QUESTION:
Can the TPROF command be used to measure activity of the entire
system for a period of time, without specifying any one command to
monitor? ie, can I run TPROF during the execution of a long batch
application to see what is happening at particular time intervals
during the batch process?
Also, is there a rule of thumb to consider for filespace to be
consumed by the output of the TPROF command? Does it grow dynamically
larger output files if it is run for longer periods of time? Or
does it just give a finite set of results regardless of the duration
of the monitoring?
---------- ---------- ---------- --------- ---------- ----------
A: These are the answers to your questions:
1. Yes, TPROF command can be used to measure the activity
of the entire system for a period of time, without
specifying a particular command to monitor. This command
will give you a profile of current system activities.
To do this, follow the following steps:
mkdir tprof
cd tprof
cp /usr/bin/sleep .
tprof -p sleep -v -x sleep
After this series of executions, TPROF will profile the
sleep command along with all other processes. The
results will be found in the file "__sleep.all".
2. We set up the next example to show source code profiling while
running TPROF during the execution of a long batch application,
you can execute the application in one window, while running
TPROF in another one. We tested this in the following way:
Window 1: compiled source code of test.c with
the command
cc -qlist -g -o test test.c
then start the application program.
Window 2: ran TPROF with the command
tprof -p test -v -k -x sleep
The following is a sample output extracted from the __test.all
file:
First report: Individual process report (edited for length)
Process PID Total Kernel User Shared Other
======= === ===== ====== ==== ====== =====
wait 514 36085 36085 0 0 0
test 69471 1870 1676 24 170 0
..X11/bin/X 13269 855 459 183 213 0
xant 71308 709 301 68 340 0
aixterm 24741 533 102 57 374 0
.
======= === ===== ====== ==== ====== =====
Total 50124 46358 1195 2571 0
Second report: Overall totals by program name
Process FREQ Total Kernel User Shared Other
======= === ===== ====== ==== ====== =====
wait 1 36085 36085 0 0 0
bsh 767 3394 3211 179 4 0
test 1 1870 1676 24 170 0
aixterm 12 1636 673 157 806 0
/bin/test 446 1044 1007 2 35 0
...X11/bin/X 1 855 459 183 213 0
.
Note: These outputs were edited to reduce length.
Also, note that the source code "hot list" will be in
"__h.".
We recommend using 'tprof' for short time intervals (a few
minutes maximum at one time). The reason for this is because
'tprof' runs in 'trace', which uses memory buffers that fill up.
3. When we tested 'tprof', the output took approximately
700,000 bytes. It grows dynamically up to the time
when it runs out of resources (diskspace, memory in
the private data segment), and will give invalid
results. If 'trace' overruns its buffers, you can get
invalid results. Check the summary report total
column. The total number of clock ticks reported should
equal *100.
---------- ---------- ---------- --------- ---------- ----------
This item was created from library item Q655076 CQGFG
Additional search words:
CMD COMMAND CQGFG IX MAR94 MEASURE OP OZNEW PERFORMANCE RISCPERF
RISCSYSTEM SOFTWARE SPECIFICS SYS TPROF TUNE
WWQA: ITEM: RTA000040246 ITEM: RTA000040246
Dated: 03/1996 Category: RISCPERF
This HTML file was generated 99/06/24~12:43:15
Comments or suggestions?
Contact us