Abstract

In autonomic computing, systems manage their own operation and adapt to changes in the environment, with only limited human guidance. Further, autonomic computing supports pervasive computing, which removes the traditional boundaries for how, when, and where humans and computers interact. One area in which autonomic behavior can be particularly helpful is communication-related software; autonomic computing can be used to enhance quality of service in the presence of dynamic or adverse conditions. However, coordinating the actions of software components that are distributed across multiple platforms is a complex task. Developers of autonomic communication services can benefit from a software infrastructure to support coordinated actions among system layers, and across system platforms.

In this dissertation, we introduce Service Clouds, which represents a new approach to supporting the development of autonomic communication services. In this approach, collaborating overlay nodes use dynamic software composition to realize autonomic distributed services. The key insight is that overlay networks provide a "blank computational canvas" on which services can be instantiated and configured as needed, and later reconfigured in response to changing conditions. We have implemented a prototype of Service Clouds and evaluated it both for high-performance communication services atop the Internet, and for mobile computing services. The Service Clouds architecture and prototype enable developers to rapidly design and deploy autonomic communication services. This dissertation addresses four aspects of autonomic communication services.

First, we explore requirements for autonomic communication services, identify constituent components for an infrastructure. We design the Service Clouds overlay-based model and architecture and describe how to organize components, specify interactions among building blocks, and provide basic algorithms and protocols to compose services. We conclude this part with a description of the prototype.

Second, we evaluate the Service Clouds model on the Internet. We use the prototype to construct and deploy two services: TCP-Relay, in which a node is selected and configured dynamically to serve as a relay to expedite data transfer; and MCON, a multipath connection algorithm that exploits network topology to dynamically establish a high-quality secondary path, as a shadow connection to the primary path to support robust streaming.

Third, we investigate extension of Service Clouds to the wireless edge of the Internet to support pervasive mobile computing. Specifically, we address multimedia streaming and demonstrate how dynamic creation and configuration of proxy services support QoS streaming in highly dynamic conditions and frequent relocation of users.

Fourth, we address composition of distributed services atop an overlay network. We introduce and evaluate Dynamis, a generic algorithmic approach to locating suitable nodes on which to map service entities with minimal probing overhead.