Consistent records in asynchronous computations

Abstract.

A global record (i.e. a set of local records, one for each process of an asynchronous computation) abstracts what is usually called global state, global checkpoint or global snapshot in particular problems. Consistent global records are important in many applications. This paper addresses the following question: “Given a set of local records, is it possible that these records belong to the same consistent global record?”. This question has been answered by Netzer and Xu in the particular context of message passing systems. This paper extends their result to a very general asynchronous computational model that encompasses shared memory systems and various message passing systems (with reliable or unreliable and point-to-point or multicast or broadcast communications).