CINXE.COM

<?xml version="1.0" encoding="UTF-8" standalone="yes"?> <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>LKML: Paul Jackson: Re: [patch] sched: auto-tune migration costs [was: Re: Industry db benchmark result on recent 2.6 kernels]</title><link href="/css/message.css" rel="stylesheet" type="text/css" /><link href="/css/wrap.css" rel="alternate stylesheet" type="text/css" title="wrap" /><link href="/css/nowrap.css" rel="stylesheet" type="text/css" title="nowrap" /><link href="/favicon.ico" rel="shortcut icon" /><script src="/js/simple-calendar.js" type="text/javascript"></script><script src="/js/styleswitcher.js" type="text/javascript"></script><link rel="alternate" type="application/rss+xml" title="lkml.org : last 100 messages" href="/rss.php" /><link rel="alternate" type="application/rss+xml" title="lkml.org : last messages by Paul Jackson" href="/groupie.php?aid=5266" /><script> var _paq = window._paq = window._paq || []; /* tracker methods like "setCustomDimension" should be called before "trackPageView" */ _paq.push(["setDoNotTrack", true]); _paq.push(["disableCookies"]); _paq.push(['trackPageView']); _paq.push(['enableLinkTracking']); (function() { var u="//m.lkml.org/"; _paq.push(['setTrackerUrl', u+'matomo.php']); _paq.push(['setSiteId', '1']); var d=document, g=d.createElement('script'), s=d.getElementsByTagName('script')[0]; g.async=true; g.src=u+'matomo.js'; s.parentNode.insertBefore(g,s); })(); </script></head><body onload="es.jasper.simpleCalendar.init();" itemscope="itemscope" itemtype="http://schema.org/BlogPosting"><table border="0" cellpadding="0" cellspacing="0"><tr><td width="180" align="center"><a href="/"><img style="border:0;width:135px;height:32px" src="/images/toprowlk.gif" alt="lkml.org" /></a></td><td width="32">聽</td><td class="nb"><div><a class="nb" href="/lkml"> [lkml]</a> 聽 <a class="nb" href="/lkml/2005"> [2005]</a> 聽 <a class="nb" href="/lkml/2005/4"> [Apr]</a> 聽 <a class="nb" href="/lkml/2005/4/3"> [3]</a> 聽 <a class="nb" href="/lkml/last100"> [last100]</a> 聽 <a href="/rss.php"><img src="/images/rss-or.gif" border="0" alt="RSS Feed" /></a></div><div>Views: <a href="#" class="nowrap" onclick="setActiveStyleSheet('wrap');return false;">[wrap]</a><a href="#" class="wrap" onclick="setActiveStyleSheet('nowrap');return false;">[no wrap]</a> 聽 <a class="nb" href="/lkml/mheaders/2005/4/3/123" onclick="this.href='/lkml/headers'+'/2005/4/3/123';">[headers]</a>聽 <a href="/lkml/bounce/2005/4/3/123">[forward]</a>聽 </div></td><td width="32">聽</td></tr><tr><td valign="top"><div class="es-jasper-simpleCalendar" baseurl="/lkml/"></div><div class="threadlist">Messages in this thread</div><ul class="threadlist"><li class="root"><a href="/lkml/2005/4/1/310">First message in thread</a></li><li><a href="/lkml/2005/4/3/81">Paul Jackson</a><ul><li><a href="/lkml/2005/4/3/86">Ingo Molnar</a><ul><li class="origin"><a href="/lkml/2005/4/5/25">Paul Jackson</a><ul><li><a href="/lkml/2005/4/5/25">Andi Kleen</a><ul><li><a href="/lkml/2005/4/5/37">Paul Jackson</a></li></ul></li></ul></li></ul></li><li><a href="/lkml/2005/4/3/89">Ingo Molnar</a><ul><li><a href="/lkml/2005/4/3/133">Paul Jackson</a><ul><li><a href="/lkml/2005/4/3/156">Nick Piggin</a><ul><li><a href="/lkml/2005/4/3/171">Paul Jackson</a></li></ul></li><li><a href="/lkml/2005/4/4/30">Ingo Molnar</a><ul><li><a href="/lkml/2005/4/4/34">Paul Jackson</a></li></ul></li></ul></li></ul></li></ul></li></ul></td><td width="32" rowspan="2" class="c" valign="top"><img src="/images/icornerl.gif" width="32" height="32" alt="/" /></td><td class="c" rowspan="2" valign="top" style="padding-top: 