Hyper-Heuristics for Performance Optimization of Simultaneous Multithreaded Processors

In Simultaneous Multi-Threaded (SMT) processor datapaths, there are many datapath resources that are shared by multiple threads. Currently, there are a few heuristics that distribute these resources among threads for better performance. A selection hyper-heuristic is a search method which mixes a fixed set of heuristics to exploit their strengths while solving a given problem. In this study, we propose learning selection hyper-heuristics for predicting, choosing and running the best performing heuristic. Our initial test results show that hyper-heuristics may improve the performance of the studied workloads by around 2%, on the average. The peak performance improvement is observed to be 41% over the best performing heuristic, and more than 15% over all heuristics that are studied. Our best hyper-heuristic performs better than the state-ofthe art heuristics on almost 60% of the simulated workloads.