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: Helge Hafting: Re: Linux 2.6.15-rc5: multiuser scheduling trouble</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 Helge Hafting" href="/groupie.php?aid=238" /><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/2005/12"> [Dec]</a> 聽 <a class="nb" href="/lkml/2005/12/10"> [10]</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/2005/12/10/74" onclick="this.href='/lkml/headers'+'/2005/12/10/74';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/10/74">[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/2005/12/4/2">First message in thread</a></li><li><a href="/lkml/2005/12/4/2">Linus Torvalds</a><ul><li><a href="/lkml/2005/12/4/62">Gene Heskett</a><ul><li><a href="/lkml/2005/12/4/89">Michael Krufky</a><ul><li><a href="/lkml/2005/12/4/110">Gene Heskett</a><ul><li><a href="/lkml/2005/12/4/113">Michael Krufky</a></li></ul></li></ul></li></ul></li><li><a href="/lkml/2005/12/4/79">Jesper Juhl</a><ul><li><a href="/lkml/2005/12/4/82">Michal Piotrowski</a><ul><li><a href="/lkml/2005/12/4/87">Jesper Juhl</a><ul><li><a href="/lkml/2005/12/4/119">Andrea Arcangeli</a></li></ul></li></ul></li></ul></li><li><a href="/lkml/2005/12/4/135">Johannes Stezenbach</a><ul><li><a href="/lkml/2005/12/5/172">Stephen Hemminger</a><ul><li><a href="/lkml/2005/12/5/188">Johannes Stezenbach</a></li></ul></li><li><a href="/lkml/2005/12/5/272">Bill Davidsen</a></li></ul></li><li><a href="/lkml/2005/12/7/333">Rich Walker</a></li><li class="origin"><a href="/lkml/2005/12/11/156">Helge Hafting</a><ul><li><a href="/lkml/2005/12/11/156">Linus Torvalds</a><ul><li><a href="/lkml/2005/12/12/13">Ingo Molnar</a><ul><li><a href="/lkml/2005/12/12/62">Helge Hafting</a></li><li><a href="/lkml/2005/12/12/67">Nix</a></li></ul></li><li><a href="/lkml/2005/12/12/55">Helge Hafting</a></li></ul></li></ul></li><li><a href="/lkml/2005/12/10/106">Brice Goglin</a><ul><li><a href="/lkml/2005/12/10/108">Lee Revell</a></li><li><a href="/lkml/2005/12/11/26">Hugh Dickins</a></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">Sat, 10 Dec 2005 17:27:59 +0100</td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: Linux 2.6.15-rc5: multiuser scheduling trouble</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Helge Hafting <></td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">2.6.15-rc5 (and earlier 2.6 kernels) seems to have a slight scheduling <br />problem, with some starvation at a load of only 2.<br /><br />All it takes is two users with separate X desktops.<br />User one plays flash games using firefox. The ill-designed flash stuff<br />typically means that his Xorg and firefox divides the cpu 100% between them.<br /><br />So I expect that I, as user two, should notice some slowness. I should get only<br />1/2 - 1/3 cpu. But I get periods of starvation. Logging in takes a long time,<br />bringing iup icewm takes 15s instead of 2, each xterm takes a long time to<br />start. They are usually instantaneous. Tha machine is unsuitable<br />for work in this mode.<br /><br />Knowing the root password I renices his Xorg and firefox by 10, and then<br />everything is fine. His games are still ok, and my xterms are snappy again.<br /><br />I have tried no preempt, voluntary preempt, and preemptible kernel. It doesn't<br />make a difference. This is an amd64 kernel on an opteron 244 (1800MHz). Everyhting<br />is 64-bit except firefox+flash which is 32-bit.<br /><br />Perhaps the way flash games work, with lots of communication with the xserver,<br />makes them get "io boost" even though they are cpu hogs. I still think<br />my xterm (or whatever I am starting up) should get its fair third of the cpu<br />though, (with firefox and xorg hogging one third each too.) Even a "600MHz opteron"<br />ought to do better than this.<br /><br />The machine isn't trashing, it is hardly touching swap. (512M memory, and swpd=16 <br />according to vmstat) The paging-in of a starting executable shouldn't be affected much<br />by the cpu load?<br /><br />Helge Hafting<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: 2005-12-10 17:30 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>