CPU Configuration Guidelines for vPars

7
partition is available to the application. The variable factor in this experiment will be the CPU’s
available to execute the workload.
The creation of the real time processor set is shown in Figure 6. Before beginning the test, “top” is
started and the file system and disk stress workloads are applied to the 8 CPU system. When the
workload has reached a steady state, determine the CPU utilization by subtracting the CPU IDLE from
100%. In Figure 7, note that with 8 CPU’s, the IDLE field shows about 85% which translates to 15%
CPU utilization. Use the “psrset” command as shown in Figure 8 to add CPU 7 and CPU 6 to the real
time processor set (pset 1).
Now run the file system and disk stress on the 6 remaining CPU’s of the default processor set and
note the CPU utilization in Figure 9. The idle percentage for each CPU other than those in the real-
time processor set (CPU 6 and CPU 7) must be summed, averaged, and subtracted from 100 to
calculate the CPU utilization. The CPU utilization for this run must be summed across the 6 CPU’s in
the default processor set and then averaged to come up with 53.5% idle which translates to 46.5%
utilized. The calculation for average CPU utilization is as follows.
%idle = ((64.8% + 42.9% + 72.6% + 69.6% + 37.4% + 33.6%)/6) = 53.5%
%utilization = 100% - 53.5% = 46.5%
Add two more CPU’s to the real-time processor set. Now run the file system and disk stress on the 4
remaining CPU’s of the default processor set. The CPU utilization for this run must be summed across
the 4 CPU’s in the default processor set and then averaged (see Figure 10) to come up with 37.6%
idle which translates to 62.4% utilized. The CPU utilization is above but close to the stated goal and
adding another CPU from the real time processor set would cause the CPU utilization to drop below
the stated goal. The steady state load with approximately 60% CPU utilization can be achieved with
4 CPU’s.
Figure 6. Create a real time processor set to isolate the excess CPU’s from processing the workload.