Where there's an app for everything, even in a business environment, it's time to observe the impact these apps and application-layer issues are having on the network -- particularly in latency-sensitive wide area network (WAN) environments. How do application-layer issues affect WANs? And how should the WAN be architected differently to accommodate all the applications across a network?
SearchEnterpriseWAN talked to Bruce Davie, the co-author of Computer Networks, Fifth Edition, to answer these questions and more. Davie is a visiting lecturer at MIT and chief service provider architect at Nicira, with a doctorate in computer sciences.
What has changed from the previous edition of Computer Networks to the fifth edition, regarding wide area networking?
Bruce Davie: With every edition we try to increase the coverage of protocols and technologies that are in used in today's networks, while trimming coverage of protocols that have fallen out of favor. In the specific case of WAN protocols, this has led us to cover MPLS more fully, and to reduce the coverage of ATM in this edition. We've also updated and expanded the coverage of BGP, the routing protocol on which the Internet depends.
Why has there been an increased focus on application-layer issues?
Davie: There are two reasons for this. One is that today's students come to the field of networking with considerable experience of networked applications -- a sharp contrast to the situation when we first wrote this book in 1995. We feel that anyone trying to learn networking from a book -- whether a college student or a professional -- will grasp the concepts better if they are related to the applications that the reader has already experienced.
The second reason is that the "core" parts of networking -- those below the application layer -- have actually been fairly steady for the last 10 years or so. There is just less innovation happening at the lower layers (with a few exceptions, such as wireless). So we try to focus more on the areas where the most innovation is happening, and, clearly, applications are one of those areas.
How do these application-layer issues affect wide area networks (WANs)?
Davie: In many respects, the WAN hasn't needed to change very much to accommodate the new applications, largely because of the end-to-end principle, which places most of the responsibility for correct application behavior in the end-systems, not in the WAN equipment itself.
How should the WAN be architected differently than before to accommodate all the applications across a network, if at all?
Davie: For some classes of applications, like Internet telephony, the WAN needs to be architected to provide adequate QoS [Quality of Service] for the applications. In some cases this is done only by provisioning adequate bandwidth throughout the network; in other cases, explicit QoS mechanisms of the sort we discuss in Chapter 6 of the book are used. And sometimes overlay networks, discussed in Chapter 9, can compensate for some of the shortcomings of the WAN. This is what Skype does, for example.
How much of network performance issues can be fixed through re-architecting or redesigning your WAN?
Davie: Skype is a good example of an application that works in spite of WAN problems using a combination of adaptive coding and overlay routing. By contrast, a lot of enterprise Voice-over-IP [VoIP] deployments have depended on the deployment of routers and switches with good QoS support.
How can you incorporate lossless compression techniques in your WAN?
Davie: This is effectively what WAN acceleration devices do. They apply the lossless compression techniques to traffic flowing across the WAN to reduce bandwidth requirements.
How much is dependent on acceleration technologies?
Davie: I see WAN acceleration as more of a cost saving technique -- figuring out how to reduce the amount of traffic sent over the WAN so that you can avoid buying more bandwidth. Certainly that is one approach to improving network performance without making significant changes to the WAN itself.