From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755404AbZGFOhR (ORCPT ); Mon, 6 Jul 2009 10:37:17 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1754775AbZGFOhF (ORCPT ); Mon, 6 Jul 2009 10:37:05 -0400 Received: from fg-out-1718.google.com ([72.14.220.158]:36121 "EHLO fg-out-1718.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754051AbZGFOhC convert rfc822-to-8bit (ORCPT ); Mon, 6 Jul 2009 10:37:02 -0400 DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type:content-transfer-encoding; b=JHIbv1ozsXK3Ua/fIc6cmSu70Hb0aW3R+L+wdGXb2a9VxQ2WvpgCNutTl9S7d8R3iN NUoCvILbQO5JfLVPsX63qz/aSttWZhTWoxqByH1OBpzJC/oJg1DPnlxeu5viztNcKdGp OJ2yZ+6Y1LkU2LQgNmJFZdFyMhkrDVeal+A/8= MIME-Version: 1.0 In-Reply-To: <4A51DC0A.10302@vlnb.net> References: <4A3CD62B.1020407@vlnb.net> <20090629142124.GA28945@localhost> <20090629150109.GA3534@localhost> <4A48DFC5.3090205@vlnb.net> <20090630010414.GB31418@localhost> <4A49EEF9.6010205@vlnb.net> <4A4DE3C1.5080307@vlnb.net> <4A51DC0A.10302@vlnb.net> Date: Mon, 6 Jul 2009 16:37:04 +0200 Message-ID: Subject: Re: [RESEND] [PATCH] readahead:add blk_run_backing_dev From: Ronald Moesbergen To: Vladislav Bolkhovitin Cc: Wu Fengguang , linux-kernel@vger.kernel.org, akpm@linux-foundation.org, kosaki.motohiro@jp.fujitsu.com, Alan.Brunelle@hp.com, hifumi.hisashi@oss.ntt.co.jp, linux-fsdevel@vger.kernel.org, jens.axboe@oracle.com, randy.dunlap@oracle.com, Bart Van Assche Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 8BIT Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org 2009/7/6 Vladislav Bolkhovitin : > (Restored the original list of recipients in this thread as I was asked.) > > Hi Ronald, > > Ronald Moesbergen, on 07/04/2009 07:19 PM wrote: >> >> 2009/7/3 Vladislav Bolkhovitin : >>> >>> Ronald Moesbergen, on 07/03/2009 01:14 PM wrote: >>>>>> >>>>>> OK, now I tend to agree on decreasing max_sectors_kb and increasing >>>>>> read_ahead_kb. But before actually trying to push that idea I'd like >>>>>> to >>>>>> - do more benchmarks >>>>>> - figure out why context readahead didn't help SCST performance >>>>>>  (previous traces show that context readahead is submitting perfect >>>>>>  large io requests, so I wonder if it's some io scheduler bug) >>>>> >>>>> Because, as we found out, without your >>>>> http://lkml.org/lkml/2009/5/21/319 >>>>> patch read-ahead was nearly disabled, hence there were no difference >>>>> which >>>>> algorithm was used? >>>>> >>>>> Ronald, can you run the following tests, please? This time with 2 >>>>> hosts, >>>>> initiator (client) and target (server) connected using 1 Gbps iSCSI. It >>>>> would be the best if on the client vanilla 2.6.29 will be ran, but any >>>>> other >>>>> kernel will be fine as well, only specify which. Blockdev-perftest >>>>> should >>>>> be >>>>> ran as before in buffered mode, i.e. with "-a" switch. >>>>> >>>>> 1. All defaults on the client, on the server vanilla 2.6.29 with >>>>> Fengguang's >>>>> http://lkml.org/lkml/2009/5/21/319 patch with all default settings. >>>>> >>>>> 2. All defaults on the client, on the server vanilla 2.6.29 with >>>>> Fengguang's >>>>> http://lkml.org/lkml/2009/5/21/319 patch with default RA size and 64KB >>>>> max_sectors_kb. >>>>> >>>>> 3. All defaults on the client, on the server vanilla 2.6.29 with >>>>> Fengguang's >>>>> http://lkml.org/lkml/2009/5/21/319 patch with 2MB RA size and default >>>>> max_sectors_kb. >>>>> >>>>> 4. All defaults on the client, on the server vanilla 2.6.29 with >>>>> Fengguang's >>>>> http://lkml.org/lkml/2009/5/21/319 patch with 2MB RA size and 64KB >>>>> max_sectors_kb. >>>>> >>>>> 5. All defaults on the client, on the server vanilla 2.6.29 with >>>>> Fengguang's >>>>> http://lkml.org/lkml/2009/5/21/319 patch and with context RA patch. RA >>>>> size >>>>> and max_sectors_kb are default. For your convenience I committed the >>>>> backported context RA patches into the SCST SVN repository. >>>>> >>>>> 6. All defaults on the client, on the server vanilla 2.6.29 with >>>>> Fengguang's >>>>> http://lkml.org/lkml/2009/5/21/319 and context RA patches with default >>>>> RA >>>>> size and 64KB max_sectors_kb. >>>>> >>>>> 7. All defaults on the client, on the server vanilla 2.6.29 with >>>>> Fengguang's >>>>> http://lkml.org/lkml/2009/5/21/319 and context RA patches with 2MB RA >>>>> size >>>>> and default max_sectors_kb. >>>>> >>>>> 8. All defaults on the client, on the server vanilla 2.6.29 with >>>>> Fengguang's >>>>> http://lkml.org/lkml/2009/5/21/319 and context RA patches with 2MB RA >>>>> size >>>>> and 64KB max_sectors_kb. >>>>> >>>>> 9. On the client default RA size and 64KB max_sectors_kb. On the server >>>>> vanilla 2.6.29 with Fengguang's http://lkml.org/lkml/2009/5/21/319 and >>>>> context RA patches with 2MB RA size and 64KB max_sectors_kb. >>>>> >>>>> 10. On the client 2MB RA size and default max_sectors_kb. On the server >>>>> vanilla 2.6.29 with Fengguang's http://lkml.org/lkml/2009/5/21/319 and >>>>> context RA patches with 2MB RA size and 64KB max_sectors_kb. >>>>> >>>>> 11. On the client 2MB RA size and 64KB max_sectors_kb. On the server >>>>> vanilla >>>>> 2.6.29 with Fengguang's http://lkml.org/lkml/2009/5/21/319 and context >>>>> RA >>>>> patches with 2MB RA size and 64KB max_sectors_kb. >>>> >>>> Ok, done. Performance is pretty bad overall :( >>>> >>>> The kernels I used: >>>> client kernel: 2.6.26-15lenny3 (debian) >>>> server kernel: 2.6.29.5 with blk_dev_run patch >>>> >>>> And I adjusted the blockdev-perftest script to drop caches on both the >>>> server (via ssh) and the client. >>>> >>>> The results: >>>> >> >> ... previous results ... >> >>> Those are on the server without io_context-2.6.29 and readahead-2.6.29 >>> patches applied and with CFQ scheduler, correct? >>> >>> Then we see how reorder of requests caused by many I/O threads submitting >>> I/O in separate I/O contexts badly affect performance and no RA, >>> especially >>> with default 128KB RA size, can solve it. Less max_sectors_kb on the >>> client >>> => more requests it sends at once => more reorder on the server => worse >>> throughput. Although, Fengguang, in theory, context RA with 2MB RA size >>> should considerably help it, no? >>> >>> Ronald, can you perform those tests again with both io_context-2.6.29 and >>> readahead-2.6.29 patches applied on the server, please? >> >> Hi Vlad, >> >> I have retested with the patches you requested (and got access to the >> systems today :) ) The results are better, but still not great. >> >> client kernel: 2.6.26-15lenny3 (debian) >> server kernel: 2.6.29.5 with io_context and readahead patch >> >> 5) client: default, server: default >> blocksize       R        R        R   R(avg,    R(std        R >>  (bytes)     (s)      (s)      (s)    MB/s)   ,MB/s)   (IOPS) >>  67108864  18.303   19.867   18.481   54.299    1.961    0.848 >>  33554432  18.321   17.681   18.708   56.181    1.314    1.756 >>  16777216  17.816   17.406   19.257   56.494    2.410    3.531 >>  8388608  18.077   17.727   19.338   55.789    2.056    6.974 >>  4194304  17.918   16.601   18.287   58.276    2.454   14.569 >>  2097152  17.426   17.334   17.610   58.661    0.384   29.331 >>  1048576  19.358   18.764   17.253   55.607    2.734   55.607 >>   524288  17.951   18.163   17.440   57.379    0.983  114.757 >>   262144  18.196   17.724   17.520   57.499    0.907  229.995 >>   