[Pacemaker] [Linux-HA] pcmk + corosync + cman for dlm support?

[Vladislav Bogdanov]

03.11.2011 15:37, Nick Khamis wrote:
> Hello Vlad,
> 
> Thank you so much for your response. I am experiencing the same hang
> as well. Did you
> have better luck with GFS2, or any other network file system?

If you see almost simultaneous kernel panic on all cluster nodes, then
you probably hit the same bug (or feature? ;) ).

I tried GFS2 as well, it also had one grave bug (under high load) but
that one is fixed in all fedoras after 13 I hope and hopefully in el6
too. You can search redhat bugzilla for bug I opened about that. Right
now I do not use any of them, but plan to return to GFS2 after some time.

>From what I saw, OCFS2 is (was) much faster (up to 10 times) then GFS2
in almost all use cases. But when it comes to stability, it is
absolutely unusable (for me).

For other filesystems - I do not know any competing to that two above
(non-distributed clustered fs on a shared/mirrored block device). You
may look at lustre, GPFS, pvfs, ceph, but they are completely different
(distributed) and wont probably fit your storage design. Also they all
have drawbacks: lustre is available only for el5 (afaik), GPFS is not
free, ceph is still alpha-quality and not recommended for production.

Best,
Vladislav