원문정보
초록
영어
Recently, cloud computing has becoming a promising networking infrastructure paradigm which enable us to deploy large-scale applications in a cost-effective manner. However, many existing cloud platforms are designed for supporting commercial applications instead of large-scale scientific computing workloads. As a result, the effectiveness of running such kind of applications on cloud platforms is still an opening issue, especially when the performance penalties introduced by virtualization technology is taken into consideration. In this work, we take effects on analyzing the scheduling performance of cloud platforms for Parameter Sweep Applications (PSA), which is one of most used program models in large-scale scientific computing applications. All the experiments are conducted on our integrated performance evaluation middleware. The experimental results indicate that many conventional scheduling algorithms which are effective in classic distributed systems (i.e. grid, cluster) need to take into account the negative effects introduced by virtualization technology. In addition, the experimental results also indicate some useful hints for improving the scheduling performance of PSA workloads in cloud environments.
목차
1. Introduction
2. Related Work
3. Cloud Performance Monitor and Evaluator
3.1 Overall Framework of CP-M&E
3.2 Working Model of CP-M&E
3.3 Working Model of CP-M&E
4. Experiments and Performance Analysis
4.1 Experimental Setting
4.2 Performance on Deadline-Missing Rate
4.3 Performance on Resource Utilization
4.4 Performance Analysis with Different VM Provision
6. Conclusion
References