Taylor R Campbell <[email protected]> wrote:

> Can you get kernel stack traces from the stuck processes using crash(8)?

Well, that requires to crash a mail server in busy hours on purpose. I
will collect if it happens again by accident.
 
> Does it unstick, or does it remain stuck forever?  That is, what makes
> you conclude `everything behaves like if all I/O were serialized'?

It seems processes get to execute after a long time, since the load
manages to go up and down. I did not conclude about serialized I/O, I
just said it behaved somehow like it: all processes doint I/O wait in
tstile and sometime get an opportunity to work. But it could also be
that there are very small windows for all processes to run.

-- 
Emmanuel Dreyfus
http://hcpnet.free.fr/pubz
[email protected]

Reply via email to