Subject: | |
From: | |
Reply To: | |
Date: | Mon, 14 Mar 2005 15:30:46 EST |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
donna writes:
> > The CPU was used a total of 55.3% of its capacity during this
> > interval <GI01>
>
> 55% is not a problem
Actually, it probably is a problem.
I've given this lecture a hundred times before, and the only result that I
can see so far is that the ruler I use to whap people's knuckles only gets
larger with each whapping.
The two primary goals of any multiuser, multitasking machine are:
o to allocate resources to the individual users in the most fair way
possible
o to keep the CPU 100% busy
A 100% busy CPU is not a sin. It's a virtue.
If you are running multiple jobs at the same time and your CPU is not running
100%, it means that your processes are sitting around waiting on something
mechanical to complete (e.g., disc heads to move, main memory page swaps, etc.).
A 55% CPU utilization means that the CPU is just twiddling its thumbs for 45%
of the time, doing nothing.
If you can determine the source of those mechanical bottlenecks and get your
CPU usage up to 100%, you can improve the speed of your throughput by approx.
2 times, as measured by wall time.
But the same number of CPU seconds (more or less) will still need to be
exhausted by your code, whether the CPU runs at 100% or 10%. If you really want to
improve the performance of your jobs, take a very careful look at your code.
Almost all real performance gains occur because inefficient database accesses
or poorly implemented algorithms are cleaned up.
Wirt Atmar
* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *
|
|
|