The Testing and Benchmarking of Speed and Efficiency of a Cluster Computer
Acadia ElzHowe, Hina Saleem, Bryan R. Kuhr · 2025
This work seeks to improve the efficiency of the Advanced Simulation for Teaching and Research Opportunities (ASTRO) cluster computer at Sweet Briar College. ASTRO, with its eight-node Raspberry Pi configuration, has the potential to provide computational power comparable to remote services, enabling faculty and students to conduct data-intensive computations in a cost-effective and energy-efficient manner. The research aims to evaluate and optimize ASTRO’s performance and energy efficiency as a case study in desktop cluster computing. This work will address the questions: How much energy does ASTRO use, and is it an efficient desktop supercomputer? High-Performance Linpack (HPL) will benchmark ASTRO’s speed in floating-point operations per second (flops) by solving linear systems of equations quickly and accurately. Optimal inputs will be identified to maximize performance and then used for all further benchmarking and trials. The energy consumption of ASTRO will be measured using a Kill-A-Watt device, recording the energy usage (watt-seconds) and the average power (watts) during the benchmark and idle states. Efficiency will be calculated in gigaflops per watt (Gflops/W). The study will provide a comprehensive review of ASTRO’s performance, including its computational speed, energy consumption, and efficiency benchmarks. The performance of ASTRO will be compared with similar cluster computer systems to determine how well a selfbuilt supercomputer functions as an alternative to purchasing cloud time or other advanced supercomputing solutions. After establishing ASTRO’s benchmarks for computational speed and energy demands, this study will provide ideas for further research and potential use for future students and professors.