Download files
Abstract
The rapid growth in the processing capability and the widespread of personal and mobile computing devices results in a corresponding growth in services that rely on collaborative computing. A number of such services rely on a combination of devices of different capabilities with varying level of trust and security, to process user's sensitive (personal or business) data. To achieve this, both privacy and trust mechanisms need to be an integral part of collaborative computing systems. This thesis addresses the problems of achieving privacy preserving computations and trust in heterogeneous collaborative environments with colluding and malicious parties.
The first contribution is a trust-based collaboration partner selection mechanism, where trust relates to the expectation of a node's behaviour during collaboration. We propose a trust evaluation mechanism, which continuously evaluates the trust value during collaboration, and a probabilistic selection mechanism, where the selection probability is proportional to a node's trust value. The evaluation in a mobile content sharing context shows that the combined mechanism provides load balancing, while still ensuring a satisfactory collaborating service performance.
The second contribution relates to data privacy protection for computations in a distributed and collaborative environment, that can be achieved using secure multi-party computation (MPC). We present MPC protocols for heterogeneous environment where the computational load of a party is proportional to its resources and monitored performance. The evaluation shows that the proposed protocol increases system's robustness but incurs a higher overhead than standard MPC protocol where the data is uniformly shared across all parties.
With the third contribution, we consider extensions to MPC computations to achieve practical federated network outage monitoring system. Our proposed system privately aggregates measurement data from multiple Internet service providers (ISPs) to provide additional information useful for troubleshooting outages. The system uses multiset union MPC protocol based on counting Bloom filter. We evaluate the system's performance based on real network data from a medium-size ISP and the trade-offs resulting from the choice of system's parameters. We show that our proposed system significantly reduces the number of outages that require troubleshooting priority while performing in near real-time.