Hi Michael,
On 12/28/2018 12:39 PM, Prof. Dr. Michael Schefczyk wrote:
Dear Ravi,
Again thank you very much for your response. I did experiment a bit further in my scenario, with the following result:
I could leave 4.1.6 for the smaller machines with all SSD. The performance does seem to be a bit lower compared to 3.12.15, but within a bearable range. For me the tradeoff is more time until EOL vs. performance.
For the larger pair of machines with 1 x SATA SSD and 1 x fast SAS HDD, I did NEED to migrate back to 3.12.15 and stay at that. There, gluster volume heal <volname> info does not always show healing, but when writing is heavy, the system temporarily lists shards for healing. That does seem to bring the overall performance down. My settings are features.shard on, features.shard-block-size 512MB, cluster.data-self-heal-algorithm full and performance.low-prio-threads 32.
Can you try disabling all client-side self-heals and see if it gives a
better performance on your 3.12.5 setup?
gluster volume set $volname data-self-heal off
gluster volume set $volname metadata-self-heal off
gluster volume set $volname entry-self-heal off
Also, since you are hosting VM images on gluster using ovirt, I hope you
have all the volume options for the virt profile
(https://github.com/gluster/glusterfs/blob/master/extras/group-virt.example)
enabled.
But if I understand your original email right, you see a degradation in
performance between 3.12.15 and 4.1.6. A good way to debug this would
be to run identical workload on both version and capture the profile
info
(https://docs.gluster.org/en/latest/Administrator%20Guide/Monitoring%20Workload/)
and see if you can figure out what is different in 4.1.6. You might want
to raise a bug with the volume info and the profile output of both
versions attached to the bug if that is something which you can
experiment and provide.
Regards,
Ravi
Is there a way to cure the performance drop while using the more modern versions of Gluster?
Regards,
Michael
-----Ursprüngliche Nachricht-----
Von: Prof. Dr. Michael Schefczyk
Gesendet: Mittwoch, 12. Dezember 2018 07:55
An: gluster-users@xxxxxxxxxxx
Betreff: AW: Gluster 4.1.6 slow
Dear Ravi,
Thank you for your response. While performance was slow, there were no volume heal tasks pending. All analysis using the usual commands (volume status, volume info, volume heal $volname info) did not show issues. One could kick off volume heal without adding "info" (e. i. gluster volume heal $volname) which would show no issues as well.
Regards,
Michael
-----Ursprüngliche Nachricht-----
Von: gluster-users-bounces@xxxxxxxxxxx <gluster-users-bounces@xxxxxxxxxxx> Im Auftrag von Prof. Dr. Michael Schefczyk
Gesendet: Dienstag, 11. Dezember 2018 08:33
An: gluster-users@xxxxxxxxxxx
Betreff: Re: Gluster 4.1.6 slow
Dear All,
Investing a few hours, I did revert to Gluster 3.12.15 from the oVirt repository. With the same settings, the system does work as fast as before. That required two tries. It looks like restarting Gluster after volume deletion and downgrade was not enough. The system needed to be rebooted.
The uncomfortable thing is that it is not obvious why the problem did happen and if it will happen again once an upgrade from 3.12.15 becomes more pressing.
Regards,
Michael
_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users
_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users
_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users
