Anonymity and Trust in Large-Scale Distributed Storage Systems

Large-Scale Distributed Storage Systems (LS-DSSs) are at the core of several Cloud services. These externalized services may run atop multiple administrative domains. While a client may trust the organization that provides a given Web service, a single server may belong to another organization that the client does not trust. The design of a Distributed Storage System is itself a challenging task, in particular when scalability, availability and consistency are required. This thesis explores three important aspects within this context: anonymity and trust in LS-DSSs, auditing for Large-Scale Distributed Aggregation Systems (LS-DASs), and the evaluation of Distributed File Storage Service (DFSS) prototypes. The three systems proposed are evaluated using prototype implementations. We present SPADS, a system that provides publisher anonymization and rate limitation to any generic LS-DSS over a Distributed Hash Table (DHT), running on unreliable and untrustworthy peers. Then, we propose CADA, a system that sits on top of a generic LS-DAS, and detects when one or several servers attempt to bias an aggregation. The system performs probabilistic auditing, and raises suspicions based on the statistical deviation from an expected behavior. Finally, we study DFSSs. Performing a fair comparison of the File System Consistency Levels (FSCLs) as supported by existing DFSSs is difficult because these systems feature different base performance and optimization levels. We make an empirical comparison of these FSCLs by instantiating them into a novel DFSS testbed named FlexiFS.