As part of planning configuration changes, you can also consider your
service level objectives or quality of service goals.
For example, as shown in Figure 2–5, CPU utilization has one peak
at 8 CPUs, with many lower peaks. If you configure your system to always meet
the demand of this single 8–CPU peak, you require 100% capacity all
the time, you must keep 8 CPUs in your configuration, and you cannot add new
processes.
| 1 | Peak value over the examined time interval.
Value for 100% utilization. |
| 2 | Value for 99.9% utilization. |
| 3 | Value for 99% utilization. |
This utilization graph shows three months of data. As there are 288
data points per day, there are about 26,000 data points in the quarter shown.
The 99.9th percentile, the value for which 99.9% of the data points are below
that value, is the 26th highest reading; the 99th percentile is the 260th
highest reading. Thus, if you set your requirement at the 99% percentile (260th
highest peak), five CPUs might be sufficient. With Capacity Advisor's visualization
and reporting tools, you can make a considered estimate of server utilization
using different scenarios. Such estimates can help you minimize overspending
for capacity you use rarely, and maximize utilization of your systems to ensure
that your systems have the capacity needed at the level you require.
Additionally, with utilization graphs and associated reports, you can
identify how to optimize use of your servers by examining your quality of
service requirements. Capacity Advisor reports provide additional detail to
connect utilization graphs with specific quality of service metrics. For additional
information, see the report examples in “Producing Graphs and Reports” later
in this document.