Comment # 4
on bug 1044
from Ken McDonell
After modifying pmdumplog to report extended timestamps (Ddd DD Mmm
HH:MM:SS.sss YYYY) with a new -x option, and fixing pmlogextract to correctly
include <mark> records in the output archive (this was a bug), I now have a
test case that uses a much smaller archive ...
kenj@bozo-vm:~/tmp$ time pmval -t 60 -a slow pmcd.numagents >/dev/null
real 0m24.469s
user 0m19.921s
sys 0m4.532s
kenj@bozo-vm:~/tmp$ time pmval -t 60 -a slow hinv.ncpu >/dev/null
real 0m0.242s
user 0m0.232s
sys 0m0.004s
kenj@bozo-vm:~/tmp$ time pmval -t 60 -a slow kernel.all.nprocs >/dev/null
real 0m0.227s
user 0m0.223s
sys 0m0.004s
kenj@bozo-vm:~/tmp$ pmlogsummary -y slow
pmcd.numagents 5.000 661 none
hinv.ncpu 4.000 22056 none
kernel.all.nprocs 617.961 21392 none
kenj@bozo-vm:~/tmp$ pmdumplog slow | grep mark | wc -l
663
kenj@bozo-vm:~/tmp$ pmdumplog -x slow | grep mark | tail -20
Fri Jan 24 19:35:34.486 2014 <mark>
Fri Jan 24 19:56:06.572 2014 <mark>
Fri Jan 24 20:35:26.576 2014 <mark>
Fri Jan 24 21:11:58.682 2014 <mark>
Fri Jan 24 21:31:32.020 2014 <mark>
Fri Jan 24 22:06:52.404 2014 <mark>
Fri Jan 24 23:06:21.748 2014 <mark>
Fri Jan 24 23:10:02.467 2014 <mark>
Fri Jan 24 23:19:16.265 2014 <mark>
Sat Jan 25 00:21:45.616 2014 <mark>
Sat Jan 25 00:37:29.268 2014 <mark>
Sat Jan 25 00:40:47.391 2014 <mark>
Sat Jan 25 00:59:58.838 2014 <mark>
Sat Jan 25 01:37:22.873 2014 <mark>
Sat Jan 25 09:16:52.833 2014 <mark>
Sat Jan 25 09:20:10.371 2014 <mark>
Sat Jan 25 09:23:20.100 2014 <mark>
Sat Jan 25 09:26:31.592 2014 <mark>
Sat Jan 25 09:54:43.677 2014 <mark>
Sat Jan 25 09:57:57.201 2014 <mark>
What this shows is that it is not the SIZE of the archive that is the problem,
it is the large number of <mark> records in the archive and metrics that are
logged ONCE each time pmlogger starts ... 661 instances of pmcd.numagents and
663 <mark> records.
This is just a status update ... I'm onto the real debugging now ... 8^)>