DataPlane Broker : Open WAN control for multi-site service orchestration

Simpson, Steven and Farshad, Arsham and McCherry, Paul and Magzoub, Abubakr and Fantom, Will and Rotsos, Charalampos and Race, Nicholas and Hutchison, David (2020) DataPlane Broker : Open WAN control for multi-site service orchestration. In: 2019 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN) :. IEEE, Dallas, USA, pp. 1-6. ISBN 9781728145457

Text (1570566040)
1570566040.pdf - Accepted Version
Available under License Creative Commons Attribution-NonCommercial.
Download (339kB)

Abstract

NFV-MANO has become the de-facto standard for network service orchestration in future programmable network infrastructures. Relevant standards define an architecture and a data model that allows an orchestration entity to deploy, dynamically configure and monitor virtual and physical network functions across virtualized datacenters. Although the model offers extensive details for network function management and host-level network configuration, end-to-end connectivity management beyond the datacenter remains limited. The responsibility for cross-site connectivity is delegated to a control abstraction, the WAN Infrastructure Manager (WIM), which is partially defined in relevant standards. The reference implementation of the NFV-MANO standards, Open Source MANO (OSM), has recently released a reference WIM driver model, but no open-source WIM implementation is currently available, thus restricting the ability of researchers to experiment with multi-site service deployments. In this paper, we present the DataPlane Broker (DPB), the first open-source WIM implementation for software-defined networks. Using an extensive data model, DPB seamlessly translates NFV-MANO requirements into SDN configuration, supporting point-to-point and multipoint connectivity with strong bandwidth guarantees. DPB is integrated with OSM FIVE via a WIM plugin. Initial experiments highlight that path computation can scale to large network topologies and a large number of services, with negligible computational overheads, while DPB increases by less than 10% the overall deployment delay of an OSM service.