Benchmarking the Processing of Aircraft Tracks with Triples Mode and Self-Scheduling

TL;DR

This paper presents a new high performance computing workflow for processing aircraft track data, significantly reducing analysis time from weeks to days, to support collision risk modeling for unmanned aircraft integration.

Contribution

It introduces a novel, optimized workflow leveraging job launch and task distribution technologies for efficient aircraft data processing.

Findings

Workflow reduced processing time from weeks to days

Benchmarking on two datasets demonstrated improved performance

New dataset focused on aerodrome environment included

Abstract

As unmanned aircraft systems (UASs) continue to integrate into the U.S. National Airspace System (NAS), there is a need to quantify the risk of airborne collisions between unmanned and manned aircraft to support regulation and standards development. Developing and certifying collision avoidance systems often rely on the extensive use of Monte Carlo collision risk analysis simulations using probabilistic models of aircraft flight. To train these models, high performance computing resources are required. We've prototyped a high performance computing workflow designed and deployed on the Lincoln Laboratory Supercomputing Center to process billions of observations of aircraft. However, the prototype has various computational and storage bottlenecks that limited rapid or more comprehensive analyses and models. In response, we have developed a novel workflow to take advantage of various job launch and task distribution technologies to improve performance. The workflow was benchmarked using two datasets of observations of aircraft, including a new dataset focused on the environment around aerodromes. Optimizing how the workflow was parallelized drastically reduced the execution time from weeks to days.