On Tue, Feb 12, 2013 at 08:16:49AM -0700, Kaspar Schleiser wrote: > Hey Chris, > > On 02/02/2013 05:02 PM, Chris Mason wrote: > > Btrfs -- 604MB/s > > MD -- 162MB/s > > > > > > MD -- 800MB/s very little system time > > Btrfs -- 3.8GB/s one CPU mostly pegged > > > Btrfs -- 380MB/s seen by fio > > MD -- 174MB/s seen by fio > > > Creating 12 million files on Btrfs raid5 took 226 seconds, vs 485 > > seconds on MD. > > Do I read these numbers incorrectly, or does even this first iteration > of btrfs' raid5/6 code run circles around MD?
Yes and no. Most of the differences were on flash, and really it just looks like MD needs tuning for IO latency and concurrency. There are some MD patches for this recently to add more threads for parity calculations, and these solve some throughput problems. But one thing that we've proven with btrfs is that helper threads mean more IO latencies. So the MD code probably needs some short cuts to do the parity inline as well. -chris -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
