Date: Tue, 23 Apr 2013 19:26:09 +0200 From: Francois Tigeot <ftigeot@wolfpond.org> To: Francois Tigeot <ftigeot@wolfpond.org> Cc: pkg@freebsd.org, kernel@dragonflybsd.org, users@dragonflybsd.org Subject: Re: DragonFly 3.3 performance improvements Message-ID: <20130423172609.GA10285@sekishi.zefyris.com> In-Reply-To: <20130421214607.GA84673@sekishi.zefyris.com> References: <20130421214607.GA84673@sekishi.zefyris.com>
next in thread | previous in thread | raw e-mail | index | archive | help
--1yeeQ81UyVL57Vl7 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline On Sun, Apr 21, 2013 at 11:46:07PM +0200, Francois Tigeot wrote: > > One of the main differences between Dragonfly-3.2 and the future 3.4 > release is the use of DPorts and binary Pkgng packages. > The packages are mass-produced by Poudriere, a tool making so heavy > use of IO and CPU resources and exercising so heavily some operating > system functions it showed major stability and performance issues on > machines which had otherwise been running perfectly fine for months. > > These issues were fixed during the DragonFly 3.3 development cycle; I > have run a few benchmarks to show the resulting performance improvements. The attached munin CPU graph may also be of interest. It represents approximately a day of CPU usage with the 48-core Opteron machine used for the Poudriere benchmarks. On the left part, Poudriere is running with the March 15, 2013 DragonFly kernel and on the right part, the same workload is running with the December 26, 2012 kernel. With user time such a little part of the available CPU time, the December kernel clearly couldn't properly scale on this machine. -- Francois Tigeot --1yeeQ81UyVL57Vl7--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130423172609.GA10285>