Distributed storage allocation problems

Abstract

We investigate the problem of using several storage nodes to store a data object, subject to an aggregate storage budget or redundancy constraint. It is challenging to find the optimal allocation that maximizes the probability of successful recovery by the data collector because of the large space of possible symmetric and nonsymmetric allocations, and the nonconvexity of the problem. For the special case of probability-1 recovery, we show that the optimal allocation that minimizes the required budget is symmetric. We further explore several storage allocation and access models, and determine the optimal symmetric allocation in the high-probability regime for a case of interest. Based on our experimental investigation, we make a general conjecture about a phase transition on the optimal allocation.