Date: Wed, 3 Mar 2004 15:35:12 -0700 (MST) From: Scott Long <scottl@freebsd.org> To: Stefan =?iso-8859-1?Q?E=DFer?= <se@freebsd.org> Cc: Vincent Poy <vince@oahu.WURLDLINK.NET> Subject: Re: buildworld times Message-ID: <20040303153011.P27388@pooker.samsco.home> In-Reply-To: <20040303221204.GA6234@StefanEsser.FreeBSD.org> References: <20040301054623.U8264-100000@oahu.WURLDLINK.NET> <Pine.NEB.3.96L.1040301201247.90719A-100000@fledge.watson.org> <20040303221204.GA6234@StefanEsser.FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 3 Mar 2004, Stefan [iso-8859-1] E=DFer wrote: > > I do not understand, why 'ratio' comes out that different (92-103% vs. 13= 7-148%) > for 5.2.1 vs. -CURRENT (for -j4 and up, where HT plays a role). > > Is the process accounting different, or has the scheduler been changed to= make > better use of logical processors ??? ULE is better in 5.2-CURRENT than in RELENG_5_2. However, the significant change that is contributing to this is the improvements to make(1). Prior to 5.2-CURRENT, 'make -jX' would poll every 20ms for completed children, and then schedule new children. Now there is a simplistic SIGCHILD handler that allows new children to be scheduled immediately, and of course eliminates the overhead of the polling. Since buildworld is such a large consumer of fork/exec, it's not surprising that this change helps a lot. Scott
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040303153011.P27388>