On 12/7/19 1:42 PM, Philippe D'Anjou wrote: > @Wido Den Hollander > > That doesn't explain why its between 76 and 92 PGs, that's major not equal. The balancer will balance the PGs so that all OSDs have an almost equal data usage. It doesn't balance that all OSDs have an equal amount of PGs. The end goal is to make sure all OSDs are filled equally. > Raising PGs to 100 is an old statement anyway, anything 60+ should be > fine. Not an excuse for distribution failure in this case. > I am expecting more or less equal PGs/OSD That will not happen. Objects are distributed over PGs based on the hash of their name and not their size. The distribution of objects over PGs is almost perfect. However, objects will vary in size. If you are low on the amount of PGs you'll have PGs which are bigger then others. This causes data distribution to be off. Therefor it can help to increase the amount of PGs as that will result in a better distribution. PGs will be less varied in size and makes them easier to balance. In addition you gain more parallel performance and suffer less from PG contention. As disks (or SSDs) become bigger and bigger you can benefit by having more PGs. Hope this explains it :-) Wido _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |