Formal Proof of Impossibility of Reliability in Crashing Protocols

In a strictly asynchronous system with process failures, it has been known that distributed consensus is impossible [FLP85]. It also has been shown that without persistence, no data link layer can work correctly (this includes all the well known and widely used protocols such as HDLC, etc) [FLMS93]. This work has been extended recently to study the fault span of crash failures [JV00]. In this paper, we present a formal proof of the non-existence of correct crashing network protocols with either unreliable FIFO, reliable FIFO or reliable non-FIFO links using the Input/Output Automata formalism in PVS, a verification system based on higher-order logic.