A Prefetching Strategy Based on LMS Rule

Existing All-Kth-Order Markov models for Web prefetching only consider one or a few important characteristics such as the length of access sequence,transition probability,page access frequency,prediction precision and output probability distribution. In order to make use of these characteristics,we present a new prefetching strategy based on the LMS (least mean square) rule by extending All-Kth-Order-Markov model. This approach defines a linear function for every state to represent its prediction capabilities and makes use of the maximum entropy to describe a state's outgoing probability distribution. In order to improve prediction precision,the function's weight values are dynamically updated according to the LMS rule. The trace-driven experiment shows that our method has a better prefetching performance.