From owner-freebsd-current Mon Jul 27 16:08:57 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id QAA02371 for freebsd-current-outgoing; Mon, 27 Jul 1998 16:08:57 -0700 (PDT) (envelope-from owner-freebsd-current@FreeBSD.ORG) Received: from dyson.iquest.net (dyson.iquest.net [198.70.144.127]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id QAA02354 for ; Mon, 27 Jul 1998 16:08:45 -0700 (PDT) (envelope-from toor@dyson.iquest.net) Received: (from root@localhost) by dyson.iquest.net (8.8.8/8.8.8) id SAA02258; Mon, 27 Jul 1998 18:05:17 -0500 (EST) (envelope-from toor) Message-Id: <199807272305.SAA02258@dyson.iquest.net> Subject: Re: New LINT options: what is VM coloring? In-Reply-To: <199807272023.NAA20781@usr08.primenet.com> from Terry Lambert at "Jul 27, 98 08:23:32 pm" To: tlambert@primenet.com (Terry Lambert) Date: Mon, 27 Jul 1998 18:05:17 -0500 (EST) Cc: dyson@iquest.net, dfr@nlsystems.com, tlambert@primenet.com, roberto@keltia.freenix.fr, freebsd-current@FreeBSD.ORG From: "John S. Dyson" Reply-To: dyson@iquest.net X-Mailer: ELM [version 2.4ME+ PL38 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-current@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG Terry Lambert said: > > For two level caches, the coloring should work fine. However, it is > > problematical with 3-level alpha caches. My guess is that one would > > want to page-color for the 2nd and 3rd level caches, but of course, > > you might want to experiment. It is critical to color for the larger > > cache. To color for all three caches, you could just expand the > > coloring scheme in FreeBSD. I don't know if it is the right approach > > though. > > The Digital literature suggests for small enough working sets, putting > the entire working set into L1 cache and keeping it there. I don't > know how you would ensure this if you were allowing interrupts and > drivers to run, except to minimize the use of seperate stacks as > much as possible, and reduce the amount of auto data usage and function > call depth for interrupt code as much as possible. > > The method they state they use in order to do this is "page coloring" > of the L1 cache. > > This is from the Digitasl UNIX 2.0 release notes. > But - but - but, the L1 cache is small!!! Is our terminology mixed-up? On X86en, L2 cache is the big one. I thought that on an Alpha, the L3 cache is the big one. I don't know how one would fit the working set into a 16K or even 128K cache?!?!?!? I can understand fitting the working set into a 1M -> 4M L3 cache though :-). -- John | Never try to teach a pig to sing, dyson@iquest.net | it makes one look stupid jdyson@nc.com | and it irritates the pig. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message