valiant - Designing a Predictable Internet Backbone Network...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Designing a Predictable Internet Backbone Network * Rui Zhang-Shen, Nick McKeown Computer Systems Laboratory Stanford University { rzhang, nickm } ABSTRACT Designing a backbone network is hard. On one hand, users expect the network to have very high availability, little or no congestion, and hence little or no queueing delay. On the other hand, traffic conditions are always changing. Over time usage patterns evolve, customers come and go, new ap- plications are deployed, and the traffic matrices of one year are quite different from the next. Yet the network operator must design for low congestion over the multiple years that the network is in operation. Harder still, the network must be designed to work well under a variety of link and router failures. It is not surprising that most networks today are enormously overprovisioned, with typical utilizations around 10%. In this paper we propose that backbone networks use Valiant Load-balancing over a fully-connected logical mesh. This is quite a radical departure from the way backbones are built today, and raises as many questions as it answers. But it leads to a surprisingly simple architecture, with pre- dictable and guaranteed performance, even when traffic ma- trices change and when links and routers fail. It is provably the lowest capacity network with these characteristics. In addition, it provides fast convergence after failure, making it possible to support real-time applications. 1. INTRODUCTION Network design can be formulated as an optimiza- tion problem where total cost is minimized subject to topology, demand, and performance constraints. The designer chooses where nodes and links are placed, their capacities, and how traffic is routed. On the face of it, this seems like a straightforward problem. We first de- termine the constraints on node and link location, and the expected demand, and then do our best to design an optimal network. In this paper, we will focus on designing a backbone network. Once deployed, the expense of the infrastructure dic- tates that the backbone topology (i.e., the set of nodes and their inter-connectivity) doesnt change for several years. As traffic patterns change, and usage grows, a new topology will eventually need to be deployed, and * This research was funded by NSF under ITR award ANI- 0331653, and by the Lillie Family Stanford Graduate Fel- lowship. the design process starts over. A well-designed network should support the traffic matrices presented to it for as long as possible. In the rest of this section, we will discuss some of the many challenges this presents to network designers today. 1.1 Obtaining a Traffic Matrix Estimation The key to designing a network is to understand how it will be used. The process starts by measuring how the network is used today, and then extrapolating based on estimates of how traffic will grow and change over time....
View Full Document

This note was uploaded on 10/26/2011 for the course CS 7260 taught by Professor Staff during the Spring '08 term at Georgia Institute of Technology.

Page1 / 6

valiant - Designing a Predictable Internet Backbone Network...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online