2003 12th International Conference on Parallel Architectures and Compilation Techniques (2003)
New Orleans, Louisiana
Sept. 27, 2003 to Oct. 1, 2003
Kevin M. Lepak , University of Wisconsin-Madison
Harold W. Cain , University of Wisconsin-Madison
Mikko H. Lipasti , University of Wisconsin-Madison
Recent work has shown that multithreaded workloads running in execution-driven, full-system simulation environments cannot use instructions per cycle (IPC) as a valid performance metric due to non-deterministic program behavior. Unfortunately, invalidating IPC as a performance metric introduces its own host of difficulties: special workload setup, consideration of cold-start and end-effects, statistical methodologies leading to increased simulation bandwidth, and workload-specific, higher-level metrics to measure performance. This paper explores the non-determinism problem in multithreaded programs, describes a method to eliminate non-determinism across simulations of different experimental machine models, and demonstrates the suitability of this methodology for performing architectural performance analysis, thus redeeming IPC as a performance metric for multithreaded programs.
H. W. Cain, K. M. Lepak and M. H. Lipasti, "Redeeming IPC as a Performance Metric for Multithreaded Programs," 2003 12th International Conference on Parallel Architectures and Compilation Techniques(PACT), New Orleans, Louisiana, 2003, pp. 232.