Performance Assessment of the Cloud for Prototypical Instant Computing Approaches in Geoscientific Hazard Simulations

Abstract

Computing forecasts of hazards, such as tsunamis, requires fast reaction times and high precision, which in turn demands for large computing facilities that are needed only in rare occasions. Cloud computing environments allow to configure largely scalable on-demand computing environments. In this study, we tested two of the major cloud computing environments for parallel scalability for relevant prototypical applications. These applications solve stationary and non-stationary partial differential equations by means of finite differences and finite elements. These test cases demonstrate the capacity of cloud computing environments to provide scalable computing power for typical tasks in geophysical applications. As a proof-of-concept example of an instant computing application for geohazards, we propose a workflow and prototypical implementation for tsunami forecasting in the cloud. We demonstrate that minimal on-site computing resources are necessary for such a forecasting environment. We conclude by outlining the additional steps necessary to implement an operational tsunami forecasting cloud service, considering availability and cost.

Leveraging Google's Tensor Processing Units for tsunami-risk mitigation planning in the Pacific Northwest and beyond

“…The TPU may increasingly become a standard hardware on which physics-based machinelearning algorithms will be built (Rasp et al, 2018;Mao et al, 2020;Wessels et al, 2020;Fauzi and Mizutani, 2020;Liu et al, 2021;Kamiya et al, 2022). Through its relative ease of access and potential for rapid simulation capabilities, cloud computing provides a valuable alternative to higherperformance computing clusters (Behrens et al, 2022;Zhang et al, 2010), particularly for communities with limited access to local clusters. By leveraging these benefits of the Google Cloud TPU, we propose our implementation as one step towards a community-based, user-friendly virtual laboratory that can be run by a minimally trained user on the cloud.…”

Machine learning-based tsunami inundation prediction derived from offshore observations

“…The inundation map is intuitively understandable to the public as it can be overlaid with geographical maps and visualized via a common online map service, such as Google Maps 17 . Furthermore, cloud computing technology for online tsunami simulations has been proposed 18 . The on-demand computing environment is suitable and cost-effective for a real-time tsunami prediction occasionally accessed only during the event.…”