LFC-K Cache Replacement Algorithm for Grid Index Information Service (GIIS)

Traditional cache replacement algorithms are not easily applicable to a dynamic and heterogeneous environment. Moreover, the frequently used hit-ratio and byte-hit ratio are not appropriate measures in grid applications, because non-uniformity of the resource object sizes and non-uniformity cost of cache misses in resource information traffic. In this paper, we propose a Least Frequently Cost cache replacement algorithm based on at most K backward references, LFC-K. We define average retrieval cost ratio (ARCR), as the cost saved by using a cache divided by the total retrieval cost if no cache was used. We compare performance of LFC-K with other caching algorithms using ARCR, hit-ratio and byte-hit ratio as performance metrics. Our experimental results indicate that LFC-2 outperforms LRU, LFU and LFU-2.