Date: Tue, 19 Nov 2013 16:30:39 +0200 From: Alexander Motin <mav@FreeBSD.org> To: Gleb Smirnoff <glebius@FreeBSD.org> Cc: svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org Subject: Re: svn commit: r258336 - head/sys/vm Message-ID: <528B760F.3010001@FreeBSD.org> In-Reply-To: <20131119121206.GH7577@FreeBSD.org> References: <201311191005.rAJA5snM048103@svn.freebsd.org> <20131119121109.GG7577@FreeBSD.org> <20131119121206.GH7577@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 19.11.2013 14:12, Gleb Smirnoff wrote: > On Tue, Nov 19, 2013 at 04:11:09PM +0400, Gleb Smirnoff wrote: > T> Alexander, > T> > T> On Tue, Nov 19, 2013 at 10:05:53AM +0000, Alexander Motin wrote: > T> A> Author: mav > T> A> Date: Tue Nov 19 10:05:53 2013 > T> A> New Revision: 258336 > T> A> URL: http://svnweb.freebsd.org/changeset/base/258336 > T> A> > T> A> Log: > T> A> Implement soft pressure on UMA cache bucket sizes. > T> A> > T> A> Every time system detects low memory condition decrease bucket sizes for > T> A> each zone by one item. As result, higher memory pressure will push to > T> A> smaller bucket sizes and so smaller per-CPU caches and so more efficient > T> A> memory use. > T> A> > T> A> Before this change there was no force to oppose buckets growth as result > T> A> of practically inevitable zone lock conflicts, and after some run time > T> A> per-CPU caches could consume enough RAM to kill the system. > T> > T> Brief review of patch tells me that system never recovers from this. uz_count > T> it decremented only and never incremented. You are wrong. uz_count is incremented in uma_zalloc_arg() same as it was always before. Probably it was too brief review. :) > Do I understand right that r258338 fixes that? Not fixes (since there was no bug), but adds second increment point. -- Alexander Motin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?528B760F.3010001>