Novel cache replacement algorithms for high performance computer systems

Type Thesis

Rate 0.0 /5

Details

Files

Reviews

Download

Summary

This thesis presents novel cache memory replacement algorithms for high performance computer systems. The proposed algorithms use neural networks (NNs) for the purpose of guiding the line replacement decisions in caches. Key to our algorithms is the correct identification and displacement of inactive lines. This allows our strategy to provide better cache performance as compared to the conventional LRU (Least Recently Used), MRU (Most Recently Used), and FIFO (First In First Out) replacement policies. Performance of the neural network based algorithms was tested against the performance of existing popular strategies through extensive trace-driven simulations in order to obtain reliable results for a wide spectrum of practical cache configurations. Simulation results show that the proposed algorithms provide substantial performance improvement in the miss ratio over the conventional algorithms for several benchmark programs. Excellent performance of neural network based replacement strategies means that the new approach can be adopted as future cache replacement algorithms. Good results, provided by the line replacement algorithms, indicate that our schemes have potential for providing promising results when applied to the page replacement algorithms in virtual memory systems and disk caches.