Proofs of Retrievability with Low Server Storage
Abstract
We investigate a novel approach to Proofs of Retrievability (PoR), protocols that allow a client to audit the cloud server storing its data remotely. These protocols allow a means of efficiently ensuring that all of the data the client believes to be stored in the cloud is still able to be retrieved by the cloud server, instead of relying on trust alone in the current model. Past PoR approaches have worked toward computational optimization for the audit; however, this requires a large amount of overhead persistent storage (up to 10x the actual database size). Our new approach instead trades higher computation for significantly decreased persistent storage. As all major cloud providers charge markedly more for storage than for computation, our new protocol offers practical efficiency. Our approach rests on treating the data as a square matrix, comparing randomized linear algebra identity tests over the matrix at the time of last check and at the current time. Honest retrieval of data, enforced through a Merkle hash tree requiring negligible extra persistent storage, and dynamic updates are supported in our approach. While audit computation now scales linearly, the required persistent storage is only 1.068x the size of the data. We demonstrate its efficiency in practice with a deployment on Google Cloud Compute Engine with test case data size of 1TB.Our approach costs $42.72 per month for storage, and an audit costs $0.23 taking 16 minutes. Previous state of the art requiring 6x storage of the data size costs $240 per month. This is a 82 percent cost savings from storage while hosting the data in the cloud. We parallelized the computation of the audit across multiple virtual machines using MPI in order to increase the I/O-bound run time performance, which resulted in a near-linear speed up. We are investigating further optimizations on client-side storage and communication costs, as well as how to deploy our approach over an entire block device.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 06, 2020
- Accession Number
- AD1136698
Entities
People
- John M. Hanling
Organizations
- United States Naval Academy