Richard McDougall wrote:
Hi Bob,

I've often wanted to see prstat changed to consistently show 'percent of a CPU core', so that values might exceed 100 for multi-threaded apps, and so per-process values from the same task running on different configurations would be more comparable.

Can you expand on the scenarios in which you would use this, an how
the information would be interpreted?

Sure.  For me, 'scenarios=all'. :-)

On a 72-way system, a single CPU hog shows as 1.4%. On an 8-way system, it will show as 12.5%. Absent the context of the core count, it's not obvious that it's a CPU- bound process, or that it's using the same absolute CPU resource on these two systems. On a (mythical) 250-way system, after rounding to the first decimal, it would show as 0.0%. I'd rather see it reported as using 100% (of a CPU) in all cases, without being
personally burdened with the task of dividing by N.

When using '-m' without '-L', I'd like to see that hog/6 is using 400% CPU without
needing to switch views.

With 32 CPU-bound threads on a T1-based system, I *think* I'd like to see each one
reported as using 25% (of a core).  With CMT, it maybe gets a bit fuzzy?

In other words, I think '%systemwide_cpu' is fundamentally annoying, as is seeing 'percent of total' (with -m) without seeing 'total' at the same time. I believe most
people would agree.  Am I right -- or just annoying?

Cheers,
-- Bob
Thanks,

Richard.



With '-m', the 'percent of total' concept needs to be maintained, but maybe adding another column would help. The issue of units extends to the '-T' rollups. Some on this thread have also wished for more 'top-like' outputs from prstat, such as aggregate
usr/sys/idle.  (I can offer free counseling for those who want %wio. ;-) )

Actually, as we evolve to systems architectures with massive thread counts, we'll
see CPU-bound threads average to 0% with the current implementation.  Yikes!

How about '-A' for 'Aggregate' and '-X' for 'eXtended'? I don't want to file an RFE
until there is some consensus on these notions.

Cheers,
-- Bob
Regards,

Richard.




On Thu, Oct 19, 2006 at 08:34:17AM -0700, Cherian Abraham wrote:
Hi All,

Thanks for all the feedback. I tried to run both prstat and prstat -m option and here are the output displays... (showing only three processes).
prstat -mL
PID USERNAME USR SYS TRP TFL DFL LCK SLP LAT VCX ICX SCL SIG PROCESS/LWPID 25952 root 30 30 0.0 0.0 0.0 0.0 30 0.0 100 6K 87K 0 processC/1
25571 root     5.1 0.7 0.0 0.0 0.0 0.0  93 0.0  57  68  5K   0 processB/1
 2003 root     0.3 1.2 0.1 0.0 0.0 0.0  97 1.9  67   8  1K  21 processA/1

prstat
PID USERNAME SIZE RSS STATE PRI NICE TIME CPU PROCESS/NLWP 2003 root 39M 38M sleep 59 0 3:05:58 0.6% processA/1
25571 root     6488K 5752K sleep   13    0   0:00:00 0.5% processB/1
25952 root     5208K 4184K sleep   23    0   0:00:00 0.1% processC/1

Is it possible to relate the CPU usage shown by prstat and prstat -m ? or am i comparing apples and oranges here ? process C shows 0.1% CPU usage (is this only usr time?) and microstate accounting shows 30% user time, 30% sys time and 30% sleep. How does one relate the two ? The above is just snapshots taken almost at the same time. My testing is normally done in a 2 hr time frame. So i could run prstat with 10 sec sampling rate and prstat -m with also a 10 sec sampling rate during the same period. The sampling instance for both maybe slightly off. If i average out during this run for each process from prstat and also the same from prstat -m, how do i relate these values ?


This message posted from opensolaris.org
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org


_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org

Reply via email to