On Mon, 11 Jun 2012, eric_yh_c...@wiwynn.com wrote: > Dear All: > > I saw rbd support "async writes" since 0.36, http://ceph.com/2011/09/ > But I cannot find related document that how to turn on it. > Should I just write "enabled" to /sys/devices/rbd/0/power/async?
This was referring to the ill-conceived 'rbd writeback window' functionality in librbd (userspace client only, never implemented in the kernel). It was removed because it was a broken and could lead to data loss. The current librbd has a proper cache with performs better and is actually safe to use. It's enabled with 'rbd cache = true', or (with upstream but unreleased qemu) via the qemu cache=writeback or cache=writethrough options. For kernel rbd, I would layer something on top of it if you want caching. Maybe bcache? > One more thing, If I want to implement iSCSI multipath with RBD, just > like http://ceph.com/wiki/ISCSI > Can I turn on the async writes under this situation? No. The caching will probably also be problematic with any uses where more than one client is accessing the device. Even file systems like gfs2 or ocfs2 (I presume) assume read-after-write, even without issuing a flush to the device. That won't work with the rbd client cache. sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html