On Mon, Jun 21, 09:02, Sage Weil wrote: > I would probably allow for 10% utilization variance for a smallish cluster > and maybe another 10% for the rest to be safe. Something along the lines > of total_disk * .8 / replication_level? You generally shouldn't fill any > file system beyond 80% or 90% anyway and expect it to perform well. Sure, but the thing is users couldn't care less about file system usage. So at least in a scenario where users write temporary data from cluster jobs to a world-writable ceph fs, the fs _will_ fill up. So "you should not fill it beyond 80% anyway" is like asking for a pony: You won't get it. It is important that filling up the fs completely does not cause any problems for independent read operations. Otherwise the fs is next to useless, at least for this /tmp-like scenario. Andre -- The only person who always got his work done by Friday was Robinson Crusoe
Attachment:
signature.asc
Description: Digital signature
