Hi,
Currently we change OSD weights so that it also affects weights of
higher elements in crush hierarchy. I'm not sure if doing simple OSD
reweight would work here, what I understood from the original CRUSH
paper, it would not propagate up so 'chose firstn' won't take such
reweights into consideration.
Also I'm sure I've seen pretty extreme weight adjustments, definitely
not proportional to the usage ratio before adjustment.
Bartek
On 11/22/2016 01:51 PM, Dan Van Der Ster wrote:
Hi,
Indeed, we reweight OSDs to balance them, just not very effectively on this particular cluster. But I'm curious if reweighting alone can fix this: if all of a host's OSDs are reweighted by, say 0.5, does that result in other hosts being selected? Or do we need to change the crush weights themselves to fix this kind of imbalance?
-- Dan
On 22 Nov 2016, at 13:44, Bartłomiej Święcki <bartlomiej.swiecki@xxxxxxxxxxxx> wrote:
Hi,
We've observed very similar problems in our clusters, it requires a lot of careful reweight to keep OSDs more or less at the same usage level.
Because of that issue, we're currently trying to keep Racks as regular as possible. Hope the patch you mentioned will address this too.
Regards,
Bartek
On 11/22/2016 01:33 PM, Dan Van Der Ster wrote:
Hi,
I have a couple questions about http://tracker.ceph.com/issues/15653
In the ticket Sage discusses small/big drives, and the small drives get more data than expected.
But we observe this at the rack level: our cluster has four racks, with 7, 8, 8, 4 hosts respectively. The rack with 4 hosts is ~35% more full than the others.
So AFAICT, because of #15653, CRUSH does not currently work well if you try to build a pool which is replicated rack/host-wise when your rack/hosts are not all ~identical in size.
Are others noticing this pattern?
Or are we unusual in that our clusters are not flat/uniform in structure?
Cheers, Dan
_______________________________________________
Ceph-large mailing list
Ceph-large@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-large-ceph.com
_______________________________________________
Ceph-large mailing list
Ceph-large@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-large-ceph.com
