A Reusable, Distributed Repository for Configuration Management Policy Programming

Abstract

Although the number and variety of available configuration management (CM) systems has grown rapidly in the past few years, the need to construct new CM systems remains. The desire to manage different kinds of artifacts other than source code, situations demanding highly specialized solutions, and the exploration of new research questions all may require the construction of a novel CM system. Unfortunately, in the face of today's move towards distributed projects, this is becoming an increasingly daunting task for which existing CM technology provides little to no support. This dissertation contributes a novel reusable testbed that supports the rapid development of—potentially distributed—prototype CM systems. The testbed separates CM repositories from CM policies by providing a generic model of a distributed repository and an associated programmatic interface. Together, the repository model and programmatic interface stipulate a precisely defined abstraction layer upon which specific CM policies are built. In particular, CM policies are programmed as unique extensions to the interface, while the underlying distributed repository is reused across different policies. Within the abstraction layer, distribution is isolated. Low-level details of distributed programming are placed within the implementation of the repository model whereas distribution aspects that are controlled at the policy programming level are placed in a separate, orthogonal functional category within the programmatic interface. Two tangible benefits result from the use of the reusable testbed. First, the effort required in constructing prototype CM systems is reduced significantly because the generic repository is reused and the CM policy is easily implemented. Second, the rapid exploration of new CM policies is enabled, leading to the creation of unique CM policies that are tailored to specific situations. The testbed is evaluated abstractly, by mapping ten CM policies onto the repository model and programmatic interface. Additionally, it is evaluated concretely through the use of a prototype, called NUCM, upon which three novel CM policies are implemented. Demonstrating the expressiveness, feasibility, utility, and validity of the testbed, these policies are characterized by their rapid development, ease of change, incremental evolution, and seamless distributed operation.