An event-driven approach to serverless seismic imaging in the cloud

This repository contains scripts and instructions to implement the AWS seismic imaging workflow described in "An event-driven approach to serverless seismic imaging in the cloud". The workflow consists of a serverless algorithm implementation using AWS Step Functions and an event-driven map-reduce implementation for computing gradients of objective functions.

• IEEE TPDS preprint on arXiv

Content

This repository is structured as follows:

• batch: Parameter files for setting up the AWS Batch computing environment.

• docker: Dockerfiles that specify the Docker images used for seismic modeling and imaging in our workflow. A separate Docker image is defined for imaging with and without MPI. This directory also contains the definitions of wave equations using Devito, as well as some auxiliary Python functions required for the workflow.

• docs: Documentation for implementing the workflow on AWS. Follow the below link to access the full documentation.

• lambda: Source code of Lambda functions used in the workflow. These include the function for the event-driven gradient summation and automatic submission of AWS Batch jobs.

• numerical_examples: Source code and scripts to reproduce the numerical examples and performance tests.

• service_roles: Parameter files for creating the necessary IAM roles to reproduce the workflow.

• step_functions: Definition of the seismic imaging workflow using AWS Step Functions. The workflow is defined as a json file.

Instructions for setting up the AWS workflow

Detailed instructions for setting up the workflow on AWS are provided in the documentation.