> On Jan 13, 2016, at 9:45 PM, Sage Weil <sweil@xxxxxxxxxx> wrote: > > On Wed, 13 Jan 2016, Zhang Huan wrote: >> Hello Sage, >> >> I have been watching the development of bluestore for a while. I am >> interested in bluestore because I believe putting object store on top of >> raw device is the right way. I would like to know if you have any >> release schedule, and which version it will be merged into? I also >> notice that there is big locks (Collection->lock and >> StupidAllocator->lock, etc) in the write path, this might be an issue >> for SSD. Do you have any plan to break it into finer locks? > > It's in master now and it will be part of the jewel release. It won't > be the default choice yet, since it is still very new and unproven, but so > far performance is looking better than FileStore on HDDs. I know it is early, but do we have any chance to know your test result? and how about the result on SSD? > > The collection->lock doesn't need to be broken up since it's hidden > beneath the OSD's per-pg lock anyway. Thanks, I found the per-pg lock. If there exist per-pg locks, the per-collection locks are actually not needed? > The allocator lock could be > sharded, though, by creating multiple allocation groups (ala XFS). This > is partly why the Allocator interface is abstract and the current > implementation is called StupidAllocator :). If you can demonstrate > that the lock is affecting performance we should definitely improve that! It would be an interesting to look into:-) > > sage > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
