Multiperspective reuse prediction.

The disparity between last-level cache and memory latencies motivates the search for efficient cache management policies. Recent work in predicting reuse of cache blocks enables optimizations that significantly improve cache performance and efficiency. However, the accuracy of the prediction mechanisms limits the scope of optimization. This paper introduces multiperspective reuse prediction, a technique that predicts the future reuse of cache blocks using several different types of features. The accuracy of the multiperspective technique is superior to previous work. We demonstrate the technique using a placement, promotion, and bypass optimization that outperforms state-of-the-art policies using a low overhead. On a set of single-thread benchmarks, the technique yields a geometric mean 9.0% speedup over LRU, compared with 5.1% for Hawkeye and 6.3% for Perceptron. On multi-programmed workloads, the technique gives a geometric mean weighted speedup of 8.3% over LRU, compared with 5.2% for Hawkeye and 5.8% for Perceptron.