Benchmarking Amazon EC2 for hig-performance scientific computing 论文

摘要

How effective are commercial cloud computers for high-performance scientific computing compared to currently avaible alternatives? I aim to answer a specific instance of this question by examining the performance of Amazon EC2 for high-performance scientific applications. I used macro and micro benchmarks to study the performance of a cluster composed of EC2 high-CPU compute nodes and compared this against the performance of a cluster composed of equivalent processors avaible to the open scientific research community. My results show a significant performance gap in the examined clusters that system builders, computational scientist, and commercial cloud computing vendors need to be aware of .