Commodity Cache Management for Storage Server

Abstract

Commodity mediums have been used as front-end cache to improve the performance of storage server during the period of high overload in distributed file systems. Comparing with traditional cache media such as SDRAM, the speed of commodity medium cache is much lower, which affects traditional cache management algorithm based on high speed cache framework. In this paper, we propose a novel cache management algorithm – C-Aware, which considers the effect of the speed characteristic of cache mediums and data source on the whole system performance. By tracing the history response times of accessing cache and source, C-Aware algorithm adaptively decides whether to cache data according to current running environment, and achieves good performance regardless of the server is busy or not. Our experiments show that C-Aware gets near 80% improvement compared with traditional methods when the cache size is half of total test data set and the server is not busy. It still presents comparable performance when there is high workload on server side.