Re: swap killer kernels

Krzysztof G. Baranowski (kgb@manjak.knm.org.pl)
Tue, 31 Mar 1998 15:13:11 +0200 (CEST)

While testing the million monkey theory, Linus Torvalds wrote:
>The 2.1.x kernels almost require a swap area. They should work without
>one, but as you see they can get rather upset if they don't find one.
[...]
>Note that this really isn't anything new as of 2.1.x - it's fairly true
>in 2.0.x too. It just becomes more obvious when the pager is more
>aggressive.
That's obvious, but the problem with my setup (32 MB RAM, 16 SWAP) was
that kernel (2.1.91) always tried to keep 6,5-7 MB of free memory,
even for price of constant swapping. The amount of free memory was
_never_ lower than about 6,5 MB. If I started an application kernel
swapped out the memory to keep 6,5-7 MB free.

.92-pre looks a little bit better, but I haven't stressed it as much
as 2.1.91. The worst thing with 2.1.91 was that it swapped to death
at night, when the machine was totally idle, no users, only few
standard daemons. I left the machine at about 2 AM (the cron-daily
has been done at 1 AM), and it woke me up at 5AM with constant
swapping. I could only use SysRq.

I think I'll wait till 2.1.92 will be released and then do some more
tests. Then I'll go back with some more reports and we will see if
kswapd should be tuned a little.

Regards,
Kris

-- 
Krzysztof G. Baranowski - President of the Harmless Manyacs' Club
"Smith & Wesson - The original point and click interface..."
http://www.knm.org.pl/                 <prezes@manjak.knm.org.pl>




-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu