From mboxrd@z Thu Jan 1 00:00:00 1970 From: Chris Mason Subject: Re: experimental branch rebased Date: Fri, 27 Feb 2009 11:25:32 -0500 Message-ID: <1235751932.12383.11.camel@think.oraclecorp.com> References: <1235663145.4631.11.camel@think.oraclecorp.com> <49A71D95.8030707@austin.ibm.com> Mime-Version: 1.0 Content-Type: text/plain Cc: linux-btrfs To: Steven Pratt Return-path: In-Reply-To: <49A71D95.8030707@austin.ibm.com> List-ID: On Thu, 2009-02-26 at 16:54 -0600, Steven Pratt wrote: > Chris Mason wrote: > > Hello everyone, > > > > I've rebased the experimental branch again with my latest performance > > fixes. > > > > I took out the delayed unlink code, it wasn't making a big enough > > difference in any benchmarks to justify the complexity. > > > > I changed the delayed backref code to do delayed processing for all > > extents. In general it is much faster and uses less stack. > > > > I pulled Josef's enospc work into the master branch and asked Linus to > > pull it. > > > > I'm going to hammer on the experimental branch for a few days and ask > > Steve to give it another run. > > > Done. Results for RAID are updated in history tree > http://btrfs.boxacle.net/repository/raid/history/History.html > > This gives back a few of the performance improvements made on the tree > from the 24th (mail server and random write). > Thanks for doing this. I'm a little confused by the output though, somehow our configs are giving opposite results ;) When I run the rand-write workload with 128 threads here, btrfs gets 6236 ops/sec, and ext4 only gets 1509 ops/sec. My box only has 5 drives, so there is probably a difference in btrfs' ability to keep all the drives in the array busy. So, I'll do some more experiments here. -chris