On Mon, Jun 29, 2009 at 06:55:21PM +0800, Vladislav Bolkhovitin wrote: > > > Wu Fengguang, on 06/29/2009 01:34 PM wrote: > > On Sat, Jun 20, 2009 at 08:29:31PM +0800, Vladislav Bolkhovitin wrote: > >> Wu Fengguang, on 06/20/2009 07:55 AM wrote: > >>> On Fri, Jun 19, 2009 at 03:04:36AM +0800, Andrew Morton wrote: > >>>> On Sun, 7 Jun 2009 06:45:38 +0800 > >>>> Wu Fengguang <fengguang.wu@xxxxxxxxx> wrote: > >>>> > >>>>>>> Do you have a place where the raw blktrace data can be retrieved for > >>>>>>> more in-depth analysis? > >>>>>> I think your comment is really adequate. In another thread, Wu Fengguang pointed > >>>>>> out the same issue. > >>>>>> I and Wu also wait his analysis. > >>>>> And do it with a large readahead size :) > >>>>> > >>>>> Alan, this was my analysis: > >>>>> > >>>>> : Hifumi, can you help retest with some large readahead size? > >>>>> : > >>>>> : Your readahead size (128K) is smaller than your max_sectors_kb (256K), > >>>>> : so two readahead IO requests get merged into one real IO, that means > >>>>> : half of the readahead requests are delayed. > >>>>> > >>>>> ie. two readahead requests get merged and complete together, thus the effective > >>>>> IO size is doubled but at the same time it becomes completely synchronous IO. > >>>>> > >>>>> : > >>>>> : The IO completion size goes down from 512 to 256 sectors: > >>>>> : > >>>>> : before patch: > >>>>> : 8,0 3 177955 50.050313976 0 C R 8724991 + 512 [0] > >>>>> : 8,0 3 177966 50.053380250 0 C R 8725503 + 512 [0] > >>>>> : 8,0 3 177977 50.056970395 0 C R 8726015 + 512 [0] > >>>>> : 8,0 3 177988 50.060326743 0 C R 8726527 + 512 [0] > >>>>> : 8,0 3 177999 50.063922341 0 C R 8727039 + 512 [0] > >>>>> : > >>>>> : after patch: > >>>>> : 8,0 3 257297 50.000760847 0 C R 9480703 + 256 [0] > >>>>> : 8,0 3 257306 50.003034240 0 C R 9480959 + 256 [0] > >>>>> : 8,0 3 257307 50.003076338 0 C R 9481215 + 256 [0] > >>>>> : 8,0 3 257323 50.004774693 0 C R 9481471 + 256 [0] > >>>>> : 8,0 3 257332 50.006865854 0 C R 9481727 + 256 [0] > >>>>> > >>>> I haven't sent readahead-add-blk_run_backing_dev.patch in to Linus yet > >>>> and it's looking like 2.6.32 material, if ever. > >>>> > >>>> If it turns out to be wonderful, we could always ask the -stable > >>>> maintainers to put it in 2.6.x.y I guess. > >>> Agreed. The expected (and interesting) test on a properly configured > >>> HW RAID has not happened yet, hence the theory remains unsupported. > >> Hmm, do you see anything improper in the Ronald's setup (see > >> http://sourceforge.net/mailarchive/forum.php?thread_name=a0272b440906030714g67eabc5k8f847fb1e538cc62%40mail.gmail.com&forum_name=scst-devel)? > >> It is HW RAID based. > > > > No. Ronald's HW RAID performance is reasonably good. I meant Hifumi's > > RAID performance is too bad and may be improved by increasing the > > readahead size, hehe. > > > >> As I already wrote, we can ask Ronald to perform any needed tests. > > > > Thanks! Ronald's test results are: > > > > 231 MB/s HW RAID > > 69.6 MB/s HW RAID + SCST > > 89.7 MB/s HW RAID + SCST + this patch > > > > So this patch seem to help SCST, but again it would be better to > > improve the SCST throughput first - it is now quite sub-optimal. > > No, SCST performance isn't an issue here. You simply can't get more than > 110 MB/s from iSCSI over 1GbE, hence 231 MB/s fundamentally isn't > possible. There is only room for 20% improvement, which should be Ah yes. > achieved with better client-side-driven pipelining (see our other > discussions, e.g. http://lkml.org/lkml/2009/5/12/370) Yeah, that's what I want to figure out why :) Thanks, Fengguang > > (Sorry for the long delay: currently I have not got an idea on > > how to measure such timing issues.) > > > > And if Ronald could provide the HW RAID performance with this patch, > > then we can confirm if this patch really makes a difference for RAID. > > > > Thanks, > > Fengguang -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
