From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from max.fys.ruu.nl (max.fys.ruu.nl [131.211.32.73]) by kvack.org (8.8.7/8.8.7) with ESMTP id SAA01301 for ; Mon, 17 Nov 1997 18:50:45 -0500 Date: Mon, 17 Nov 1997 22:15:46 +0100 (MET) From: Rik van Riel Subject: Re: memory management wishes... In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-linux-mm@kvack.org To: Gordon Oliver Cc: linux-mm List-ID: On Mon, 17 Nov 1997, Gordon Oliver wrote: > ... Rik van Riel said ... > >Send Linux memory-management wishes to me: I'm currently looking > >for something to hack... > > I'd like to see an experiment done where some aggressive swapping code marks > pages for swapping and makes them "non-resident" without actually doing the > swapping. I.e. > 1) mark pages for swapping aggressively, marking them non-resident > in the page tables at the same time. > 2) gather statistics for pages that have been marked non-resident, > trying to figure out the "value" of a page. > 3) Use these statistics to swap out little used pages rapidly... > > The advantage is that it gives the possibility of aggressively swapping without > taking the entire penalty... I'm not sure if this will actually help in the > end, but it is cool research, and might get a big win. I believe this is what 'real' unixen already do, they have an 'inactive' list of not-so-often used pages that are ready to be swapped out. They even prepage the head of the inactive list. If a page from the inactive list _is_ used before being swapped out, it is 'reactivated'. To implement this we would need: - a big chunk of memory to hold the list - a mechanism to build the list - a preswapping/freeing daemon (easy) - the willingness to code all of this Rik. ---------- Send Linux memory-management wishes to me: I'm currently looking for something to hack...