[Gluster-users] SSD tier experimenting

Mohammed Rafi K C rkavunga at redhat.com
Thu May 5 10:53:09 UTC 2016

On 05/05/2016 03:59 PM, Sergei Hanus wrote:
> Hi, Dan, Mohammed.
> Sorry for the long pause, was busy with another project.
> I was able to test the situation again - and looks like you are right.
> When I create non-tiered volume, put load on it and then attach tier -
> performance gets worse.
> Speaking numbers:
> I use fio for testing, fio --direct=1 --rw=randrw --norandommap
> --randrepeat=0 --ioengine=sync --bs=4k --rwmixread=70 --iodepth=4
> --numjobs=8 --runtime=20 --group_reporting --size=2G --name=randwrite
> --refill_buffers
> I get around 140 iops with non-tiered disk, and around 70 iops - when
> adding tier.

As I mentioned in previous mail, there are couple of things related to
metedata, that tiering does for first time after attaching, Eventually
you should get an ideal performance benefit close to that 900 iops.

> When I create tiered volume from the very beginning, the same load
> profile gives performance around 900 iops
> Also, I'm planning to experiment with io-cache parameter in the
> future, to see how it changes the behavior. If you are interested - I
> could share results.

great, I would be very much inerested to see this. Also other community
users can also benefit.

> Sergei.

More information about the Gluster-users mailing list