[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Bugfix request, re: 'time'-builtin: compute CPU% as man page documents
From: |
Linda Walsh |
Subject: |
Bugfix request, re: 'time'-builtin: compute CPU% as man page documents |
Date: |
Thu, 27 Nov 2008 16:38:40 -0800 |
User-agent: |
Thunderbird 2.0.0.18 (Windows/20081105) |
I use a short time format that displays the available data:
TIMEFORMAT='%2Rsec %2Uusr %2Ssys (%P%% cpu)'
I was running some timing tests under Cygwin/Windows (but similar exists on
SuSE-Linux).
The CPU% isn't display the correct value:
Examples:
15.78sec 21.32usr 10.77sys (100.00% cpu)
11.15sec 20.35usr 07.63sys (100.00% cpu)
8.37sec 13.77usr 5.42sys (100.00% cpu)
4.97sec 2.12usr 3.06sys (100.00% cpu)
---
In each of the above examples (last line from linux, obvious, huh? :-)).
The usr + sys values are > sec, so so why am I seeing a "faked-up value?"
The real %cpu should have numbers all over 100%.
From the manpage:
%P The CPU percentage, computed as (%U + %S) / %R.
I should see something like these for cpu%:
(203.36% cpu)
(250.94% cpu)
(229.27% cpu)
(103.26% cpu)
It would be more useful than fixing the manpage to say:
%P = percentage except when it would be over 100, in
which case we throw away the real value and display,
the completely wrong and un-useful value of 100.00%.
It tells me how much parallelism is going on, and also tells me how much
it might slow down on single-cpu machines. But it certainly isn't telling
me how busy any single cpu was (which was probably the context of
decision to hard-code 100.00% as the max).
:^? :)
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- Bugfix request, re: 'time'-builtin: compute CPU% as man page documents,
Linda Walsh <=