Mike Prausa (The MITRE Corporation), Norman Facas (The MITRE Corporation)
Keywords: Astrodynamics
Abstract:
This paper examines a horizontal scaling approach for Conjunction Assessment (CA) processing using the Apache Hadoop framework. A key focus of this paper is the Input/Output (I/O) characteristics of CA, which directly impact the performance and scalability of the distributed system. To accomplish this and maintain a focus on I/O, a synthetic data model and simulated CA algorithm are used. These components form the basis for a series of scalability tests that result in high collocation rates and near linear scaleout for the 16 node/64 core cluster used. Amdahls law is used to examine these results as well as provide a basis for extrapolated results. These extrapolated results model larger clusters and identify potential degradation as node count increases.
Date of Conference: September 13-16, 2011
Track: Astrodynamics