A New Algorithm to Solve Synchronous Consensus for Dependent Failures

Fault tolerant algorithms are often designed under the t-out-of-n assumption, which is based on the assumption that all processes or components fail independently with equal probability. However, real systems may exhibit dependent failures. Cores and survivor sets are used to build an abstraction model for dependent process failures. Using this abstraction, we design an algorithm to solve consensus problem crash failures. Our algorithm uses the processes in all cores to broadcast messages. Each core reaches agreement separately and even simultaneously in some round with no failure in the core. In the worst-case, the decision can be made in the round that is equal to the size of the minimal core. Our algorithm guarantees that all processes eventually decide on the same value regardless the initial values. We prove the correctness of our algorithm and give the lower bound of the number of rounds to solve consensus problem.