[Gluster-users] self-heal performance regression in 3.6

[Florent Monbillard]

Hi all,

Has anybody observed a performance regression in the self-heal process between 3.4 and 3.6 gluster releases ?

The best self-heal performance I achieved was using 3.4.2 and "self-heal-window-size=4” on a large amount of data (several TB) with a mix of small and large files (from 100KB to 4MB).
Using self-heal-window-size default value (16) on 3.4.2 or any setting (4 or 16) with 3.6.2 lead to 33% (up to 50%) slowdown compared to the best performance baseline with the same dataset.
With several TB of data, that requires many extra hours on 3.6.2 for the self-heal to complete.

Other than self-heal-window-size parameter, is there any self-heal performance settings or speed-up tricks I could use either on 3.4 or 3.6 ?
Any feedback appreciated.

Best regards,
Florent Monbillard
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20150410/065256dc/attachment.html>