--A100B2ED3.852753805=_/vger.rutgers.edu
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Content-ID: <Pine.SOL.3.91.970108163900.16991H@fan1.fan.nb.ca>
Message contained in attachment.
-- William Burrow -- Fredericton Area Network, New Brunswick, Canada Copyright 1997 William Burrow This line left intentionally blank. And the one below it.--A100B2ED3.852753805=_/vger.rutgers.edu Content-Type: MESSAGE/RFC822 Content-ID: <Pine.SOL.3.91.970108163900.16991J@fan1.fan.nb.ca>
Received: from unb.ca ([131.202.3.20]) by vger.rutgers.edu with ESMTP id <213252-26718>; Wed, 8 Jan 1997 15:03:20 -0500 Received: (from aa126@fan1.fan.nb.ca [131.202.9.99]) by unb.ca (8.7.6/961016-08:40) id QAA02827; Wed, 8 Jan 1997 16:08:28 -0400 (AST) Received: by fan1.fan.nb.ca (8.6.12/951109-23:05) id QAA15879; Wed, 8 Jan 1997 16:08:24 -0400 Date: Wed, 8 Jan 1997 16:08:22 -0400 (AST) From: William Burrow <aa126@fan.nb.ca> To: Wolfram Gloger <Wolfram.Gloger@dent.med.uni-muenchen.de> cc: linux-devel@vger.rutgers.edu Subject: Re: Zero page In-Reply-To: <u4916480o4.fsf@md.dent.med.uni-muenchen.de> Message-ID: <Pine.SOL.3.91.970108160324.14582C-100000@fan1.fan.nb.ca> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII
On Wed, 8 Jan 1997, Wolfram Gloger wrote:
> William Burrow <aa126@fan.nb.ca> wrote: > > > Someone suggested to me that a system running X, httpd and few users might > > swap about 6-10% zero filled pages. > > I doubt this, if the programs in question are linked against recent > versions of libc. This is because with the new malloc(), calls to > calloc() do not necessarily touch the pages by memset()ing them to > zero. So all `excess' pages allocated remain copies of the global > zero page, and are not swapped at all.
Yes, I ran into the calloc() problem with an older copy of libc when trying to run tests on empty pages. This might significantly skew the results given the new kernel yet to be released requires the new libc. It would be interesting to see what the stats would be with the new libc.
> As you said, it is really more a user-space problem. With libc, I [...] > > With Fortran -- hmm, don't know for sure, but if it allocates large > sparse matrices with brk() system calls, it should also work without > dirtying all pages. (If not, maybe consider using f2c.) Also, when > you write zeros to a sparse matrix from your code, you could check for > it yourself before performing any (unnecessary) assignments.
Fortran is the amongst the problems of the special case I am considering, large processes. Some people use published libraries and are unwilling to change them. Checking each value stored for zero may also be perceived as a performance hit. Consider that checking the page for zeroes only on swap out consolidates the hit on performance in exchange for reducing the hit on another action: swapping. More data on how much this helps would be nice before dismissing it out of hand.
-- William Burrow -- Fredericton Area Network, New Brunswick, Canada Copyright 1997 William Burrow This line left intentionally blank. And the one below it.--A100B2ED3.852753805=_/vger.rutgers.edu--