The performance of a processor is limited by the specific bottlenecks that a benchmark exposes while running on that processor. Since the quantification of these bottlenecks can be extremely time-consuming, our prior work proposed using the Plackett and Burman design as a statistically-rigorous, but time-efficient method of determining the processor's most significant performance bottlenecks. In this paper, we use the Plackett and Burman design to quantify the magnitude of the bottlenecks in the SPEC CPU 2000 benchmark suite from the viewpoints of both performance and energy consumption. We then use Principal Components Analysis, and hierarchical and K-means clustering algorithms to determine the similarity of the benchmarks based on their energy-delay production bottlenecks.
|Original language||English (US)|
|State||Published - Dec 1 2006|
|Event||SPEC Benchmark Workshop 2006 - Austin, TX, United States|
Duration: Jan 23 2006 → Jan 23 2006
|Other||SPEC Benchmark Workshop 2006|
|Period||1/23/06 → 1/23/06|