Combating state explosion in the detection of dynamic properties of distributed computations

Abstract

In the context of asynchronous distributed systems, many important applications depend on the ability to check that all observations of the execution of a distributed program, or distributed computation, satisfy a desired (or undesired) temporal evolution of states, or dynamic property. Examples include the implementation of distributed algorithms, automated testing via oracles, debugging, and building fault-tolerant applications through exception detection and handling. When a distributed program exhibits a high degree of concurrency, the number of possible observations of an execution can grow exponentially, quickly leading to an explosion in the amount of space and time required to check a dynamic property. In the worst case, detection of such properties may be defeated. This is the run-time counterpart of the well-known state explosion problem studied in model checking. In this thesis, we study the problem of state explosion as it arises in the detection of dynamic properties. In particular, we consider the potential of applying well-known techniques for dealing with state explosion from model checking to the case of dynamic property detection. Significant semantic similarities between the two problems means that there is great potential for deriving techniques for dealing with state explosion in dynamic property detection based on existing model checking techniques. However, differences between the contexts in which model checking and dynamic property detection take place mean that not all approaches to dealing with state explosion in model checking may carryover to the run-time case. We investigate these similarities and differences and provide the development and analysis of two approaches for combating state explosion in dynamic property detection based on model checking methods: on-the-fly automata theoretic model checking, and partial order reduction.