Scyld - Weird Beostatus bit here
niles at scyld.com
Tue Apr 3 21:37:39 PDT 2001
> What was not expected was that the cpu0 counter doesn't seem to track
> properly. I started up an instance of distributed.net client on both nodes,
> and node 11 shot to 100%, like it should. Node 10 though seems stuck at 5%,
> but the other counters (mem, disk etc) all change normally.
Can you try watching "bpsh 10 cat /proc/stat" and see if the first
number (user counts) seems to change faster than the four number (idle
counts) while your running a busy job. This will prove that the
CPU is busy.
How does it compare with the output of "beostat -N 10 -s" ?
> The process is definitely running - I have it logging to the local ramdisk,
> and can watch the packets gets crunched and so forth. loadavg also shows
> the correct amount of utilization.
Well, if you checked the loadavg, then I guess it's busy.
> So what got left behind (and where) in beostatus' config ? How can it be
> reset or cleared out ? This problem (pretty minor actually) stays around
> even after a reboot of master and slaves.
Hmm...every time you restart the "recvstats" daemon all status
information should be reset to zero.
(i.e. "/etc/rc.d/init.d/beowulf restart")
Scyld Computing Corporation.
More information about the Beowulf