Subject: Re: compressed virtual memory - interested?
To: None <email@example.com>
From: Ignatios Souvatzis <firstname.lastname@example.org>
Date: 05/31/1995 09:10:00
Hi VaX#n8 <email@example.com>,
> ... contemplating working on a compressed virtual memory system for NetBSD.
> This would be for 3 hours of credit over the summer semester (i.e. expected
> target date around August).
> I'm curious how receptive people would be to using something like this.
> I believe the professor I would be working under already has a working port
> on another Unix system, but to be honest I'm not sure exactly which one.
> The compression ratio is somewhere around 3:1, using little braindead
> algorithms (i.e. not LZW) that run very quickly.
3:1, with a algorithm worse than lzw? I don't believe that for most of
my VM memory space. lzw and friends do about 40-50% on normal text, c
source text, or binaries... a bit more on News batches, of course, due
to lots of repetitive information in the headers (and bodies :-) gzip is
a bit better, but uses LOTS of cycles.
As you probably will only compress small amounts at a time (the 8k
pages), the ratio will even be worse.
> I'm also looking for feedback on the degree of difficulty of this project,
> and relevant references. I've not done very much kernel hacking.
> I've got a spare machine, spare drive space, Leffler et. al. Design & Impl,
> and the (perhaps not relevant) i386 data book. Opinions?
You're sure you want to trade yet more cpu cycles against swap space?
Then again, you would save on the paging i/o operations. Hm... I guess
there are other things to do in NetBSD, but if you insist (and get
credits for it at your university), it might be worth a try. At least
for fast CPU's like i486 or Pentium or 68040/60 above 50 MHz.
Btw. this would be a quiestion for tech-kern, not current-users, I
guess... I've set the Reply-To: and cc: