Utilizations for NT on SMP machine

Steven P. Woolet (woolet@raleigh.ibm.com)
Wed, 18 Mar 1998 12:59:57 -0500

I'm running on two dual-processor SMP machines. I didn't realize that
there was a parameter to indicate number of processors, so the first set
of runs assumed the default of 1. Utilizations for the local machine
appeared to "bounce" around quite a bit between successive runs, but the
remote utilization appeared very consistent.

After realizing that I needed to tell netperf that there were 2
processors, I reran the LOC_CPU and REM_CPU tests using the -n
parameter. I typically run these tests 5 times, just to see if
everything is stable. The runs on the local machine gave much lower
numbers than those without the -n parameter and there was quite a bit of
variation between the runs (the numbers dropped from values in the
242000 to 243000 range without the -n to values in the 203000 to 205000
range with the -n parameter). The remote machine numbers remained at
the same level as those without the -n parameter and there was very
little variation between the runs (numbers in the 242000 range with
variation over all 5 samples of only 30).

This brings up my questions: Does netserver need to be started with the
-n parameter to get the remote machine to know that it has
multiprocessors? How does the -n parameter to netperf indicate whether
there are MP's on the local machine versus the remote machine (i.e. what
if the local machine is a uniprocessor but the remote machine is an
SMP)?

Steve Woolet