Speaker:Jahid Hasan

Multi-Server Oblivious RAM Approach for Secure Cloud Storage

This thesis presents an empirical research on how to apply typical single-server ORAM models to multi-server cloud storage to improve performance and security in large-scale storage systems. Our method used a node-splitting mechanism to decrease the storage tree size and the strain on each server by dividing original blocks into smaller ones. Maintaining high parallelism across multiple servers reduces query latency while increasing system performance, especially for large block sizes. To improve security, we use a double encryption approach and piece-wise eviction processing to ensure data confidentiality and access pattern privacy. Our experimental results verify the performance improvement brought by the multi-server design. Specifically, as the block sizes range from 1 KB to 32 KB and the number of servers increases from 4 to 12, the size of data needed to be stored by each server drops from 10 GB to 110 MB, with block sizes ranging from 85 to 2730 bytes. With these settings, the system attains the data access throughput between 195.29 KB/s and 5101.32 KB/s, as well as the average query latency between 0.01 ms and 0.025 ms per block.

Committee: Wensheng Zhang (major professor), Nicole Hashemi, and Ying Cai

Join on Zoom: https://iastate.zoom.us/j/99950731906