On Wed, Jun 1, 2016 at 8:32 AM, Alexandre DERUMIER <aderumier@xxxxxxxxx> wrote: > Hi, > > I'm begin to look at rbd mirror features. > > How much space does it take ? Is it only a journal with some kind of list of block changes ? There is a per-image journal which is a log of all modifications to the image. The log will be trimmed as the remote peer consumes the entries. We have a feature tracker ticket to add support for disconnecting laggy peers if they are behind by some roughly configurable amount of space. Unfortunately, until that feature lands, the journal can continue to grow if your peer cannot keep up with the journal entries. > and how much io/s does it take ? > worst case, 4k block write , how much write in journal ? A 4K write will result in a slightly more than 4K journal entry (the journal entries have a small header, CRC, etc). > > My osd are ssd, 1journal + data for each osd/ssd, but I don't overloaded them with write currently. The journal entries are append-only operations so they should be fairly light ops on the OSD side as compared to RBD block (over)writes. You can also optionally configure journal event batching via the 'rbd_journal_object_flush_age' config option if you can support losing (crash consistent) X seconds of updates. > > Regards, > > Alexandre > > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com -- Jason _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |