Parallel and survivable multipath circuit provisioning in ESnet’s OSCARS

Abstract

Data generation is approaching petascale and exascale rates by cutting-edge science and research applications varying from material informatics to physics. With data generation and management comes the necessity to transmit such vast collections of information across the world's networks for processing, analysis, storage, or peer-sharing. This practice is becoming the norm to the large-scale scientific community, but complications can arise during networking. There are countless situations such as component failure due to a harmless construction accident or a devastating natural disaster that may lead to catastrophic interruption of service. Furthermore, given the size of datasets, there is a strong need to support intelligent and fast parallelism throughout the network to allow end users to efficiently consume available bandwidth. We therefore propose a multipath extension for ESnet's On-demand Secure Circuits and Advance Reservation System (OSCARS), the network research community's most popular long-lived circuit-provisioning software package. Presently, OSCARS supports purely point-to-point circuits; however, our proposed client software provides an overlay onto the default OSCARS path computation engine that enables end users to route their data along multiple linkdisjoint paths to provide session survivability and increase the degree of parallelism. We have also adapted the proposed multipath extension to an existing anycast OSCARS deployment, which allows for the selection of one preferred destination node from among a set of potential candidates. Through thorough simulation analysis and exposure to realistic failure event distributions, we quantitatively evaluate the multipath client performance and showcase the relative benefits when compared to the standard single-path OSCARS deployment.