One of the main issues in computer system management is planning to support peak usage demands. This is particularly important for customer-facing systems, where performance problems due to overload could materially affect the business. The difficulty is in knowing how big the peaks are going to be, so that you can plan to have enough capacity in place to support them.

Typically, the total CPU (or other resource) usage is "driven" by the workload. However, the relationship is rarely as simple as "double the workload, double the CPU use". A small change in a key workload can have a large effect on total processing costs. This is where Correlation Analysis comes in. It lets you determine the extent to which different quantities vary in sympathy with one another. Then, if you can predict a change in one quantity (such as a small increase in a key workload), you can determine the likely knock-on effect in the other quantity (such as total processing cost). Knowing that two metrics are correlated, lets you make this kind of prediction with confidence.

Doing Correlation Analysis by hand is not recommended unless you are a statistician! However there are many tools available that can help to determine correlations. Metron's Athene package will enable the user to make such checks against any of the data stored in its performance database with expert assistance in the ensuing analysis.