I also would like to see cephfs stable, especially with the snapshot function.
I tried to figure out the roadmap but couldn't get a clear picture?
Is there a target date for production-ready snapshot-functionality?
until than a possible alternative (sorry without ceph :-/)
is using glusterfs which can be really fast.
2 years ago I had a setup utilizing raid6 bricks consisting each of 7+1hotspare disks (1TB sata)
several of them in a gluster stripe connected via 4GB FC (needed to be cheap ;-) to a server (debian)
that exported the space via samba.
I liked it because it was:
- really fast
- really robust
- as cheap as poss (for huge productive data IMHO)
- easy to set up and maintain
- smoothly scalable ad infinitum
(one can start with one server, one raid-array than grow for volume and redundancy/off-site replication)
big drawback: no snapshots, no easy readonly / cow functionality,
that's what I hope cephfs will bring us!
I tried it since some days, and it works, mds hasn't crashed (yet ;-)
it took 2TB of data with acceptable performance - BUT
erasing that data is a no go :-( 13MB/s??
Again, is there any roadmap on cephfs (incl. snaps?)
best regards
Bernhard
Actually #3 is a novel idea, I have not thought of it. Thinking about the difference just off the top of my head though, comparatively, #3 will have1) more overheads (because of the additional VM)2) Can't grow once you reach the hard limit of 14TB, and if you have multiple of such machines, then fragmentation becomes a problem3) might have the risk of 14TB partition corruption wiping out all your shares4) not as easy as HA. Although I have not worked HA into NFSCEPH yet, it should be doable by drdb-ing the NFS data directory, or any other techniques that people use for redundant NFS servers.- WPOn Fri, Nov 15, 2013 at 10:26 PM, Gautam Saxena <gsaxena@xxxxxxxxxxx> wrote:Yip,I went to the link. Where can the script ( nfsceph) be downloaded? How's the robustness and performance of this technique? (That is, is there are any reason to believe that it would more/less robust and/or performant than option #3 mentioned in the original thread?)On Fri, Nov 15, 2013 at 1:57 AM, YIP Wai Peng <yipwp@xxxxxxxxxxxxxxx> wrote:On Fri, Nov 15, 2013 at 12:08 AM, Gautam Saxena <gsaxena@xxxxxxxxxxx> wrote:1) nfs over rbd (http://www.sebastien-han.fr/blog/2012/07/06/nfs-over-rbd/)We are now running this - basically an intermediate/gateway node that mounts ceph rbd objects and exports them as NFS. http://waipeng.wordpress.com/2013/11/12/nfsceph/
- WP
--
Bernhard Glomm
Network & System Administrator
Ecologic Institute
bernhard.glomm@xxxxxxxxxxx
www.ecologic.eu
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
