Comparing Self-Stabilizing Dining Philosophers through Simulation

We evaluate the performance of five of the most well-known selfstabilizing dining philosophers algorithms in a read/write atomicity model. The algorithms are evaluated under interleaving, power-set and synchronous execution semantics. For all algorithms, we compute latency and throughput of critical section access, as well as the number of safety violations. These metrics are computed under various load and fault scenarios. On the basis of the evaluation, we propose a combined algorithm that switches between different algorithms to achieve optimal performance.