Looking for Novel Ways to Obtain Fair Measurements in Multithreaded Architectures

Current methodologies do not provide representative results for the evaluation of multithreaded architectures, which could lead to unfair or misleading conclusions. This paper presents FAME, a novel evaluation methodology aimed to fairly measure the performance of multithreaded processors. FAME reexecutes all threads in a multithreaded workload until all of them are fairly represented in the final measurements taken from the workload. We compare FAME with previously used evaluation methodologies in architectural research scenarios. Our results show that FAME provides the most accurate measurements, becoming an ideal evaluation methodology to analyze novel design ideas implemented in multithreaded architectures. Palabras clave— Evaluation Methodology, Performance Metric, Multithreaded Architetures, Fairness.