[Bugs] [Bug 1200457] gluster performance issue as data is added to volume. tar extraction of files goes from 1-minute on empty volume to 20-minutes on volume with 40TB.

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1200457



--- Comment #10 from Shyamsundar <srangana at redhat.com> ---
C) Strace from David points to the following,

linkat and symlink take a high amount of average latency on the bricks that
belong to the slow volume and are fast on the bricks that belong to the fast
volume.

The order of magnitude is very high as follows,

Fast brick:
Avg. linkat latency from strace: 0.000012 seconds
#calls: 22025
Time taken for linkat: 0.000012 * #calls = 0.26914 seconds

Slow brick:
Avg. linkat latency from strace: 0.0103351812 seconds
#calls: 22025
Time taken for linkat: 227 seconds

Similar data is seen for symlink into the .glusterfs directories.

Overall the latency here is for a single brick, extrapolating this (i.e
multiplying this by 2) shows that we would save ~450 seconds on linkat if this
was fast, which points to the problem being linkat and symlink into the
.glusterfs directories.

Looking at option B (in comment #9) and possibly getting xfs_metadump from
David and also some entry counts in the 5 best and worst linkat directories to
see how to proceed with looking at the same from an XFS/VFS perspective.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
Unsubscribe from this bug https://bugzilla.redhat.com/token.cgi?t=XiQSMp6QCv&a=cc_unsubscribe