job better as ssds have lower latencies .You can use them both for OS (minimum needed is 60 GB) and the rest will be plenty for an arbiter.
By the way, if you plan using gluster snapshots - use thin LVM for the brick.

Best Regards,
Strahil Nikolov

On Apr 24, 2019 16:20, Leo David <leoalex@gmail.com> wrote:
Hello Everyone,
I need to look into adding some enterprise grade sas disks ( both ssd and spinning  ),  and since the prices are not too low,  I would like to benefit of replica 3 arbitrated.
Therefore,  I intend to buy some smaller disks for use them as arbiter brick.
My question is, what performance ( regarding iops,  througput ) the arbiter disks need to be. Should they be at least the same as the real data disks ?
Knowing that they only keep metadata, I am thinking that will not be so much pressure on the arbiters.
Any thoughts?

Thank you !


--
Best regards, Leo David