A Distributed Network Services Stack for SD-WANs

Lakshmi Subramanian


Lakshminarayanan Subramanian, New York University


Majority of software-defined network design and implementation efforts have focused on local area networks. These environments are tightly managed, with reasonable control over network failures, and relatively few unpredictable event occurrences. SD-WANs extend the traditional notion of SDNs to the wide area network, where SDNs will face three fundamental challenges: increased latency, significantly reduced and highly variable network  bandwidth, and an increased probability of failures. This talk will describe our early work on a Distributed Network Service (DNSs) stack for the design of a distributed, scalable, and fault-tolerant architecture for SDWANs. Availability and tolerance towards network partitions have been well-documented, but their impact on consistency in moderate bandwidth, increased latency environments is not well studied.  The DNS stack is a generalization of the Network Service Virtualization (NSV) vision, where service capabilities are tied to each instance of an application and are orchestrated and instantiated with the application. One can imagine the DNS paradigm to be a fluent combination of both SDN and NFV; the paradigm provides a clean decoupling between network control and forwarding, while providing infrastructural cost reduction due to virtualization of network functionalities (or applications). We will explain the realization of our DNS stack vision across three SD-WAN contexts: extreme Internet caching, programmable IoT platforms and programmable cellular networks.