[Gluster-devel] Hammering start() calls

[Anand Avati]

> I have a workload where around 10000 stat() calls per second must be
> hammered on a mounted glusterfs share. So far I suffer very bad performance,
> way worse than NFS. I tried with and without all performance translators
> (besides booster), as well as increasing the -e -a options..


-a and -e helps only for the 2nd stat and beyond. are the 10000 stat() calls
being performed on 10000 different files? what are the total number of files
and are they re-stat()ed? how many mount points and threads per moutn point?

avati