← Back to Sessions

Constructing Cloud Resources for the Individual Researcher From the Ground Up: An Example of Earthquake Detection in the Cloud

Description: 

The commercial cloud could potentially unlock boundless seismic processing at scale. However, only a few introductory educational examples of cloud-based workflows exist that fit the specific needs of most independent seismic researchers. To this end, we share our experience simultaneously storing large time-series datasets and distributing their processing across many individual cloud computing instances on the Microsoft Azure cloud computing platform. Our data processing workflow performs earthquake detection using both template-matching, through EQcorrscan (Chamberlain et al., 2018), and a machine learning detector, Earthquake Transformer (Mousavi et al., 2020), with the ability to flexibly scale computation by varying individual virtual machine and machine pool sizes. Our documented example shows how to containerize locally-developed codes and manage their use with cloud-stored data and cloud-hosted computing pools. We explore how the performance of CPU-only pools compares to GPU pools and examine whether the speed-up from GPU resources justifies their higher development and deployment cost. We highlight the complexity of navigating commercial cloud resources as an individual researcher and discuss how contrasting workflow set-ups, i.e. a few large computing nodes versus a big network of small instances, can be equally appropriate depending on the researcher’s time constraints and technical ability. The presented work is both a commercial cloud resource structure that can be readily modified by other researchers, and a preliminary comparison of two highly effective but rarely compared earthquake detection methods.

Session: Geophysical Data Analysis in Cloud Computing Environments [Poster]

Type: Poster

Date: 4/18/2023

Presentation Time: 08:00 AM (local time)

Presenting Author: Zoe Krauss

Student Presenter: Yes

Invited Presentation: 

Authors