Reverting back to filestore is quite a lot of work and time again. Maybe see first if with some tuning of the vms you can get better results? What you also can try is for io intensive vm's add an ssd pool? I moved some exchange servers on them. Tuned down the logging, because that is writing constantly to disk. With such setup you are at least secured for the future. -----Original Message----- From: Stuart Longland [mailto:stuartl@xxxxxxxxxxxxxxxxxx] Subject: Re: Future of Filestore? > > Maybe a bit of topic, just curious what speeds did you get previously? > Depending on how you test your native drive of 5400rpm, the > performance could be similar. 4k random read of my 7200rpm/5400 rpm > results in ~60iops at 260kB/s. Well, to be honest I never formally tested the performance prior to the move to Bluestore. It was working "acceptably" for my needs, thus I never had a reason to test it. It was never a speed demon, but it did well enough for my needs. Had Filestore on BTRFS remained an option in Ceph v12, I'd have stayed that way. > I also wonder why filestore could be that much faster, is this not > something else? Maybe some dangerous caching method was on? My understanding is that Bluestore does not benefit from the Linux kernel filesystem cache. On paper, Bluestore *should* be faster, but it's hard to know for sure. Maybe I should try migrating back to Filestore and see if that improves things? -- Stuart Longland (aka Redhatter, VK4MSL) I haven't lost my mind... ...it's backed up on a tape somewhere. _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |