Date of Award

Spring 1-1-2011

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical, Computer & Energy Engineering

First Advisor

Dirk Grunwald

Second Advisor

Li Shang

Third Advisor

Jeremy Siek

Abstract

Computer systems are resource constrained. Application adaptation is a useful way to optimize system resource usage while satisfying an application’s performance requirements. Current multicore computer systems supporting these applications, however, are not designed to reliably meet these requirements. Meanwhile, these computer systems are resource-limited, e.g., have power-induced energy and thermal constraints. Compounding the application’s performance requirements are increasingly-stringent microprocessor thermal constraints. Previous application adaptation efforts, however, were ad-hoc, time-consuming, and highly application-specific, with limited portability between computer systems.

This thesis presents OCCAM, a software platform for developing multicore adaptable applications. OCCAM’s design-time platform consists of design patterns, APIs, and data structures that allow application developers to specify the performance constraints and application-specific optimization techniques. OCCAM generates a run-time controller offline, using profiling data. It then uses this profiling data to generate an internal model that it subsequently employs to generate a robust Markov Decision Process-based Model Predictive Controller. Using a set of Recognition, Mining, and Synthesis benchmarks, the experimental study demonstrates that OCCAM can successfully optimize the system while meeting the systems performance requirements across a wide range of computer platforms, ranging from an energy-constrained single-core system to a high-performance 16-core system. Finally, OCCAM presents a simulation-based, stochastic model checking-based framework for quantifying the robustness of the controller.

Share

COinS