Yes that seems about right. Erasure coding works similar to Raid5/6 so data will be striped, whereas replicated pools are simple copies of data. When you write to a 3x replicated pool you are having to write 3 times as much and so performance is lower. When writing to an erasure coded pool (k=8 m=2 for example) you are only having to write 20% worth of redundancy data. As for reads, in an ideal world both would have similar performance. But currently erasure coded pools require all chunks to be returned to satisfy a read request, even if the request is smaller than a whole object. Also the read request hits the primary OSD and then it forwards the remaining requests onto all the other OSD's which hold the remaining erasure chunks. Another factor is that in erasure pools, as the object is split into smaller chunks you will be getting less performance from spinning disks as disks give the most bandwidth when doing 4MB+ IO's. All of these reasons tend to make reading slightly slower. > -----Original Message----- > From: ceph-users [mailto:ceph-users-bounces@xxxxxxxxxxxxxx] On Behalf Of > M Ranga Swami Reddy > Sent: 24 January 2016 15:09 > To: ceph-users <ceph-users@xxxxxxxx> > Subject: Performance - pool with erasure/replicated type pool > > Hello, > I have 2 pools as below: > > 1. pool1 with erasure type > 2. pool2 with replicated type > > I ran the "rados bench" with above 2 pool and the results came as below: > > - Read performance - around 60% better for replicated type pool ie pool2 > - Write performance - around 50 % better for erasure type pool ie pool1. > > Can someone help me, if the above is correct behavior with ceph? > Or do I miss something here? > > Thanks > Swami > _______________________________________________ > 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
 |