I'm having several problems concerning scheduling on my linux box. I'm
running it as a server and therefore I'd like it to be as reliable as
possible, but several niggling factors force me to reboot the box
The biggest problem I have is if there is no CPU idle time (as reported
by top(1)) processes take a very long time to start up, and sometimes
even fail to start entirely. I can trigger this very easily and it is
very reproducible. I tried to run the Distributed.net client at some
point but it ground my system to a halt (even nice'd to 19). This
prevents me from logging in using SSH to kill the offending process, or
checking my email, or any other task that requires starting processes.
The next biggest problem is related to the above, I think. When I
connect to my box using SSH, and the connection is broken somehow, the
SSHd process takes up all the available CPU after just a couple of
seconds. All I need to do to trigger this is to connect to the server,
then kill the SSH client process. After a few seconds, the server
process reaches 99.9% CPU usage:
[bootc@arcadia bootc]$ pa | grep notty
bootc 24408 99.9 0.4 10688 3928 ? R 14:15 0:38 sshd:
bootc 24448 0.0 0.0 1576 496 pts/7 S+ 14:16 0:00 grep
I'm running Debian Sarge on Linux 2.6.5 with no special patches. The
machine is a Shuttle XPC with an Athlon XP2000+ and 1GB of RAM. The
problems have occurred for as long as I can remember, definitely all
along the 2.6 kernel series, but most probably before then too.
Thanks for any pointers people can give me...