Hi. I'm looking for a distributed filesystem that would distribute a single namespace with POSIX access. It should be scaleable upwards to be easily to add more physical drives as storage fills up with time. The deepscrub and snapshotting are essential. I have been studying different options and would consider Ceph and CephFS the best option. I have a two site setup with fairly poor connection in between. That would have two MDS in active-active configuration. I'm considering to have two copies on each site, so total of 4 copies of the data. For reading this is not a problem, but for writing as I understand in my case the client would have to write 4 copies until the write is acked. Is it possible to set Ceph up so that to write would be satisfied by two writes on the local site and Rados would then complete the write to remote site asyncronously? This is a fairly small system and the requirements for different servers is probably a bit overkill... As the manual suggests that clients should not run on the OSD's. Is it sufficient to run OSD on a server and then have a Linux container to run the client etc? If container is not a sufficient barrier would a virtualized server do? Does this sound feasible? Would you have any suggestions on this before I start deploying and testing this... Thank you all... And sorry for the newbie q. -- -Matti _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com