M-ORAM: A matrix ORAM with Log N bandwidth cost

Oblivious RAM can hide a client’s access pattern from an untrusted server. However current ORAM algorithms incur large communication or storage overheads. We propose a novel ORAM construction using a matrix structure for server storage where a client downloads blocks from each row, choosing the column randomly to hide the access pattern. Both a normal and recursive construction are presented, achieving bandwidth cost of O(1) and O(logN), respectively, and client storage similar to Path ORAM. We show under the same conditions, our matrix ORAM reduces bandwidth cost compared to Path ORAM by log N/2.