From mboxrd@z Thu Jan 1 00:00:00 1970 Date: Tue, 23 May 2000 12:51:15 -0400 (EDT) From: Chuck Lever Subject: Re: [PATCH--] Re: Linux VM/IO balancing (fwd to linux-mm?) (fwd) In-Reply-To: <392AA3D5.FD6B5399@norran.net> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-linux-mm@kvack.org Return-Path: To: Roger Larsson Cc: Rik van Riel , linux-mm@kvack.org, Matthew Dillon List-ID: hi roger- list manipulations are probably more expensive than maintaining a "load average" value associated with a page. usually a list manipulation will require several memory writes into areas shared across CPUs; maintaining a weighted load average requires a single write. this was an issue with andrea's original LRU implementation, IIRC. On Tue, 23 May 2000, Roger Larsson wrote: > From: Matthew Dillon > > The algorithm is a *modified* LRU. Lets say you decide on a weighting > > betweeen 0 and 10. When a page is first allocated (either to the > > buffer cache or for anonymous memory) its statistical weight is > > set to the middle (5). If the page is used often the statistical > > weight slowly rises to its maximum (10). If the page remains idle > > (or was just used once) the statistical weight slowly drops to its > > minimum (0). > > My patches has been approaching this a while... [slowly...] > The currently included patch adds has divided lru in four lists [0..3]. > New pages are added at level 1. > Scan is performed - and referenced pages are moved up. > > Pages are moved down due to list balancing, but I have been playing with > other ideas. > > These patches should be a good continuation point. > Patches are against pre9-3 with Quintela applied. - Chuck Lever -- corporate: personal: The Linux Scalability project: http://www.citi.umich.edu/projects/linux-scalability/ -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux.eu.org/Linux-MM/