If you're looking for something that will dump out the accumulated
time of the LWPs in an MT app then another mechanism might better.
a /proc reader that iterated over all the LWPs for the process
would be one possibility. DTrace may well not be the best way
to do this thing.
Jim
---
Jim Mauro wrote:
What problem are you trying to solve?
Use "prstat -Lm", or "prstat -Lmp <PID>",
at 1 second intervals for a fixed duration,
and do the math. The various columns produced
by "pstat -m" (USR SYS SLP, etc) represent the
percentage of time the sampled thread spent in each
microstate for the last sampling period (except the
last 4 columns, which are counts).
Chad Myhier improved ptime to include reporting
of time spent in various microstates. I can send you
a binary - are you on x86 or SPARC?
You should also explore extended accounting.
HTH,
/jim
tester wrote:
Hi,
Need some dtrace pointers to track time spent by some programs. N
processes with N threads each running on a multiprocessor system.
What probes to use to decompose total run time to on-proc, sleep, io
etc?
Since it's a muliprocessor system, I am guessing that the toal units
of time will be in the terms of " clock ticks" and so will be the
break up.
So for example on a 2X2Ghz system a program running for 10 sec has,
theoretically 10*2*2*1024*1024 clock ticks potentially available. How
to track how much of this the program is using (with a breakup of
on-cpu and io wait etc)
Or am I completely off on this with a different approach?
Also what will the additional impact of tracking such usage with Dtrace.
Thanks
_______________________________________________
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org
_______________________________________________
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org
