2.4.0-release (and a few priors) stalls

Recently, about test 12 I believe, I started experiencing stalls. I believe it has to do with VM pressure but I'm not sure.

What happens: 5-60 second instant dead stall, nothing at all happens. No sound/key/disk/anything activity, screen updates stop in the middle of an update. Until recently I was mistaking this as a lockup and rebooting (thank you reiserfs). Now I just wait it out. I'm away for holiday/vacation so I can't hookup kbd. :(

Environment: normally swap is full -but- buffers can have any amount of free listed, sometimes well over 100 megs.

After the stall resumes, everything starts right up and sings along fine. There aren't any kernel messages indicating something was funny.

This is different from running out of normal pages, the system starts to thrash and slows considerably, almost unusable, but over a period of 10-20 seconds. If I add swap quickly, X won't get killed. (unfortunately, X seems to get killed 9/10 times because netscape took too long to die so the kernel kept killing).

-d

begin:vcard
n:Ford;David
x-mozilla-html:TRUE
url:www.blue-labs.org
adr:;;;;;;
version:2.1
email;internet:[email protected]
title:Blue Labs Developer
note;quoted-printable:GPG key: http://www.blue-labs.org/[email protected]=0D=0A
x-mozilla-cpt:;9952
fn:David Ford
end:vcard

Probably an interaction between the fact that the
VM tries to write out pages but reiserfs blocking
this because of write ordering constraints...

It should become better once reiserfs has a smart
->writepage() function that will push the "right"
page to disk and allows the system to continue
faster.

The fact that swap is full may also have something to do with it, though on my tests here the system continued to run fine with swap space filled up.

regards,

Rik
--
Hollywood goes for world dumbination,

	Trailer at 11.

		http://www.surriel.com/

http://www.conectiva.com/ http://distro.conectiva.com.br/

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/