Gluster users - May 2014 - Cap on random write at 3,5 MB/s ?

Hi all,
I am currently working on my bachelor thesis, and using glusterfs 
between frontend servers that are connected to backend slow storage.

The frontend servers have a SSD cache and are iSCSI initiators to the 
slow backend storage(HDD RAID1+0).

We are using a xfs/bcache setup.

When testing performance on a gluster mounted folder from a 
clientserver, the performance shows 3,5 MB/S on my random write tests. 
This is the same result for every random write test i run.

Is this something that is limited by GlusterFS in itself? or should I 
turn my gaze at something else ?

Thanks in advance,
Kim Holmebakken

On 05/12/2014 01:57 PM, Kim Holmebakken wrote:
> Hi all,
> I am currently working on my bachelor thesis, and using glusterfs between 
> frontend servers that are connected to backend slow storage.
>
> The frontend servers have a SSD cache and are iSCSI initiators to the slow 
> backend storage(HDD RAID1+0).
>
> We are using a xfs/bcache setup.
>
> When testing performance on a gluster mounted folder from a clientserver,
the
> performance shows 3,5 MB/S on my random write tests. This is the same
result
> for every random write test i run.
>
> Is this something that is limited by GlusterFS in itself? or should I turn
my
> gaze at something else ?
>
> Thanks in advance,
> Kim Holmebakken
Hi Kim,

Very hard to understand from the above where your bottleneck is.

The usual way to tackle performance debugging is to test each layer of the 
system where possible so you can isolate the issue.

What exactly is your random write test? What specific IO pattern does it
generate?

Can you run it against the XFS/bcache setup locally without glusterfs? How does 
it do there?

Thanks!

Ric