Dai,
For the spin mutexes (smtx), you might consider running
'/usr/sbin/lockstat -CcwP -n 50000 -D 20 -s 40 sleep 2'. This is what
Sun's Guds script runs, and it's handy for lock examination.
The migrations could be caused by whatever's waiting on the above
locks. If not, you might search around for some dtrace scripts for
examining the migrations. I'm sure someone has a script that makes use
of the "on-cpu" and "off-cpu" triggers. Otherwise, I have attached a
general purpose profiling dtrace, for examing all things running on all
cpus. This might help pinpoint the cause. If you don't care about
application stacks, just remove the "@ustacks[ustack()] = count();" line.
Jim
dai wenjun wrote:
Maybe something causes this problem ? Could you give some advise?
/* To run:
* dtrace -s ./procstacks.d <process name>
*/
fbt:::entry
/execname == $$1/
{
@stacks[stack()] = count();
@ustacks[ustack()] = count();
}
profile:::tick-15sec
{
exit(0)
}
END
{
printa(@stacks);
printf("---------------------------------");
printa(@ustacks);
}
_______________________________________________
networking-discuss mailing list
[email protected]
