CBR and neural networks based technique for predictive prefetching

Abstract

Cache prefetching in memory management greatly relies upon effectiveness of prediction mechanism to fully exploit available resources and for avoiding page faults. Plenty of techniques are available to devise strong prediction mechanism for prefetching but they either are situation specific (Locality of reference principle) or inadaptable (Markovian model) and costly. We have proposed a generic and adaptable technique benefiting from past experience by employing hybrid of Case Based Reasoning (CBR) and Neural Networks (NNs). Here we will be concerned with improving adaptation phase of CBR using NN and its impact on predictive accuracy for prefetching. The level of predictive accuracy attained (specifically in case adaptation of CBR) is ameliorated by handsome margin with declined cost than contemporary techniques as would be affirmed by results. © 2010 Springer-Verlag.