Hi all, after basically throwing away the SSDs we had because of very poor journal write performance, I tested our test systems with spindle drives only. The results are quite horrifying and i get the distinct feeling that i am doing something wrong somewhere. So read performance is great, giving me full 10gbe (1078MB/s avg), no problem at all… But doing a rados bench gives me 190MB/s write, max.
I have 14 OSDs all SATA drives. and even if its only 20MB/s per drive throughput, I would expect above 240MB/s but it looks to me like the cluster never really uses more than 1 or 2 nodes, i have 1 pool for testing with 256 PGs and its set to replicate x2… yet if i make 3 replicas, the speed remains the same. Is there any kind of checking i can do if all drives really are used or can i force this? Thanks already //F _______________________________________________ ceph-users mailing list [email protected] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
