Cloud Computing for COVID-19: Lessons Learned From Massively Parallel Models of Ventilator Splitting
Michael Kaplan, Charles Kneifel, Victor Orlikowski, James Dorff, Mike Newton, Andy J. Howard, Don Shinn, Muath Bishawi, Simbarashe Chidyagwai, Péter Balogh, Amanda Randles
Abstract
A patient-specific airflow simulation was developed to help address the pressing need for an expansion of the ventilator capacity in response to the COVID-19 pandemic. The computational model provides guidance regarding how to split a ventilator between two or more patients with differing respiratory physiologies. To address the need for fast deployment and identification of optimal patient-specific tuning, there was a need to simulate hundreds of millions of different clinically relevant parameter combinations in a short time. This task, driven by the dire circumstances, presented unique computational and research challenges. We present here the guiding principles and lessons learned as to how a large-scale and robust cloud instance was designed and deployed within 24 hours and 800 000 compute hours were utilized in a 72-hour period. We discuss the design choices to enable a quick turnaround of the model, execute the simulation, and create an intuitive and interactive interface.