131072  18.342   18.259   17.551   56.751    1.131  454.010 >>    65536  17.733   18.572   17.134   57.548    1.893  920.766 >>    32768  19.081   19.321   17.364   55.213    2.673 1766.818 >>    16384  17.181   18.729   17.731   57.343    2.033 3669.932 >> >> 6) client: default, server: 64 max_sectors_kb, RA default >> blocksize       R        R        R   R(avg,    R(std        R >>  (bytes)     (s)      (s)      (s)    MB/s)   ,MB/s)   (IOPS) >>  67108864  21.790   20.062   19.534   50.153    2.304    0.784 >>  33554432  20.212   19.744   19.564   51.623    0.706    1.613 >>  16777216  20.404   19.329   19.738   51.680    1.148    3.230 >>  8388608  20.170   20.772   19.509   50.852    1.304    6.356 >>  4194304  19.334   18.742   18.522   54.296    0.978   13.574 >>  2097152  19.413   18.858   18.884   53.758    0.715   26.879 >>  1048576  20.472   18.755   18.476   53.347    2.377   53.347 >>   524288  19.120   20.104   18.404   53.378    1.925  106.756 >>   262144  20.337   19.213   18.636   52.866    1.901  211.464 >>   131072  19.199   18.312   19.970   53.510    1.900  428.083 >>    65536  19.855   20.114   19.592   51.584    0.555  825.342 >>    32768  20.586   18.724   20.340   51.592    2.204 1650.941 >>    16384  21.119   19.834   19.594   50.792    1.651 3250.669 >> >> 7) client: default, server: default max_sectors_kb, RA 2MB >> blocksize       R        R        R   R(avg,    R(std        R >>  (bytes)     (s)      (s)      (s)    MB/s)   ,MB/s)   (IOPS) >>  67108864  17.767   16.489   16.949   60.050    1.842    0.938 >>  33554432  16.777   17.034   17.102   60.341    0.500    1.886 >>  16777216  18.509   16.784   16.971   58.891    2.537    3.681 >>  8388608  18.058   17.949   17.599   57.313    0.632    7.164 >>  4194304  18.286   17.648   17.026   58.055    1.692   14.514 >>  2097152  17.387   18.451   17.875   57.226    1.388   28.613 >>  1048576  18.270   17.698   17.570   57.397    0.969   57.397 >>   524288  16.708   17.900   17.233   59.306    1.668  118.611 >>   262144  18.041   17.381   18.035   57.484    1.011  229.934 >>   131072  17.994   17.777   18.146   56.981    0.481  455.844 >>    65536  17.097   18.597   17.737   57.563    1.975  921.011 >>    32768  17.167   17.035   19.693   57.254    3.721 1832.127 >>    16384  17.144   16.664   17.623   59.762    1.367 3824.774 >> >> 8) client: default, server: 64 max_sectors_kb, RA 2MB >> blocksize       R        R        R   R(avg,    R(std        R >>  (bytes)     (s)      (s)      (s)    MB/s)   ,MB/s)   (IOPS) >>  67108864  20.003   21.133   19.308   50.894    1.881    0.795 >>  33554432  19.448   20.015   18.908   52.657    1.222    1.646 >>  16777216  19.964   19.350   19.106   52.603    0.967    3.288 >>  8388608  18.961   19.213   19.318   53.437    0.419    6.680 >>  4194304  18.135   19.508   19.361   53.948    1.788   13.487 >>  2097152  18.753   19.471   18.367   54.315    1.306   27.158 >>  1048576  19.189   18.586   18.867   54.244    0.707   54.244 >>   524288  18.985   19.199   18.840   53.874    0.417  107.749 >>   262144  19.064   21.143   19.674   51.398    2.204  205.592 >>   131072  18.691   18.664   19.116   54.406    0.594  435.245 >>    65536  18.468   20.673   18.554   53.389    2.729  854.229 >>    32768  20.401   21.156   19.552   50.323    1.623 1610.331 >>    16384  19.532   20.028   20.466   51.196    0.977 3276.567 >> >> 9) client: 64 max_sectors_kb, default RA. server: 64 max_sectors_kb, RA >> 2MB >> blocksize       R        R        R   R(avg,    R(std        R >>  (bytes)     (s)      (s)      (s)    MB/s)   ,MB/s)   (IOPS) >>  67108864  16.458   16.649   17.346   60.919    1.364    0.952 >>  33554432  16.479   16.744   17.069   61.096    0.878    1.909 >>  16777216  17.128   16.585   17.112   60.456    0.910    3.778 >>  8388608  17.322   16.780   16.885   60.262    0.824    7.533 >>  4194304  17.530   16.725   16.756   60.250    1.299   15.063 >>  2097152  16.580   17.875   16.619   60.221    2.076   30.110 >>  1048576  17.550   17.406   17.075   59.049    0.681   59.049 >>   524288  16.492   18.211   16.832   59.718    2.519  119.436 >>   262144  17.241   17.115   17.365   59.397    0.352  237.588 >>   131072  17.430   16.902   17.511   59.271    0.936  474.167 >>    65536  16.726   16.894   17.246   60.404    0.768  966.461 >>    32768  16.662   17.517   17.052   59.989    1.224 1919.658 >>    16384  17.429   16.793   16.753   60.285    1.085 3858.268 >> >> 10) client: default max_sectors_kb, 2MB RA. server: 64 max_sectors_kb, RA >> 2MB >> blocksize       R        R        R   R(avg,    R(std        R >>  (bytes)     (s)      (s)      (s)    MB/s)   ,MB/s)   (IOPS) >>  67108864  17.601   18.334   17.379   57.650    1.307    0.901 >>  33554432  18.281   18.128   17.169   57.381    1.610    1.793 >>  16777216  17.660   17.875   17.356   58.091    0.703    3.631 >>  8388608  17.724   17.810   18.383   56.992    0.918    7.124 >>  4194304  17.475   17.770   19.003   56.704    2.031   14.176 >>  2097152  17.287   17.674   18.492   57.516    1.604   28.758 >>  1048576  17.972   17.460   18.777   56.721    1.689   56.721 >>   524288  18.680   18.952   19.445   53.837    0.890  107.673 >>   262144  18.070   18.337   18.639   55.817    0.707  223.270 >>   131072  16.990   16.651   16.862   60.832    0.507  486.657 >>    65536  17.707   16.972   17.520   58.870    1.066  941.924 >>    32768  17.767   17.208   17.205   58.887    0.885 1884.399 >>    16384  18.258   17.252   18.035   57.407    1.407 3674.059 >> >> 11) client: 64 max_sectors_kb, 2MB. RA server: 64 max_sectors_kb, RA 2MB >> blocksize       R        R        R   R(avg,    R(std        R >>  (bytes)     (s)      (s)      (s)    MB/s)   ,MB/s)   (IOPS) >>  67108864  17.993   18.307   18.718   55.850    0.902    0.873 >>  33554432  19.554   18.485   17.902   54.988    1.993    1.718 >>  16777216  18.829   18.236   18.748   55.052    0.785    3.441 >>  8388608  21.152   19.065   18.738   52.257    2.745    6.532 >>  4194304  19.131   19.703   17.850   54.288    2.268   13.572 >>  2097152  19.093   19.152   19.509   53.196    0.504   26.598 >>  1048576  19.371   18.775   18.804   53.953    0.772   53.953 >>   524288  20.003   17.911   18.602   54.470    2.476  108.940 >>   262144  19.182   19.460   18.476   53.809    1.183  215.236 >>   131072  19.403   19.192   18.907   53.429    0.567  427.435 >>    65536  19.502   19.656   18.599   53.219    1.309  851.509 >>    32768  18.746   18.747   18.250   55.119    0.701 1763.817 >>    16384  20.977   19.437   18.840   51.951    2.319 3324.862 > > The results look inconsistently with what you had previously (89.7 MB/s). > How can you explain it? I had more patches applied with that test: (scst_exec_req_fifo-2.6.29, put_page_callback-2.6.29) and I used a different dd command: dd if=/dev/sdc of=/dev/zero bs=512K count=2000 But all that said, I can't reproduce speeds that high now. Must have made a mistake back then (maybe I forgot to clear the pagecache). > I think, most likely, there was some confusion between the tested and > patched versions of the kernel or you forgot to apply the io_context patch. > Please recheck. The tests above were definitely done right, I just rechecked the patches, and I do see an average increase of about 10MB/s over an unpatched kernel. But overall the performance is still pretty bad. Ronald.