Step 7:
Validate analysis results

Before the results of any performance analysis are announced, they should be validated. The validation step answers the question:

Do we believe these results?

Validation is a joint effort between the performance analyst and developers of the product. Data from one analysis method should be compared to results from a second method. Model results are checked by comparison to performance measurements, while measurements are checked against model predictions. In addition, other checks should be made as well:

• Are the results for different components of the system consistent? For example, if a simulation of a network device (that doesn't do fragmentation or reassembly) reports that number of packets sent is twice the number received, changes are good there is a logic error somewhere in the simulation. In many cases, perfect consistency is unrealistic, but any gross discrepancies should be investigated.
• Are the qualitative trends as expected? For example, increasing the time needed to install new software in an availability model should decrease system availability
• Are the results in line with intuitive predictions? If not, be open to the possibility that the model or measurements are right and the intuition is not.

Once the results have been validated, it is safe to analyze the data completely and present them to the appropriate audiences.