CINXE.COM
LKML: Willy Tarreau: Re: [PATCH] strict VM overcommit accounting for 2.4.32/2.4.33-pre1
<?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" /><!--Matomo--><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><!--End Matomo Code--></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/2005/12/31/14" onclick="this.href='/lkml/headers'+'/2005/12/31/14';">[headers]</a>聽 <a href="/lkml/bounce/2005/12/31/14">[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/2005/12/30/212">Alan Cox</a><ul><li><a href="/lkml/2005/12/30/243">Al Boldi</a><ul><li class="origin"><a href="/lkml/2005/12/31/64">Willy Tarreau</a><ul><li><a href="/lkml/2005/12/31/64">Al Boldi</a><ul><li><a href="/lkml/2005/12/31/66">Willy Tarreau</a></li><li><a href="/lkml/2005/12/31/67">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 08:38:18 +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 07:59:02AM +0300, Al Boldi wrote:<br />> Alan Cox wrote:<br />> > On Gwe, 2005-12-30 at 23:06 +0300, Al Boldi wrote:<br />> > > > +3 - (NEW) paranoid overcommit The total address space commit<br />> > > > + for the system is not permitted to exceed swap. The machine<br />> > > > + will never kill a process accessing pages it has mapped<br />> > > > + except due to a bug (ie report it!)<br />> > ><br />> > > This one isn't in 2.6, which is critical for a stable system.<br />> ><br />> > Actually it is<br />> ><br />> > In the 2.4 case we took "50% RAM + swap" as the safe sane world 'never<br />> > OOM kill' and to all intents and purposes it works. We also had a 100%<br />> > paranoia mode.<br />> ><br />> > When it was ported to 2.6 (not by me) whoever did it very sensibly made<br />> > the percentage tunable and removed "mode 3" since its mode 2 0% ram and<br />> > can be set that way.<br />> <br />> Only, doesn't this imply that you cannot control overcommit unless backed by <br />> swap? i.e Without swap the kernel cannot use all of ram, because it would <br />> overcommit no-matter what, thus invoking OOM-killer.<br />> <br />> Which raises an important question: What's overcommit to do with limiting <br />> access to physical RAM?<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!<br />> <br />> --<br />> Al<br /><br />Cheers,<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 08:43 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>