[Gluster-users] Is my (low) performance expected?

[Nick Demou]

Hi on a test setup of 2 nodes with replication over a 1Gbit link I'm seeing
extremely low performance while deleting thousands of small files (in the
order of a few hundred times slower than hitting the disks directly ~300
unlinks/min to give you a number). The setup is very basic -- I've just
followed instructions from the CentOS wiki. Before profiling and
investigating I wanted to know if this speed is to be expected with one
dedicated 1Gbit link between the 2 nodes using the FUSE client (my CPU,
disks and memory are far from being over-utilized). If my performance is
normal is there something I can do to improve performance for small file
operations on this H/W?

Nick

P.S.
Here are a few more details for the curious: While working directly on the
filesystem that hosts the bricks I can delete tens of thousands of small
files instantly. I've mounted the volume from one of the nodes with the
fuse client and I'm deleting from there. strace shows that most unlinks
take ~0.1-0.2sec and every once in a while one unlinkat call may take
several second.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20170505/532dea49/attachment.html>