On Sun, 30 Jun 1996 12:51:26 GMT, steve@mctavish.demon.co.uk (Steve
Woodford) wrote:

>Is this a problem specific to the i386 port?

At least one person has reported it on a Sparc, so it isn't
i386-specific.

>I'm running 1.1 on my server mvme68k, with a couple of diskless
>mvme68k's over nfs. Earlier last week, I decided to put -current
>on one of the nfs cpus. This basically involved compiling the
>whole of -current from scratch, on a cpu with 8Mb RAM, and 8Mb
>swap over nfs.

>The build took around two days all together, with some manual
>intervention. At one point, swap usage was up around 95% on the
>build machine (ie. the one with 8Mb swap in total), but things
>were motoring on without any noticable slow down.

The problem isn't a constant thing - it generally happens at the point
when it first starts swapping, after which it will work just fine for
some time. You probably wouldn't notice it unless you were constantly
watching top and systat (as we do during these performance runs), or
using the same machine to run an editor or something.

>Obviously, you lot seem to be pushing the VM system harder than
>me, but I'm still pretty amazed that systems with bags more RAM
>and swap than I have can suffer like you say.

Possibly the amount of suffering is in direct proportion to the large
amount of RAM/swap we have...