Hello Martin, The proper way is to perform the following process: For all Pools utilizing the same bucket of OSDs: (Pool1_pg_num * Pool1_size) + (Pool2_pg_num * Pool2_size) + ... (Pool(n)_pg_num * Pool(n)_size) -------------------------------------------------------------------------------------------------------------------------------------- OSD count
This value should be between 100 and 200 PGs and is the actual ratio of PGs per OSD in that bucket of OSDs. For the actual recommendation from Ceph Devs (and written by myself), please see: http://ceph.com/pgcalc/ NOTE: The tool is partially broken, but the explanation at the top/bottom is sound. I'll work to get the tool fully functional again. Thanks, Michael J. Kidd Sr. Software Maintenance Engineer Red Hat Ceph Storage +1 919-442-8878 On Tue, Mar 15, 2016 at 11:41 AM, Martin Palma <mar...@palma.bz> wrote: > Hi all, > > The documentation [0] gives us the following formula for calculating > the number of PG if the cluster is bigger than 50 OSDs: > > (OSDs * 100) > Total PGs = ------------ > pool size > > When we have mixed storage server (HDD disks and SSD disks) and we > have defined different roots in our crush map to map some pools only > to HDD disk and some to SSD disks like described by Sebastien Han [1]. > > In the above formula what number of OSDs should be use to calculate > the PGs for a pool only on the HDD disks? The total number of OSDs in > a cluster or only the number of OSDs which have an HDD disk as > backend? > > Best, > Martin > > > [0] > http://docs.ceph.com/docs/master/rados/operations/placement-groups/#choosing-the-number-of-placement-groups > [1] > http://www.sebastien-han.fr/blog/2014/08/25/ceph-mix-sata-and-ssd-within-the-same-box/ > _______________________________________________ > ceph-users mailing list > ceph-users@lists.ceph.com > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com >
_______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
