Sanity check

Greetings, A colleague and I are doing a rough bit of performance analysis on an 8-way RS6000, in particular trying to determine the number of CPU seconds that must be attributed to a non-multithreaded process. Here's what I did to measure CPU : 1. I took measurements from single-thread tests, ie there is only one user on the system executing an application that makes calls to Informix and the background load is low enough to ignore. 2. For the duration of the test, I used the AIX command iostat at 1-second intervals to determine the CPU (user + system) % utilisation 3. I sum all of the per-second utilisations from step (2). 4. Dividing the result in (3) by 100 gives me the number of CPU seconds used by this test. Now ... an 8-way processor is capable of 8 CPU seconds per elapsed second. This means that a 100% busy system is actually clocking up 8 seconds of processor time every second. So .... the "pure" number of CPU seconds used by this test is the result from (4) multiplied by 8. Does this appraoch seem reasonable ? I've convinced myself that it is - but I have a few nagging doubts. Any input welcomed. Many thanks John PS : I've also just started looking at the syscpu and usercpu results from Informix's onstat -g glo. These figures tie-in well with the results from (4), so I guess that the approach is not too far off.