On Thu, Feb 14, 2013 at 11:56:34AM +0100, Kevin Wolf wrote: > Am 14.02.2013 11:43, schrieb Stefan Hajnoczi: > > On Wed, Feb 13, 2013 at 02:21:50PM +0100, Kevin Wolf wrote: > >> iozone results with and without this series show significant difference > >> for allocating writes: > >> > >> random random > >> KB reclen write rewrite read reread read write > >> > >> base 65536 8 1727 1945 12546 12539 2449 1836 > >> patch 65536 8 1934 1949 12263 12521 2463 1796 > >> > >> base 1048576 256 22344 38437 105823 106135 37743 32167 > >> patch 1048576 256 35989 38542 105231 105994 38301 33036 > > > > I imagine the benefit is even greater for 4 KB or 8 KB blocksize? > > The upper two lines are 8k block size. I suspect that the benefit would > be the greatest for misaligned 64k writes, then you have one overlapping > allocation for each issued request.
I see. I haven't looked at IOzone output for too long, forgot how to interpret it. Stefan