I am sure we would all like to be able to compare the throughput ratings of one machine against another and get a satisfactory result. As I am sure you are aware, the reality is something totally different!

An exercise involving purpose-built capacity planning models will offer results with the best compromise between accuracy and effort spent to get predictions.

In outline one would model the current platform first using data collected from it. One can then adjust the parameters of this model to represent the target hardware environment. The results of the model will show device utilization at processor and I/O device level, as well as the predicted response times for the individual workloads.

This approach inevitably means that you have to make some assumptions about the ways in which the application will be implemented in the new environment. However, if it is also possible to take measurements from a Solaris machine running either a development version of the application, or to measure a Solaris machine running a similar application, the models can then be further tuned for accuracy. The differing performance characteristics of each platform can then be factored in to increase the accuracy of the predictions.

The process can also be refined at the development stage when there is a sample system on the target hardware to measure. Data can be captured at the test phase and a performance model built. This model can then be adjusted to reflect the live workload on the target environment.