Bluestore's deferred write queue doesn't act like Filestore's journal because a) it's very small = 64 requests b) it doesn't have a background flush thread. Bluestore basically refuses to do writes faster than the HDD can do them _on_average_. With Filestore you can have 1000-2000 write iops until the journal becomes full. After that the performance will drop to 30-50 iops with very unstable latency. With Bluestore you only get 100-300 iops, but these 100-300 iops are always stable :-)
I'd recommend bcache. It should perform much better than ceph's tiering.
--
With best regards,
Vitaliy Filippov
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
