100% of _one_ CPU actually, on a multiprocessor system.
That is, if I run the distributed.net client (which
is multithreaded and sucks up all of ALL the CPUs, but
at a low priority so as not to compete with anything else)
on a two CPU system where nothing else is going on, I could
see something like the following:
$ ps -p 498;sleep 10;ps -p498
PID TTY TIME CMD
498 ? 10821:47 rc5des
PID TTY TIME CMD
498 ? 10822:07 rc5des
Note that it used 20 seconds of CPU time (that is, 10
seconds times two CPUs) in 10 seconds of clock time.
Another implication is that 1 second of CPU time on
a system with a fast CPU gets more done than 1 second
of CPU time on a system with a slower CPU. That could
get interesting if one were billing based on CPU time
across multiple, different speed systems. One might
want to charge more for CPU time on the faster system.
The problem is that a "fair" ratio would be difficult to
arrive at; the relative speeds of different systems can
depend on the particular type of activity being compared.
And billing based on an amortization of the system may not
make sense either, since by whatever benchmark one compares,
cost per unit work drops as systems get faster.
--Quote:> Say I have TIME=00:20, does it mean the process consumed as much CPU as if
> it was running with 100% of the CPU during 20 seconds ?
Bogonics: the primary language inside the Beltway
1. How to hide from 'who', 'finger', 'ps' commands?
How to hide from 'who', 'finger', 'ps -u' commands?
3. Limiting 'w', 'top' and 'ps'
4. Problem with MOL on Suse 7.0
5. Help with 'user', 'w', 'who' commands
7. Piping results of a 'find' command to a 'mv' command...
8. Please help! How to get environment variables into pkgadd?
9. Process START times (ps aux) don't agree with 'date'
10. Request headers - 'Reload', <SHIFT>+'Reload'
11. 'last' command doesn't show complete login times
12. How do I re-format the 'time' command's output?
13. 'time' command doesn't display memory usage