The notion of knowledge-based program introduced by Halpern and Fagin provides a useful formalism for designing, analysing, and optimising distributed systems. This paper formulates the two phase commit protocol as a knowledge-based program and then an iterative process of model checking and counter-example guided refinement is followed to find concrete implementations of the program for the case of perfect recall semantic in the Byzantine failures context with synchronous reliable communication. We model several different kinds of Byzantine failures and verify different strategies to fight and mitigate them. We address a number of questions that have not been considered in the prior literature, viz., under what circumstances a sender can know that its transmission has been successful, and under what circumstances an agent can know that the coordinator is cheating, and find concrete answers to these questions. The paper describes also a methodology based on temporal-epistemic model checking technology that can be followed to verify the shortest and longest execution time of a distributed protocol and the scenarios that lead to them.
Download