Scalable Storage Systems

Petabyte and Exabyte storage systems are a reality. These systems introduce unique challenges to the systems architect because of their size and unique requirements. In this thesis, I suggest that access patterns to very large storage systems are long-tailed distributions. I explore three live systems and show that each of them has a very strong long-tailed access distribution. Based on this finding, I prove there is no operational benefit to be gained by applying a directed placement policy allocating items to storage units. I conclude this thesis with a discussion of impacts and future research opportunities based on these findings, including large scale caches, backup techniques and distributed databases.