All-atom forcefield and electronic structure studies of small molecules, proteins, and polymers receive a much smaller proportion of supercomputer allocations than atmospheric modeling, reacting flows, and some solid-state materials studies. Part of this gap is the difficulty in organizing high-throughput simulation campaigns capable of running millions of batched computations efficiently on existing supercomputing resources. This talk presents three recently published open-source tools for such `big-data' campaigns. They have been used in production on Summit supercomputer and Google Cloud to produce and analyze multi-terabyte datasets of molecular dynamics simulations on 25 SARS-Cov-2 protein systems as well as billion-compound docking studies to six of those protein targets. Each of the three tools employs a different scheduling and storage management paradigm. We contrast the scheduling paradigms to explain how they address challenges in data provenance, reproducibility, and mapping to existing facility infrastructure.

New tools for billion-molecule analytics at scale

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.