Causality Domains: Scaling Clock Vectors for Open Distributed Processing

Clock vectors provide the information required to determine the possible causality and hence concurrency of events in a distributed computation. The causality of any pair of events can be determined by comparing event timestamps based on clock vectors. One of the major diiculties of deploying clock vectors in large-scale distributed systems is that they must be transmitted with all messages between processes, and they grow linearly with the number of processes in the system. This thesis uses the concept of causality domains to reduce the size of clock vectors and support the use of event timestamps in open distributed processing. A causality domain is an abstraction of a set of processes that exhibits process-like behaviour to processes outside the set. Causality domains provide an approximation of causality with well deened semantics. With appropriate structuring of domains, the clocks resulting from this approximation grow logarithmically with the number of processes|a signiicant reduction in size. Causality domains also provide a basis for encapsulation of behaviour to support clock vectors in open systems. This thesis describes existing theory of clock vectors, introduces and formalises the new concept of causality domains, and provides algorithms for several alternative implementations of causality domains in a distributed system.