From: Brewer, Edward \(NIH/OD\) (BREWERE@OD.NIH.GOV)
Date: Mon Jan 31 2005 - 15:03:06 EST
To all,
I am trying to find a way to ascertain what contribution a single user is
making to the CPU load of a system over time. I have collect data from
previous days. I have a script or command line that will pull out all of
the cpusum data for every 30 seconds. From this data I can graph the
CPU+SYS+WAIT for the CPU load on the system for that day. Since I have the
process info in the same collect data, I wrote a awk script to pull out the
selected users %CPU value and summing them for an accumulated value. Then I
take this value and divided by the number of CPU's. At first the graphs for
several days had very similar shapes and gave me encouragement that this was
working. Then I ran across a day where for a sustained period for the
derived value for the user was greater than the total CPU load for the
machine. So I was stumped... I do see at times, there are some negatives
values (ones where the user load is higher than the total CPU usage) on the
other graphs, but there are spots where the CPU usage made a radical jump up
or down. I think that the way the original collect data is computed is an
average over the 30 second interval from the collected data and the
instantaneous value from the awk script that I use for the user is just
higher at times than the average for that interval. So I would expect this
to be a crude estimate. The long period where the user values from the
script was higher I can not explain unless my awk script is faulty. Does
anyone have any ideas.
TIA,
Lee Brewer
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:50:14 EDT