Yes, you are experiencing issues with I/O because of a slow peering. You may put monstores behind a faster storage if they are served from rotating disks right now or greatly decrease number of placement groups. if it is possible - with 100 OSDs I would try something like 4096 and 8192, though it may impact data placement flatness. I`ve seen a couple of off-list reports where slow peering on a large number of placement group caused persistent problems, for example if user added new OSDs in the middle of slow-going peering process, it stood still forever. If none of those suggestions helps, please feel free to report this problem to a bugtracker, possibly it would give a bump bump a very nice blueprint initiative for reducing overall peering time (https://wiki.ceph.com/Planning/Blueprints/Infernalis/osd%3A_Faster_Peering).
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
