Hello, We are building a ~2PiB 200 OSD cluster that will be used entirely by an FS for hosting data storage. I have a question regarding whether there is any specific reason/advantage to use subvolumes vs plain directories for segregating different departments, envs, etc. Currently, the plan is to create subtrees like: /web/home/env1 /web/logs/env1 /web/logs/env2 /mail/home/env1 ..etc Clients will be path-restricted to the bottom directory with MDS caps, and namespace restricted with OSD caps. Namespace layouts will be set on the mounted dir when new users/dirs are provisioned. Snapshots will be managed by storage admins from the top of the trees. Static MDS pinning will be used throughout. As far as I understand, if I create subvolume groups and subvolumes, I would have to flatten the <department>/<service>/<env> structure a bit to make it work, but it may simplify the manual management of snapshots, caps, and namespaces. Are there any other reasons to use subvolumes via manual dir management? Are subvolumes treated any differently by the cluster than dirs? I am especially concerned with io performance, as well as snap mirroring, as we are building an identical cluster to receive mirrors. Appreciate any insight. Thank you. *Jesse Galley* *Director, Mail Platform & Migrations, Hostopia* *C: 416.816.0866* *jesse.galley@xxxxxxxxxxxx <jesse.galley@xxxxxxxxxxxx>* _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |