| Computing environments have gone through radical changes in the last two decades. There has been widespread production and deployment of elaborate and innovative systems and technologies. As a result, there has been a huge proliferation of new and more complex computing devices. Typically, a large number of these heterogeneous devices are interconnected to make up a large distributed system. Managing, maintaining, protecting and securing such complex systems is quite challenging even to the most skilled IT professionals. Moreover, users tend to have even stricter expectations from today's computing systems in terms of performance, availability, reliability, and security. Therefore, it is a vital necessity that current and future computer systems be built with capabilities of self-management, self-organization, self-protection, and self-healing. That is exactly the vision of Autonomic Computing. This dissertation presents a novel and robust approach to autonomic computing through analytic performance models with a greater emphasis on the self-managing and self-configuring aspects. It starts by introducing a generic architecture for a controller system that allows for self-management and self-organization and that takes into account several design considerations for the controller including the use of workload forecasting, frequency of control, and robustness of the controller. The dissertation shows how our approach to autonomic computing achieves the expected results for three instances of autonomic computing systems, namely a multi-threaded server, an Internet data center, and a virtualized server. |
