Ceph » CephFS

Once we've updated our MemoryModel (#4502, #4503) and have selected groups of in-memory data that we believe we can split out (#4499, #4500, #4501), update the memory model and the cache objects so that we can see projected use following that split. Probably this will be something like add booleans for each grouping in the cache object and set them to "true" when the grouping is in use, or create an atomic for each grouping and increment and decrement it appropriately.
This way we can look at actual running systems and get estimates of which groupings are in use least often (because if a grouping is a lot of memory but is in use by 90% of cached inodes, we can't save much RAM).