1em"><table><tr><td><table><tr><td class="lp">Date</td><td class="rp" itemprop="datePublished">Sun, 3 Apr 2005 15:30:56 -0700</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Paul Jackson <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: [patch] sched: auto-tune migration costs [was: Re: Industry db benchmark result on recent 2.6 kernels]</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">Ingo wrote:<br />> if_ there is a significant hierarchy between CPUs it<br />> should be represented via a matching sched-domains hierarchy,<br /><br />Agreed.<br /><br />I'll see how the sched domains hierarchy looks on a bigger SN2 systems.<br /><br />If the CPU hierarchy is not reflected in the sched-domain hierarchy any<br />better there, then I will look to involve the "SN2 sched domain<br />hierarchy experts" in improving SN2 the sched-domain hierarchy.<br /><br />Ok - that works. Your patch of yesterday provides just the tool<br />I need to measure this. Cool.<br /><br />> i'll first try the bottom-up approach to speed up detection (getting to<br />> the hump is very fast most of the time).<br /><br />Good.<br /><br />> then we can let the arch override the cpu_distance() method<br /><br />I'm not aware we need that, yet anyway. First I should see if<br />the SN2 sched_domains need improving. Take a shot at doing it<br />'the right way' before we go inventing overrides. I suspect<br />you agree.<br /><br />> the migration cost matrix we can later use to tune all the other <br />> sched-domains balancing related tunables as well<br /><br />That comes close to my actual motivation here. I hope to expose a<br />"cpu_distance" such as based on this cost matrix, to userland.<br /><br />We already expose the SLIT table node distances (using SN2 specific<br />/proc files today, others are working on an arch-neutral mechanism).<br /><br />As we push more cores and hyperthreads into a single package on one end,<br />and more complex numa topologies on the other end, this becomes<br />increasingly interesting to NUMA aware user software.<br /><br />-- <br /> I won't rest till it's the best ...<br /> Programmer, Linux Scalability<br /> Paul Jackson <pj@engr.sgi.com> 1.650.933.1373, 1.925.600.0401<br />-<br />To unsubscribe from this list: send the line "unsubscribe linux-kernel" in<br />the body of a message to majordomo@vger.kernel.org<br />More majordomo info at <a href="http://vger.kernel.org/majordomo-info.html">http://vger.kernel.org/majordomo-info.html</a><br />Please read the FAQ at <a href="http://www.tux.org/lkml/">http://www.tux.org/lkml/</a><br /><br /></pre></td><td width="32" rowspan="2" class="c" valign="top"><img src="/images/icornerr.gif" width="32" height="32" alt="\" /></td></tr><tr><td align="right" valign="bottom"> 聽 </td></tr><tr><td align="right" valign="bottom">聽</td><td class="c" valign="bottom" style="padding-bottom: 0px"><img src="/images/bcornerl.gif" width="32" height="32" alt="\" /></td><td class="c">聽</td><td class="c" valign="bottom" style="padding-bottom: 0px"><img src="/images/bcornerr.gif" width="32" height="32" alt="/" /></td></tr><tr><td align="right" valign="top" colspan="2"> 聽 </td><td class="lm">Last update: 2005-04-06 13:31 聽聽 [from the cache]<br />漏2003-2020 <a href="http://blog.jasper.es/"><span itemprop="editor">Jasper Spaans</span></a>|hosted at <a href="https://www.digitalocean.com/?refcode=9a8e99d24cf9">Digital Ocean</a> and my Meterkast|<a href="http://blog.jasper.es/categories.html#lkml-ref">Read the blog</a></td><td>聽</td></tr></table><script language="javascript" src="/js/styleswitcher.js" type="text/javascript"></script></body></html>