sec-cs: Getting the Most out of Untrusted Cloud Storage

Abstract

We present sec-cs, a hash-table-like data structure for contents on untrusted storage that is provably secure and storage-efficient. We achieve authenticity and confidentiality with zero storage overhead using deterministic authenticated encryption. State-of-the-art data deduplication approaches prevent redundant storage of shared parts of different contents irrespective of whether relationships between contents are known a priori. Instead of just adapting existing approaches, we introduce novel (multi-level) chunking strategies, ML-SC and ML-CDC, which are significantly more storage-efficient than existing approaches in presence of high redundancy. We prove sec-cs's security, publish an implementation, and present evaluation results indicating suitability for, e.g., future backup systems that should preserve many versions of files on little available storage.