From owner-freebsd-hackers Fri Aug 9 10:45:44 1996 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id KAA26463 for hackers-outgoing; Fri, 9 Aug 1996 10:45:44 -0700 (PDT) Received: from phaeton.artisoft.com (phaeton.Artisoft.COM [198.17.250.211]) by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id KAA26458 for ; Fri, 9 Aug 1996 10:45:42 -0700 (PDT) Received: (from terry@localhost) by phaeton.artisoft.com (8.6.11/8.6.9) id KAA19017; Fri, 9 Aug 1996 10:40:57 -0700 From: Terry Lambert Message-Id: <199608091740.KAA19017@phaeton.artisoft.com> Subject: Re: "Panick" - help needed... To: joerg_wunsch@uriah.heep.sax.de Date: Fri, 9 Aug 1996 10:40:57 -0700 (MST) Cc: freebsd-hackers@FreeBSD.org, rminnich@Sarnoff.COM In-Reply-To: <199608090547.HAA02326@uriah.heep.sax.de> from "J Wunsch" at Aug 9, 96 07:47:57 am X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-hackers@FreeBSD.org X-Loop: FreeBSD.org Precedence: bulk > > If anyone runs this on a freebsd desktop i'd be interested in what you > > observe -- how does interactive response function as this program runs. > > About the same as you observed on Linux. As i wrote in the other > mail, i had to force a `top' to get CPU cycles using `rtprio'. The problem appears (from a first perusal of the code) that the cache is being thrashed. The need to be able to support a per vnode working set quota on page usage by: 1) reclaiming in a local LRU 2) inserting pages reclaimed from this LRU overflow at the *head* of the free pool LRU so that it will be the next to be reused. This should move the locality from global to per vnode, and prevent a given vnode from monopolizing the available pages from a single process instead of preserving other processes locality. This technique was used to great effect on a UnixWare 2.x I had hacked to resolve the fact that ld mmap'ed a bunch of files and thrashed the hash to the point where the X server failed to move the cursor when you moved the mouse. Unfortuantely, my changes were ignored, and they implemented a "fixed" scheduling class instead -- their solution, which I must say, doesn't fix the problem. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.