Quoting Alan Brown <ajb2@mssl.ucl.ac.uk>:
I can't speak for OCFS2, but after several years' experience with
the filesystem I strongly recommend NOT using GFS/GFS2.
GFS2 works for me. GFS was a bit slow, but GFS2 meets my (perhaps low) needs. Of course, my use case is different (e.g., I don't use it with quotas, in fact I've stayed away from it anywhere I need quotas).
Its locking model is incredibly slow (500 locks/second on a
filesystem mounted with quotas enabled and noatime) and results in
dire performance - plus
Did you raise plock_rate_limit from its defaults? The defaults will indeed suck.
there's a known crash vulnerability if files are repeatedly renamed
in large directories (this bites us regularly...)
Never had that happen... But then, my applications don't repeatedly rename files in large directories (e.g., I don't use maildir) and I don't use quotas on GFS.
GFS2 isn't an "enterprise" filesystem by any stretch of the
imagination, despite what a number of enthusiastic salespeople might
try to convince you of. We're lucky to keep the GFS servers up for
more than a week at a time.
GFS isn't for all applications. I've used it for 2 different applications for which it has proven well suited. Every discussion I've seen about GFS has always said "don't use it with large directories of small files" so if that is your use case, then you must be ignoring common wisdom...
Given the right use case (including dovecot with mbox and dovecot indexes) it seems to work fine... I've used it for another project also without problems (been running for years now in both cases).
-- Eric Rostetter The Department of Physics The University of Texas at Austin
Go Longhorns!