Replica determinism and non-determinism

Abstract

This chapter is aimed at a fundamental discussion of replica determinism and non-determinism without resort to actual implementations and methodologies that enforce replica determinism. The first section introduces a definition of replica determinism. It will be shown that a generic definition of replica determinism can only be a framework. Any detailed or formal definition is strictly application specific. A refinement of the generic definition of replica determinism to certain application scenarios will be given. The second section lists possible modes for non-deterministic behavior of replicated servers. This list shows that replica non-determinism is introduced by standard functions such as on-line scheduling, timeouts or readings of replicated sensors. In the third section the fundamental limitations of replica determinism are discussed. It will be shown that it is impossible to achieve total replica determinism. By defining a cause consequence relation on the possible effects of replica non-determinism a notion of the atomic source for replica non-determinism is introduced. The important problem of how to characterize all possible atomic sources for replica non-determinism will be discussed. It will be shown that replica determinism, knowledge and simultaneity are closely related. The fourth section will consider the question whether there are replicated systems that do not require the enforcement of replica determinism. Two possible cases for such systems will be discussed. It will be shown, however, that any “real” non-trivial replicated system has to enforce replica determinism.