Model-based Performance Characterization of Software Correlators for Radio Interferometer Arrays. (1st October 2022)

• Record Type: Journal Article
• Authors: Vázquez, A. J.; Elosegui, P.; Lonsdale, C. J.; Crew, G. B.; Fish, V. L.; Ruszczyk, C. A.
• Is Part Of: Publications of the Astronomical Society of the Pacific. Volume 134:Number 1040(2022)
• Publication Date: 2022-10-01
• Subjects: Astronomy -- Periodicals; Astronomy; Periodicals; Periodicals; 520.5
• DOI: 10.1088/1538-3873/ac8dc2
• Languages: English
• ISSNs: 0004-6280

Abstract

Abstract: Correlation for radio interferometer array applications, including Very Long Baseline Interferometry (VLBI), is a multidisciplinary field that traditionally involves astronomy, geodesy, signal processing, and electronic design. In recent years, however, high-performance computing has been taking over electronic design, complicating this mix with the addition of network engineering, parallel programming, and resource scheduling, among others. High-performance applications go a step further by using specialized hardware like Graphics Processing Units (GPUs) or Field Programmable Gate Arrays (FPGAs), challenging engineers to build and maintain high-performance correlators that efficiently use the available resources. Existing literature has generally benchmarked correlators through narrow comparisons on specific scenarios, and the lack of a formal performance characterization prevents a systematic comparison. This combination of ongoing increasing complexity in software correlation together with the lack of performance models in the literature motivates the development of a performance model that allows us not only to characterize existing correlators and predict their performance in different scenarios but, more importantly, to provide an understanding of the trade-offs inherent to the decisions associated with their design. In this paper, we present a model that achieves both objectives. We validate this model against benchmarking results in the literature, and provide an example for its application for improving cost-effectiveness in the usage of cloud resources.