SORT: A Similarity-Ownership Based Routing Scheme to Improve Data Read Performance for Deduplication Clusters

Existing data routing schemes developed for deduplication clusters have never addressed the data read performance, although it has been a well-known problem that the reads require non-trivial random disk seeks significantly affecting the data read performance in deduplication systems. In this paper, we propose SORT, a Similarity-Ownership based Routing scheme that exploits both the data similarity and ownership to improve the data read performance for deduplication clusters. Our experimental results fed with real-world datasets show that SORT reduces about 10% of random disk seeks while at the cost of only 0.11% of deduplication efficiency, achieving an optimal trade off between the deduplication efficiency and data read performance compared to other existing routing schemes. This result indicates that the exploration of data ownership to data routing schemes has the potential benefit in optimizing the data read performance for deduplication clusters.