As cloud storage continues to grow in usage and be adopted by new users with unique demands, it is increasingly relevant to ensure that access patterns are concealed as these may reveal sensitive information. Many research efforts have tailored traditional oblivious storage algorithms to the cloud setting. While these schemes provide an extremely high level of privacy, there is little or no research allowing them to support fault-tolerance. However, in order to meet the stringent availability demands of the cloud, a practical storage system must incorporate fault tolerance while also delivering a high level of performance.
This work looks to promote further research in oblivious storage systems by exploring fault-tolerant oblivious storage in the cloud setting. We study the use of quorum-based models in oblivious data storage systems. We then extend the open source implementation of the oblivious data storage system TaoStore to support these models. These models support configurable parameters that allow us to alter the tradeoff between availability and performance. We evaluate the performance trade-offs of this new system under various configurations through experimental results.