ON THE BREAK-EVEN POINTS BETWEEN CLOUD-ASSISTED AND TRADITIONAL ROUTING

Abstract

Concerns regarding the routing scalability has increased recently, as there are more than 40K service providers advertising more than 600K IP prefixes. Today, a typical edge router receives thousands of updates per second on how to route traffic towards these prefixes. Multi-homing and peering practices as well as the demand on more routing flexibility (e.g., source routing, multi-path routing, QoS routing) have been contributing to this routing complexity. Operators rightfully expect more flexibility and programmability in router con- figurations, which further challenges router architectures by inclusion of software-based designs and virtualization of routing as a service. To explore a spectrum of designs, we consider a Cloud-Assisted Routing (CAR) framework which follows a hybrid and oppor- tunistic approach by keeping the high priority tasks at the router and use an adaptive router- cloud integration when beneficial. Instead of a complete separation of the two planes, it maintains most of the control plane functions at the cloud and least of it at local router and vice versa for the data plane. We present the architectural view of CAR and results from our initial prototype of forwarding table size reduction using CAR. We discuss possible intra- and inter-domain optimizations and perform a cost comparison between legacy router vs. CAR. Comparing the performance and monetary cost benefits between CAR, we discuss: i) What is the break-even point? ii) What are the key components of CAR to be monetarily beneficial? iii) How much savings CAR can offer regarding FIB table size? iv) What are the constraints that will make the traditional routing favorable than CAR?