Hey, Im running a new ceph 13 cluster, using just one cephfs, 6.3 erasure encoded stripe pool, each osd is 10T hdd, 20 total, all on there own host. Storing mostly large files ~20G. I'm running mostly stock except that I've optimized for the low (2G) memory hosts based an old threads recommendations.
I'm trying to fill it and test various failure scenarios and by far my biggest bottleneck is iops for both writing and recovery. I'm guessing because of the journal write + block write (seeing roughly 30MiB/s for 100iops). SSD for the journal is not possible. Am I correct in saying that I'm really only able to reduce/influence iops/MiB for the block write? Is the correct way to increase that is to increase the stripe_unit by say 3x to achieve 100MiB/s per osd? Daniel
_______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
