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: Ingo Molnar: 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 Ingo Molnar" href="/groupie.php?aid=18" /><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/2005/3"> [Mar]</a> 聽 <a class="nb" href="/lkml/2005/3/31"> [31]</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/2005/3/31/349" onclick="this.href='/lkml/headers'+'/2005/3/31/349';">[headers]</a>聽 <a href="/lkml/bounce/2005/3/31/349">[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/2005/3/28/161">First message in thread</a></li><li><a href="/lkml/2005/3/31/235">Linus Torvalds</a><ul><li><a href="/lkml/2005/3/31/266">"Chen, Kenneth W"</a><ul><li><a href="/lkml/2005/3/31/295">Nick Piggin</a><ul><li><a href="/lkml/2005/4/1/16">Paul Jackson</a><ul><li><a href="/lkml/2005/4/1/20">Nick Piggin</a></li><li><a href="/lkml/2005/4/1/24">Ingo Molnar</a></li><li><a href="/lkml/2005/4/1/28">Ingo Molnar</a></li></ul></li></ul></li><li class="origin"><a href="/lkml/2005/4/1/4">Ingo Molnar</a><ul><li><a href="/lkml/2005/4/1/4">"Chen, Kenneth W"</a></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">Fri, 1 Apr 2005 06:52:19 +0200</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Ingo Molnar <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: Industry db benchmark result on recent 2.6 kernels</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody"><br />* Chen, Kenneth W <kenneth.w.chen@intel.com> wrote:<br /><br />> The low point in 2.6.11 could very well be the change in the <br />> scheduler. It does too many load balancing in the wake up path and <br />> possibly made a lot of unwise decision. For example, in <br />> try_to_wake_up(), it will try SD_WAKE_AFFINE for task that is not hot. <br />> By not hot, it looks at when it was last ran and compare to a constant <br />> sd->cache_hot_time. The problem is this cache_hot_time is fixed for <br />> the entire universe, whether it is a little celeron processor with <br />> 128KB of cache or a sever class Itanium2 processor with 9MB L3 cache. <br />> This one size fit all isn't really working at all.<br /><br />the current scheduler queue in -mm has some experimental bits as well <br />which will reduce the amount of balancing. But we cannot just merge them <br />an bloc right now, there's been too much back and forth in recent <br />kernels. The safe-to-merge-for-2.6.12 bits are already in -BK.<br /><br />> We had experimented that parameter earlier and found it was one of the <br />> major source of low point in 2.6.8. I debated the issue on LKML about <br />> 4 month ago and finally everyone agreed to make that parameter a boot <br />> time param. The change made into bk tree for 2.6.9 release, but <br />> somehow it got ripped right out 2 days after it went in. I suspect <br />> 2.6.11 is a replay of 2.6.8 for the regression in the scheduler. We <br />> are running experiment to confirm this theory.<br /><br />the current defaults for cache_hot_time are 10 msec for NUMA domains, <br />and 2.5 msec for SMP domains. Clearly too low for CPUs with 9MB cache. <br />Are you increasing cache_hot_time in your experiment? If that solves <br />most of the problem that would be an easy thing to fix for 2.6.12.<br /><br /> Ingo<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>