Chatz, are you still there?
I was trying to use pmdumptext to validate the derived metric expressions
for kernel CPU times expressed as a percentage, and could make no sense of
the numbers.
Looking at the code I notice this in checkUnits() ...
// Only scale units if interactive and not raw
if (rawFlag || !niceFlag)
return;
Now this just looks wrong ... I can see no reason for -i enabling scaling
(especially for time utilization) ...
Can anyone explain the rationale for the current code, before I go change
it?
The example below shows that pmdumptext -I produces numbers in the same
range as pmval, but pmdumptext without -I produces numbers that depend on
the units of kernel.all.cpu.user from the kernel PMDA.
kenj@bozo-vm:~/tmp$ pmval kernel.all.cpu.user
metric: kernel.all.cpu.user
host: bozo-vm
semantics: cumulative counter (converting to rate)
units: millisec (converting to time utilization)
samples: all
0.0
1.995E-02
1.995E-02
9.984E-03
9.986E-03
^C
kenj@bozo-vm:~/tmp$ pmdumptext kernel.all.cpu.user
Sat Jun 28 13:44:55 ?
Sat Jun 28 13:44:56 9.995
Sat Jun 28 13:44:57 0.000
Sat Jun 28 13:44:58 10.014
Sat Jun 28 13:44:59 9.991
Sat Jun 28 13:45:00 10.000
Sat Jun 28 13:45:01 10.011
^C
kenj@bozo-vm:~/tmp$ pmdumptext -i kernel.all.cpu.user
Sat Jun 28 13:45:13 ?
Sat Jun 28 13:45:14 0.01
Sat Jun 28 13:45:15 0.02
Sat Jun 28 13:45:16 0.01
Sat Jun 28 13:45:17 0.01
|