Managing Buffer Cache by Block Access Pattern

As buffer cache is used to overcome the speed gap between processor and storage devices, performance of buffer cache is a deciding factor in verifying the system performance. Need of improved buffer cache hit ratio and inabilities of the Least Recent Used replacement algorithm inspire the development of the proposed algorithm. Data reuse and program locality are the basis for determining the cache performance. The proposed algorithm determines the temporal locality by detecting the access patterns in the program context from which the I/O request are issued, identified by the program counter signature, and the files to which the I/O request are addressed. For accurate pattern detection and enhanced cache performance re-reference behavior exploited in the cache block are associated with unique signature. Use of multiple caching policies is supported by the proposed algorithm so that the cache under that pattern can be best utilized.