Design, Measurement and Management of Large-Scale IP Networks: Bridging the Gap Between Theory and Practice
What makes the information in Design, Measurement and Management of Large-Scale IP Networks unique is that it is based on actual measurements collected from the Sprint IP backbone. Operational measurements when used as a basis for the design of networks has two important benefits: first, one can ensure that a network design can scale and, second, one can predict and quantify a performance gain of tradeoffs that could be observed under realistic conditions.
Design, Measurement and Management is organized into three sections. The first section covers the state of the art in the measurement of IP backbone networks, and identifies protocols and tools. The second section builds on the first and presents techniques for making use of information collected by protocols and tools. The third section addresses aspects of designing high performance IP networks, in particular toward management of a system that provides services while avoiding overload.
Background and context is provided on the Internet itself, including the OSI protocol, TCP/IP, ISPs, and the Sprint IP backbone. What is most surprising for a non-expert is in learning that IP network management today is primarily reactive in nature. When problems arise, network management relies on trial and error. Why is this so? The authors answer that question.
First, the designers of IP networks have attached less importance to network monitoring and resource accounting than to distributed management and robustness. This choice results in routers and end hosts that have not been designed to retain the right detailed information about the traffic flowing through them. As well Internet protocols also do not provide enough of the right detailed information about the state of the underlying network.
Second, is that although the network itself is automatically corrective, which is a good thing, self-correction has a bad side effect, that of making it difficult for a network administrator to track down the cause of network failure.
Third, current network administration tools pose tradeoffs among quality of collected information, requirements, overheads, and costs.
What is most contradictory in this situation is noted by the authors but not identified until the very middle of the book, and that fact is: One of the biggest infrastructure costs faced by network providers is with deployment and maintenance of its links. The authors claim that to solve that problem what is needed is a monitoring structure that reduces network maintenance costs. This solution is not at manageable by single technique but must be heterogeneous in nature, a variety of tools working at different conceptual and physical levels, providing a range of variable timescale statistics, aggregate flow statistics, real-time packet capture, and coordinated router-assisted measurements. From this point in the text, the book goes through the details of identifying exactly what is needed before such tools can be provided, such as determining end-to-end packet delays, which must be accumulated from individual components of through-router delays.
Operational end-to-end data was collected by the authors and presented as their starting point for their wide ranging solution, and was obtained through measurements that were only made possible through the authors’ access to the Sprint IP backbone network. From the data collected the authors were able to make accurate models for router-delay performance and reveal to the reader insights about generalized IP backbone networks that this reviewer will not spoil. You just have to read the book yourself to find out.
From the insight gathered from operational data, a more rational network design is proposed and laid out in a sequence of steps, each with their own chapter, as follows: 1. Topology design, 2. Topology resiliency, 3. Performance enhancement, and 4. Capacity planning. Each step is detailed, formalized and modeled and compared to the observations and information gathered from the Sprint data network.
The last section of the book, titled “From bits to services: information is power” addresses network security first as security on data used at the endpoints and identifies common attacks on the endpoints, then as security of the IP backbone infrastructure along and identifies common attacks on the infrastructure.
Note that the politics and economics of networks, the technical aspects of services related to providing (or not) network neutrality is not addressed by this text. Apart from that one hole (which might in fact be an interesting topic for a book of its own right), this book is a valuable addition to the bookshelf for network designers. Note also that each chapter in the text ends not with problems for students but with a section titled “Lessons Learned,” indicating intent of audience as the practicing engineer. This reviewer thinks that the authors have succeeded.
Reviewer Robert Schaefer is a Research Engineer at MIT Haystack Observatory.