Re: mmap/write vs read/write supprise
Rogier Wolff (R.E.Wolff@BitWizard.nl)
Fri, 11 Jul 1997 17:46:54 +0200 (MET DST)
Eric.Schenk@dna.lth.se wrote:
>
>
> mcculley@iag.net <mcculley@iag.net> writes:
> >If we had madvise, the responsibility for determining the access
> >pattern could be put on the application. We could still use a better
> >default access pattern for mmap'd pages, though. Is the VM code
> >sophisticated enough yet to take advantage of an madvise syscall?
>
> Another approach to this would be to use a predictive algorithm
> to determine which pages to prefetch. There was a neat article
> on doing this in JACM recently. The idea is to use the core of
> a good compression algorithm as the predictor. Apparently
> this can be proved to be within a constant of the optimal
> prefect strategy (i.e. the one that knows all future
> fetches in advance). I don't know off hand what the real
> performance of this would be like for kernel VM, but
> someone ambitious with a lot of time on their hands
> may want to play :)
Compression algorithm (compress, gzip) work by assigning "short
bitstreams" to "strings that we've recently seen".
I don't see how this can be adapted to predicting access patterns.
ROger.
>
> --
> Eric Schenk www: http://www.dna.lth.se/~erics
> Dept. of Comp. Sci., Lund University email: Eric.Schenk@dna.lth.se
> Box 118, S-221 00 LUND, Sweden fax: +46-46 13 10 21 ph: +46-46 222 96 38
>
>