On 3 May 2017 at 17:24, Wido den Hollander <w...@42on.com> wrote: > Is this a HDD or SSD cluster? I assume the latter? Since usually HDDs are > 100% busy during heavy recovery.
HDD with SSD journals. Our experience at this scale, ~900 OSDs over 33 hosts, is that it takes a fair percentage of PGs to be involved before disks are saturated. At this point I'm mainly basing the observations on backfill after reweight, as that is something we have been doing periodically. We have some comparative rados bench and fio data but have not yet graphed it. > Do you also know how much more power these machines started to use? Your > iDRAC might be able to tell you this. No have not really looked into this yet, though the iDRAC certainly can provide an indication. I think these are 45Watt TDP CPUs (E5-2630v2) so would expect it to be less than 2X power draw increase after factoring in 30+ spindles, RAM, etc. > Would you maybe want to write a pull request to get this in to docs.ceph.com? Absolutely, actually I have a small list of outstanding Ceph doc contributions that I am planning to work on during next couple of weeks travel. Do you have a suggestion for an appropriate place in the docs to add this? -- Cheers, ~Blairo _______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
