Re: Linux's problems with too much memory - summary /2
Rogier Wolff (R.E.Wolff@BitWizard.nl)
Sun, 27 Apr 1997 15:19:55 +0200 (MET DST)
Pavel Machek wrote:
>
> It is *not* cache problem (present even with l2 disabled).
> It is *not* interleaving problem. Why do I think so?
>
> Compile times (compile of 2.1.36 kernel):
>
> mem=20M - 30 minutes
> mem=16M - 22 minutes
> [well, that proves nothing, but look at following]
> mem=16490k - 22 minutes
> mem=16800k - 29.8 minutes
> mem=17M - 30 minutes
Some critical kernel-datastructures might end up a few kilobytes below
end-of-memory.
It is cache related. The problem is that some BROKEN '486 motherboards
actually activate a pin on the '486 to indicate that they can't cache
that address. The engineers misread the '486 manual, because that
pin causes the '486 to prevent caching internally. This could
still be happening with L2 cache disabled. L1 cache is still a large
win on a '486.....
Roger.