On Mon, Aug 3, 2015 at 2:39 PM, Kenneth Waegeman <kenneth.waegeman@xxxxxxxx> wrote: > Another question: When using cephfs with it, we have to use a cache pool on > top of it. But this forms a huge bottleneck for read-only operations. Is it > possible to use the ec pool directly and bypass the cache for read-only > mounts/ scenarios ? File layouts that would work directly with EC pools (e.g. an append-only file layout for archival use cases) have been discussed, but not actively worked on. Implementing new layouts is expensive because we have both the userspace client and the kernel client to consider, and the use cases that would benefit can be a bit niche. As for bypassing the cache for reads: the OSDs already know how to do that, with so-called "proxy reads" where the cache tier OSD reads through instead of promoting an object (it's a mode you can set on your cache tier). Adding support to cephfs to enable per-layout use of IO hints (e.g. "proxy reads on this directory") is mentioned here: https://wiki.ceph.com/@api/deki/files/71/Ceph_Io_Hints_Improvements_r1.pdf John _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |