I had similar concerns when we moved to bluestore. We run filestore clusters with HDD OSD's and SSD journals and I was worried bluestore wouldn't perform as well with the change from journals to WAL. As it turns out our bluestore clusters outperform our filestore clusters in all regards, latency, throughput and IOPS. Deferred writes aren't eliminated they go to the WAL instead of a journal. The difference is that with filestore all writes would hit the journal where with bluestore only small writes will go to the WAL while large wites will go directly to disk. _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |