RACE: A Robust Adaptive Caching Strategy for Buffer Cache

• Authors: Yifeng Zhu;Hong Jiang
• Venue: IEEE Transactions on Computers
• Year: 2008

While many block replacement algorithms for buffer caches have been proposed to address the well-known drawbacks of the LRU algorithm, they are not robust and cannot maintain an consistent performance improvement over all workloads. This paper proposes a novel and simple replacement scheme, called RACE (Robust Adaptive buffer Cache management schemE), which differentiates the locality of I/O streams by actively detecting access patterns inherently exhibited in two correlated spaces: the discrete block space of program contexts from which I/O requests are issued and the continuous block space within files to which I/O requests are addressed. This scheme combines global I/O regularities of an application and local I/O regularities of individual files accessed in that application to accurately estimate the locality strength, which is crucial in deciding which blocks are to be replaced upon a cache miss. Through comprehensive simulations on eight real-application traces, RACE is shown to significantly outperform LRU and all other state-of-the-art cache management schemes studied in this paper, in terms of absolute hit ratios. Specifically, it improves the absolute hit ratios of LRU, UBM, PCC and AMP by as much as 56.9%, 22.5%, 42.7% and 39.9%, with an average of 15.5%, 3.3%, 9.4% and 8.8%, respectively. Given the relatively high buffer cache miss penalties, which typically are six orders of magnitude higher than buffer cache hit times, these gains of hit ratios obtained by RACE are likely to imply significant performance gains in applications' response times as well.