Training of NNPCR-2: An improved neural network proxy cache replacement strategy

Proxy servers are designed with three goals: decrease bandwidth, lessen user perceived lag, and reduce loads on origin servers by caching copies of Web objects. To achieve these goals an efficient cache replacement technique should be utilized. Squid is a widely used proxy cache software. Squid's default cache replacement strategy is least recently used. While this is a simple approach, it does not necessarily achieve the targeted goals. We use a different approach to address the cache replacement problem by training neural networks to make cache replacement decisions. In this paper we present the many improvements to our neural network proxy cache replacement strategy. We focus on the training of the neural networks and demonstrate the results for the effect of the number of hidden nodes, input node, the sliding window length and the learning rate on the neural network.