Visualization and Analysis of Parameter Optimization Logs

Goal

Heuristic optimization algorithms try to find the best solution for a problem (or an approximation for it) in a specified time frame. Such a problem is to find a good garbage collector configuration for a given scenario, i.e., application and work load. In this case, the optimizer repeatedly executes the scenario and changes parameter values in order to maximize the throughput, i.e., minimize the overall garbage collection time.

However, the reasons for parameter changes made during the optimization process remain a mystery to the outside observer, leaving him with the final result only. The goal of this thesis is to develop a tool that visualizes different aspects of the optimization process based on the created log files. It must show the path the optimizer has taken through the n-dimensional solution space (for n <= 2), show the performance depending on a single parameter, and identify parameters having the biggest effect on performance. Furthermore, the parser must be easily exchangeable in order to support log formats of other optimizers.