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: Willy Tarreau: Re: [PATCH] strict VM overcommit accounting for 2.4.32/2.4.33-pre1</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 Willy Tarreau" href="/groupie.php?aid=143" /><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/2005/12/31"> [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/2005/12/31/66" onclick="this.href='/lkml/headers'+'/2005/12/31/66';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/31/66">[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/2005/12/30/122">First message in thread</a></li><li><a href="/lkml/2005/12/31/14">Willy Tarreau</a><ul><li><a href="/lkml/2005/12/31/64">Al Boldi</a><ul><li class="origin"><a href="">Willy Tarreau</a></li><li><a href="/lkml/2005/12/31/67">Arjan van de Ven</a><ul><li><a href="/lkml/2005/12/31/76">Willy Tarreau</a><ul><li><a href="/lkml/2005/12/31/79">Arjan van de Ven</a></li></ul></li><li><a href="/lkml/2005/12/31/108">Al Boldi</a><ul><li><a href="/lkml/2006/1/1/6">Arjan van de Ven</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">Sat, 31 Dec 2005 15:18:41 +0100</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Willy Tarreau <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: [PATCH] strict VM overcommit accounting for 2.4.32/2.4.33-pre1</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">On Sat, Dec 31, 2005 at 05:02:20PM +0300, Al Boldi wrote:<br />> Willy Tarreau wrote:<br />(...)<br />> > As shown in my previous mail, it allows malloc() to return NULL. I've<br />> > also successfully verified that it allows mmap() to fail if there is<br />> > not enough memory. I disabled swap, and set the overcommit_ratio to 95<br />> > and could not kill the system. Above this, it becomes tricky. At 97, I<br />> > see the last malloc() calls take a very long time, and at 98, the<br />> > system still hangs. But 95% without swap seems stable here.<br />> <br />> Thanks, for confirming this! And I agree that this patch and 2.6 offer an <br />> important and necessary workaround to inhibit OOM-killer, but it's no more <br />> than a workaround.<br />> <br />> And so the question remains: Why should overcommit come into play at all <br />> when dealing with physical RAM only?<br /><br />It is very important when you have many processes wasting lots of unused<br />memory. For instance, when firefox allocates 10 MB and uses only 3, then<br />the remaining 7 MB can be used by another process. But if finally firefox<br />tries to use them, and there is no memory left, then chances are that some<br />processes will be killed. I believe the same problem happens after a fork()<br />because data gets COWed but I'm not certain about this.<br /><br />I'd bet that using a heavy GUI under X with no swap an overcommit_ratio<br />set around 95%, you could get occasionnal malloc() failures. But once<br />again, I may be wrong.<br /><br />> Shouldn't it be possible to disable overcommit completely, thus giving kswapd <br />> a break from running wild trying to find something to swap/page, which is <br />> the reason why the system gets unstable going over 95% in your example.<br /><br />I think it's going unstable above 95% because there are lots of other<br />areas consuming memory. I don't know for example if dentry caches,<br />network buffers, various hash tables, etc... are accounted for.<br /><br />> Thanks!<br />> <br />> --<br />> Al<br /><br />Willy<br /><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: 2005-12-31 15:25 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>