On 04/24/2013 05:18 AM, Maik Kulbe wrote: > Hi, > > I'm currently looking into several options on how to use ceph in a small > to mid size web cluster. > > I've ruled out CephFS as it is sadly not stable enaugh. > > Then I went with RBD and different approaches. OCFS2 on RBD did the job > well but had extrem performance issues when two processes where > accessing files in the same folder(especially when doing writes). Any idea if this was more due to OCFS2 or more due to Ceph? I confess I don't know much about how OCFS2 works. Is it doing some kind of latency sensitive operation when two files are being written per directory?
I can't say for sure but it seemed that the locking mechanisms of OCFS2 locked the whole folder. I tested concurrent writes in one folder and in several folders and using multiple folders I could reach about 3K Ops per second, with a single folder those rates dropped to under 2 (single!) Ops per second, which is kind of strange but I wouldn't think that an rbd volume could be responsible for that..
> > At the moment I'm trying a solution that uses RBD with a normal FS like > EXT4 or ZFS and where two server export that block device via NFS(with > heartbeat for redundancy and failover) but that involves problems with > file system consistency. > > My question here is, what kind of software stack other users here would > suggest for this kind of workload? > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
