Minimizing the cost of configuration changes in self-managing systems via closures, matchings, and marriages

The transition planning problem for system configuration management is to determine, given a current system state and a set of new requirements, how to best satisfy those requirements. In this thesis, we draw upon several approaches to solving this problem, including the theory of closures in configuration management, and graph algorithms including bipartite matching and stable marriage, in order to provide minimum-cost or distributed solutions. We begin by designing and implementing an "IP Address Closure" that manages IP address assignments for a network. This prototype proves that such a closure is feasible, but there is no good solution for handling infeasible transition requests. We next develop a theory of exceptions that could be used to deal with infeasible transitions, but conclude that infeasible requests are best handled through virtualization, which allows transitions to be expressed as role mappings. Transitions between mappings can be computed via a series of polynomial reductions to a bipartite matching problem, whose solution is approximable in a distributed environment via the stable marriage algorithm. This technique shows promise for transition planning in distributed environments with no centralized control.