Application of neural networks to cache replacement

In this paper, we propose neural network-based schemes for the replacement of cache lines in computer systems. Our research indicates that Neural Networks (NN) have great potential in the area of statistical prediction [1]. This attribute of neural networks was used in our work to develop neural network-based replacement policies, which can effectively eliminate dead lines from the cache memory by predicting the sequence of memory addresses referenced by the Central Processing Unit (CPU) of a computer system. The proposed strategies will, therefore, provide better cache performance as compared to the conventional LRU (Least Recently Used) scheme. In fact, we observed from the simulation experiments that neural network-based replacement schemes provide excellent performance as compared to the ubiquitous LRU algorithm. This new approach can also be applied to the page replacement and prefetching algorithms in virtual memory and shared-memory multiprocessor systems.