Hawk-i HPC Cloud Benchmark Tool
Visakh Muraleedharan · 2012
Scientific computing unravels the mysteries of science by constructing mathematical models and numerical algorithms to simulate scientific scenarios. This requires massive computational power and High-Performance Computing (HPC) solutions. Cluster and grid have been the answer for massive computing needs for a long time. These HPC solutions are often hard and expensive to setup, maintain and use. Cloud computing is a model of delivering the compute infrastructure in which computation and storage can be dynamically provisioned on a pay as you go model. Using a vendor cloud service like Amazon Web Service (AWS) can significantly reduce the effort to access these ondemand high performance resources. In this work we create a live benchmark tool (Hawk-i) to study how we can usefully compare the suitability of Amazon EC2 instance types for scientific applications. We studied the memory scalability and performance of Spectral methods and N-Body Methods in a cluster built on top of cloud infrastructure. We have done a detailed analysis of the performance results and found some possible solutions to get better performance out of the cloud. Hawk-i tool proved to be very useful to generate large amount of benchmark data which made this study possible. With the simple user interface, it is also very easy to use service for non-technical users who have little knowledge of cloud or cluster environments.