| Computer systems have limited amounts of resources to serve applications' growing demands. Most systems tend to allocate resources to applications by offline analysis of application requirements, which often results in inefficient resource usage due to the dynamic, time-varying nature of workloads. Formal control theory is known to effectively support the desired performance of the controlled system. However, it is challenging to support the performance of computational systems, since there is no definitive methodology to model computer system dynamics unlike physics laws applied to model physical systems such as a cruise control system. Modeling computer systems, selecting proper control theoretic tools and tuning them according to the needs of specific applications - the research problems to be investigated in this proposed work- are the key ingredients for successful application of control theory to computer system performance management.;To support the desired system performance even in the presence of dynamic workloads, we have applied advanced control theoretic approaches, namely fuzzy control theory, model predictive control theory and event-driven control theoretic techniques. Specifically, we apply these techniques to manage the CPU utilization in a real-time operating system, network bandwidth consumption for video streaming, and link congestion in network gateways.;I aim to demonstrate the applicability of formal control theoretic techniques to support desired system behaviors even in the presence of dynamic workloads and uncertain environments. In this way, I intend to improve the predictability and reliability of computer systems that need to process highly dynamic workloads in uncertain environments. |
