The proposed research seeks to investigate mathematical and formal methods for evaluating new computer system architectures. It involves adapting fundamental techniques from existing statistical and mathematical theory to permit the use of simulation-based evaluation in exploratory design. Specific issues to be studied deal with workload compression using sampling, and benchmark synthesis based on workload and parameterized models.
Detailed full-system simulation with real-world workloads is the preferred method of performance analysis in the microprocessor and computer system design community today, in preference to analytic modeling or the use of synthetic benchmarks. However, the very large simulation times taken by this methodology prohibits good design space exploration needed for increasingly complex designs. In addition, the project will address the training of computer architecture students in neglected areas of statistical or analytical modeling techniques.
