Re: netscape + SIGUNUSED -> 96% CPU

Stephen C. Tweedie (sct@dcs.ed.ac.uk)
Sat, 3 Aug 1996 10:56:41 +0100


Hi,

"Aaron" == Aaron Ucko <UCKO@VAX1.ROCKHURST.EDU> writes:

>> That's one of the reasons why linux's interactive behaviour is really
>> pretty good even under load. If I've got a large simulation plus a
>> compile going on in the background, response time in interactive
>> sessions is still instantaneous.

> Depends on hardware. I've got a 486DX2/66 with 8M RAM swapping to IDE;
> if I compile the kernel and run Netscape at the same time, disk activity
> goes up markedly, both processes run at a snail's pace, and interactive
> performance isn't too hot either. OTOH, doing significantly better in
> this case would take a miracle.

Of course. We were talking about the scheduling algorithm. Memory
size versus working set size is another important factor for
performance, but it's a completely separate one. Performance is
always like this --- it's hard to analyse because there are so many
contributing factors. Knowing _that_ something is slow is far removed
from knowing _why_ it is slow. :)

Are you running the 2.0 kernel, btw? Later kernels do swap much
better than 1.2 (and I mean *much* better) under heavy load, but with
that kind of workload in 8MB you are never going to see really good
performance.

Cheers,
Stephen.

--
Stephen Tweedie <sct@dcs.ed.ac.uk>
Department of Computer Science, Edinburgh University, Scotland.