hi,

your observations support a theory I've had for a while,
which is that the interactive performance problem is really the combination
of two different phenomena:

 - application data gets paged out to swap when arguably it shouldn't.
 - other i/o in the system is swamping the disk, so reading anything back in
   takes a really long time.

it's the combination of these two things that is really bad.
everyone's been focusing on the first issue,
but alleviating the second issue should also improve
interactive performance a lot.  we need to address the
second issue even if the first issue is addressed,
and maybe the second issue will turn out to be easier.

-Chuck


On Thu, Oct 25, 2001 at 09:19:19PM -0500, Peter Seebach wrote:
> >I just pulled some iso files over to the -current system. This caused
> >netscape to page out until there was 5 megs left in memory. When the
> >ftp was done I switched to the netscape window and it refreshed in less
> >than 2 seconds.
> 
> Yeah.  The problem is definitely worse when I'm doing, say, a "make build"
> while trying to do stuff, but it would be nice to be able to tell the buffer
> cache to be a tad less aggressive.
> 
> -s