LPS-ORAM: Perfectly Secure Oblivious RAM with Logarithmic Bandwidth Overhead

Abstract

Oblivious Random Access Machine (ORAM) is a cryptographic tool used to obfuscate the access pattern. In this paper, we focus on perfect security of ORAM. A perfectly secure ORAM is an ORAM that can resist against an adversary with unlimited computing power, and the failure probability of ORAM is zero rather than negligible. Since all existing perfectly secure single-server ORAM solutions require at least sublinear worst-case bandwidth overhead, we pose a natural and open question: can we construct a perfectly secure single-server ORAM with logarithmic worst-case bandwidth overhead? In this paper, we propose the first tree-based perfectly secure ORAM scheme, named LPS-ORAM. To meet the requirements of perfectly secure ORAM, two techniques are presented. One technique is dynamic remapping associated with a mutable scope, and the other is dynamically balanced eviction. Their combined effect allows the root bucket to never fill up while maintaining its statistical security in tree-based ORAM. In the worst case, our solution achieves logarithmic bandwidth overhead. Therefore, our solution answers the open question in the affirmative. In terms of overhead for temporary storage on the client side, compared with the latest perfectly secure ORAM solution, our solution is reduced from sublinear to logarithmic, and even if the server storage overhead scales lightly, it is still at the same level of quantity as the state of the art. Finally, the evaluation results show that our LPS-ORAM has a significant advantage in terms of bandwidth overhead and overhead for temporary storage on the client side.