Subject: Re: Do some disk accesses miss the UVM?
To: Perry E. Metzger <firstname.lastname@example.org>
From: Chuck Silvers <email@example.com>
Date: 01/26/2002 15:06:52
On Sat, Jan 26, 2002 at 12:59:42PM -0500, Perry E. Metzger wrote:
> Chuck Silvers <firstname.lastname@example.org> writes:
> > actually the generational idea isn't a replacement for the knobs.
> > the generational bit still depends on the assumption that past behaviour
> > is a good predictor of future behaviour, so it falls down just like the
> > pre-knob code when that assumption isn't true.
> Absolutely, but *every* policy does. The best part of the research of
> the last 30 years on the subject is the realization that even if you
> have a psychic computer that knows what the entire future of page use
> will be, in the most general case it is still extremely
> computationally expensive to determine what pages to evict.
of course, but that's only marginally related to what I said.
or are you implying that using information about the type of data
that pages contain in the replacement code is necessarily too
computationally expensive, or that the class of applications
that require it is unimportant? I don't agree with that.
my point is that the current page replacement code uses both
page access history and page content type to make decisions,
and I believe that both of these are required to have a robust system.
the processing of both types of information can and should be improved
in our implementation, and the generational idea has potential improve
the usage of page access history. I have ideas on how to improve how
we use page content type too. but we need to have both.