Hi, I seem to have a bad edge effect in my setup, don't know if it's a RBD problem or a Xen problem. So, I have one Ceph cluster, in which I setup 2 different storage pools : one on SSD and one on SAS. With appropriate CRUSH rules, those pools are complety separated, only MON are commons. Then, on a Xen host A, I run "VMSSD" and "VMSAS". If I launch a big reballance on the "SSD pool", then the "VMSSD" *and* "VMSAS" will slow down (a lot of IOWait). But if I move the "VMSAS" on a different Xen host (B), then "VMSSD" will still be slow, but the "VMSAS" will be fast again. The first thing I checked is the network of the Xen host A, but I didn't find any problem. So, is there a queue, shared by all RBD kernel clients running on a same host ? Or something which can explain this edge effect ? Olivier PS : one precision, I have about 60 RBD mapped on the "Xen host A", don't know if it can be the key of the problem. _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |