Date of Award
Doctor of Philosophy (PhD)
School of Computing
Performance, usability, and accessibility are critical components of high performance computing (HPC). Usability and performance are especially important to academic researchers as they generally have little time to learn a new technology and demand a certain type of performance in order to ensure the quality and quantity of their research results. We have observed that while not all workloads run well in the cloud, some workloads perform well. We have also observed that although commercial cloud adoption by industry has been growing at a rapid pace, its use by academic researchers has not grown as quickly. We aim to help close this gap and enable researchers to utilize the commercial cloud more efficiently and effectively.
We present our results on architecting and benchmarking an HPC environment on Amazon Web Services (AWS) where we observe that there are particular types of applications that are and are not suited for the commercial cloud. Then, we present our results on architecting and building a provisioning and workflow management tool (PAW), where we developed an application that enables a user to launch an HPC environment in the cloud, execute a customizable workflow, and after the workflow has completed delete the HPC environment automatically. We then present our results on the scalability of PAW and the commercial cloud for compute intensive workloads by deploying a 1.1 million vCPU cluster. We then discuss our research into the feasibility of utilizing commercial cloud infrastructure to help tackle the large spikes and data-intensive characteristics of Transportation Cyberphysical Systems (TCPS) workloads. Then, we present our research in utilizing the commercial cloud for urgent HPC applications by deploying a 1.5 million vCPU cluster to process 211TB of traffic video data to be utilized by first responders during an evacuation situation. Lastly, we present the contributions and conclusions drawn from this work.
Posey, Brandon M., "Evaluating and Enabling Scalable High Performance Computing Workloads on Commercial Clouds" (2019). All Dissertations. 2497.