Linus wrote
>On 15 Apr 1996, really kuznet@ms2.inr.ac.ru wrote:
>> I have Pentium 133 64Mb AIC7881 and run program
>> that eats >=3D128Mb of virtual memory (certainly, matrices).
>
>Agreed. This is one case where linux doesn't do anything clever at all=
,=20
>and one place where we should probably do some work. It happens for a =
lot=20
>of scientific calculations.
[...]
Here is a user's comment, I hope you don't mind my interfering with you=
r
discussions:
I am a scientist and I do a lot of calculations. I know a _lot_ of othe=
r
scientist who, like me, would love to be able to do their calculations =
on
their local pc rather than sharing cpu time on a mainframe. One of the =
main
reasons we can't do that is lack of memory on local pc's and the
unbearably low speed of jobs that go over the physical memory limit in
comparison to ones that fit into the memory. I haven't run any tests an=
d can
therefore not comment on the performance of linux. I would think that a=
good
share of the potential serious user's of linux out there have something=
to
do with academia, science, research, or otherwise with number crunching=
or
large data set analysis. For instance, a friend of mine is thinking of
setting up a P200 with linux for medical MRI image processing of 3D
high-resolution scans of the human skull. He, like me, would use the li=
nux
machine mainly as a single user single task performance machine and, li=
ke
me, wouldn't mind the minor delay time increase when changing tasks as =
much
as a speed decrease for large single task jobs.
In other words, I merely want to second the suggestion that a single ta=
sk
optimised/improved swapping algorithm might be desirable. Maybe as a=20
compile time option ?
Sincerely and gratefully for your outstanding development work,
Dr. Peter J=E4ckel
Joint European Torus