Power-efficient and scalable virtual router architecture on FPGA

In the recent years, networking infrastructure has advanced in such a way that router hardware management and power efficiency issues have gained considerable attention. Router virtualization alleviates these issues by allowing a single hardware router to serve packets from multiple networks. We propose a power-efficient scalable architecture for implementing router virtualization using the Virtualized Merged (VM) approach on Field-programmable Gate Array (FPGA). Three novel optimizations are incorporated into the basic VM approach to reduce the dynamic power dissipation of the router hardware and deliver higher throughput per unit power consumed. The reduction in power consumption is in part due to the savings in memory required to store the merged lookup table, which makes our optimized VM approach more scalable with respect to the number of virtual routers per FPGA chip. Also, by exploiting the low power features and clock gating techniques, the optimized VM approach achieves significant power savings. To illustrate the improvements achieved, we tested the optimized VM router using 75 routing tables on a single FPGA, utilizing 50% less memory and consuming 20% less power compared with the basic VM approach.

I. Introduction

In recent years, the Internet has been growing at a rapid pace, and currently consists of over 5 billion nodes [11]. This growth has been supported by the development of high performance networking hardware, made possible by the continuous advancements in semiconductor technology. Network virtualization [1], [2] was proposed to take advantage of these advancements to improve network efficiency. This has been realized via router virtualization, at the hardware level. Router virtualization allows multiple network hardware to be consolidated onto a single shared platform, thereby enabling centralized management of multiple networks. Reduced hardware costs, reduced power consumption, and ease of management are a few additional benefits. These features have become attractive especially in the context of datacenter networking and Software Defined Networking (SDN) [4].