MPI performance analysis of amazon EC2 cloud services for high performance computing

Abstract

Cloud computing offers a highly scalable infrastructure for high performance computing. Using the Amazon EC2 Cloud service, it is possible to get a set of computing instances on demand without requiring a lot of maintenance and financial resources a common cluster would need. The drawback for communicational intensive algorithms is, that the distribution of the instances is arbitrary and the communication speed might vary a lot which might affect the overall speed of the algorithms significantly. We present a benchmark for cloud computing on EC2 using the MPI to measure communication speeds of a set of cloud instances. This benchmark can visualize the actual network infrastructure. By adapting the measured data to the algorithms constraints it can help to manage the distribution of tasks to cloud instances to get an optimal distribution concerning network communication speeds. © 2011 Springer-Verlag.