High Performance Optimization on Cloud for a Metal Process Model

• Main Author: Saxén, Adam
• Format: Others
• Language: English
• Published: Uppsala universitet, Avdelningen för beräkningsvetenskap 2014
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-228088

The Amazon Elastic Compute Cloud (EC2)is a service providing on-demand compute capacity to the public. In this thesis a scientific software, performing global optimization on a metal process model, is implemented in parallel using MATLAB and provisioned as a service from AmazonEC2. The thesis is divided into two parts. The first part concerns improving the serial software, analyzing different optimization methods, and implementing a parallel version; the second part is about evaluating the parallel performance of the software, both on different computer resources in Amazon EC2 and on a local cluster. It is shown that parallel performance of the software in Amazon EC2 is similar and even surpasses the local cluster for some provisioned resources. Factors affecting the performance of the global optimization methods are found and related to network communication and virtualization of hardware, where the method MultiStart has the best parallel performance. Finally, the runtime for large optimization problem was successfully reduced from 5 hours(serial) to a few minutes (parallel) when run on Amazon EC2; with the total cost of just 25-30$.