We often want to calibrate a performance model to measured data. The traditional way to do this is to measure the CPU demands directly by profiling. However this is difficult, costly, and cannot be used to maintain the model as the software evolves or it installed in new environments.
We want to calibrate the model parameters to fit to performance test results (performance measurements). Standard regression techniques and be used for this, and the seminar will explain them, and show the results. Basically, sensitive parameters are estimated with better accuracy. A MATLAB program is available to experiment with regression fitting.
When parameters are changing over time, an iterative tracking filter (Kalman filter) can be used to update the parameter estimates, and this will be briefly indicated also.
These techniques can be applied to any kind of performance model.