Hi folks,
I've converted yesterday my router, which until now had been happily running
ext2+2.4.13-pre2 on 8 MB of RAM + 200 MB of swap, to the ext3 + 2.4.16 (and
2.4.17-pre2) combinations (still eight megs of RAM, unfortunately 8-bit
SIMMs ain't cheap nowadays).
Now, as soon as the system gets some use (inetd kicks exim in, one ssh
attempt, etc.), most processes go freeze themselves into
<shrink_caches +57/80>
I can very temporarily regain some control over the system by SAKing it, but
eventually, everything userland is frozen up (packets still get routed, but
after a while I get a load of netfilter-related messages questioning Rusty's
sanity, which I'm willing to ignore as long as the VM is misbehaving).
Going a little backwards, to 2.4.13-ac8 (of course) solves the problem (but
incidentally, the interactive feel is much worse than what 2.4.16 gives
before it freezes).
What can I do to further isolate the problem ?
Thanks for any help.
-- Cyrille
--
Grumpf.
On Wed, Dec 05, 2001 at 06:55:09AM +0100, Cyrille Chepelov wrote:
> Hi folks,
>
> I've converted yesterday my router, which until now had been happily running
> ext2+2.4.13-pre2 on 8 MB of RAM + 200 MB of swap, to the ext3 + 2.4.16 (and
> 2.4.17-pre2) combinations (still eight megs of RAM, unfortunately 8-bit
> SIMMs ain't cheap nowadays).
>
> Now, as soon as the system gets some use (inetd kicks exim in, one ssh
> attempt, etc.), most processes go freeze themselves into
> <shrink_caches +57/80>
I never reproduced anything wrong here, testing on highmem and non
highmem. Just in case can you also reproduce with 2.4.17pre1aa1?
Andrea
Le mer, déc 05, 2001, ? 09:46:32 +0100, Andrea Arcangeli a ?crit:
> > Now, as soon as the system gets some use (inetd kicks exim in, one ssh
> > attempt, etc.), most processes go freeze themselves into
> > <shrink_caches +57/80>
>
> I never reproduced anything wrong here, testing on highmem and non
> highmem. Just in case can you also reproduce with 2.4.17pre1aa1?
Here are the results; basically, I've pitted 2.4.17-pre2 against
2.4.17-pre2+pre1aa1. The short story is, pre1aa1 works great for the
workload I've thrown at it (Debian 'woody'; long netfilter rules;
console-only runlevel2; exim; apache; apt-get update && apt-get -uy
dist-upgrade; ssh; one updatedb ; top). pre2 doesn't
even make it to runlevel2 (I've made two tests with the "andrea" kernel, one
in single mode for comparison with "marcelo" kernel, one in normal
multi-user mode).
Incidentally, I perceive a much better performance with 2.4.17-pre2+pre1aa1
than with 2.4.13-ac8 (Rik's VM).
-- Cyrille
--
Grumpf.