IP Quality of Service (Networking Technology)

Features

• Cover Type: Hard Cover with 368 pages
• Published by: Cisco Press February 2, 2001
• Written in: English
• ISBN 10 Number: 1578701163
• ISBN 13 Number: 978-1578701162
• Book Dimensions: 9.4 x 7.5 x 1 inches
• Weighs: 2 pounds

Reader Reviews
For those who need a general overview of how QoS is implemented in Cisco devices and software, this book would serve as a pretty good introduction. Some of the discussion is generic enough to be useful to those who are interested in QoS, but not necessarily implemented in a Cisco environment. Naturally then the reader is assumed to have a good background in Cisco network architectures. The author implements case studies in the book, with it being assumed that the reader is also comfortable with the actual administration of Cisco network devices. Therefore the book is really useful for those readers who are involved in the practical implementation of QoS schemes. Those interested in developing new approaches to QoS may still find the book helpful as an introduction to what is known. There are more specialized treatments that can be found online if one is willing to spend the time finding and downloading the documents (from the Cisco website). I only read the first six chapters of the book, which deal with IP QoS, and so my review will be restricted to these. Remembering that the Internet is a best-effort service only, the author introduces the IP QoS functions in chapter 1. The advent of voice and video traffic over IP for example, requires the need for QoS in modern networks. QoS services are divided into levels: best-effort, which does not guarantee traffic delivery; differentiated service, which groups traffic into classes but does not guarantee its delivery; and guaranteed service, which allocates network resources to ensure specific service requirements. Bandwidth, packet latency, and packet loss are measures used to characterize connection performance. Chapter 2 gives a more detailed overview of the differentiated services architecture for delivering QoS on the Internet. Called `diffserv' by the IETF, the author discusses the historical origins of it, and how it provides traffic differentiation by breaking traffic up into a small number of classes, with relative service priority existing among these classes. The presentation is straightforward to follow, once one gets used to remembering all of the many acronyms that are employed, such as PHB (per-hop behavior), DSCP (Differentiated Services Code Point), etc. The traffic conditioners, which are QoS functions that set the DSCP field and monitor traffic for profile compliance, and discussed in detail. Network provisioning, signaled QoS, and QoS policy manager are all discussed as resource provisioning policies. In chapter 3, the author overviews the use of traffic conditioning functions at the network boundary as a tool for providing differentiated services. In one, called `packet classification', the packets are identified using one or more fields in a packet. This could be the MAC address, URL, IP Precedence, etc. Then `packet marking' is used to mark classified packets according to their traffic class. Traffic rate management, another conditioning function, is discussed via the token bucket scheme, along with the CAR traffic policing function. The strategy of borrowing of tokens for token buckets with extended burst capability is very interesting and is a good candidate for improvement using techniques and concepts from financial engineering and artificial intelligence. The token bucket scheme is also discussed in the context of traffic shaping. In the next chapter on resource allocation, the author discusses how weighted fair queuing can be employed as a scheduling discipline in which flow differentiation occurs. The author makes some interesting and somewhat controversial remarks in this chapter, one being that after stating that some flows are delayed to offer a particular bandwidth to other flows, he concludes to the effect that a preferential treatment will result in another suffering. This is not really true as one can show using techniques from game theory. In fact, the max-min fair-share allocation scheme that he discusses next is a step in this direction. This is true also for the weighted max-min fair share allocation, in which each user is assigned a weight, with the fair-share being proportional to this weight. Generalized processing sharing (GPS) is discussed as an ideal scheduling mechanism that services an infinitesimally small amount of data from each nonempty queue via round-robin. This unrealistic requirement is then ameliorated by using fair queuing, a strategy that takes all flows to have the same weight, and simulates GPS by computing a sequence number for each arriving packet. Flow-based weighted fair queuing is also discussed in detail in this chapter. For those readers worrying about QoS for interactive voice traffic, the author is careful to point out that WFQ maybe unable to achieve the low-jitter requirements. Therefore, he includes a discussion of WFQ with priority queue in this chapter. A short overview of flow-based distributed WFQ is also included. The latter does not make use of the CPU, unlike ordinary WFQ. The author then outlines the class-based WFQ, which can be implemented in both distributed and nondistributed modes. Priority queuing, which divides queues into subqueues of decreasing order of priority, is also treated. The author also gives an overview of custom queuing, which is a strategy for guaranteeing a minimum bandwidth for each traffic classification. Chapter 5 is an overview of the scheduling algorithms on routers that employ a switching architecture that is not bus-based. In this regard, the author gives a very detailed discussion of the use of Modified Weighted Round Robin and Modified Deficit Round Robin algorithms. The TCP congestion control mechanism and how it deals with packet drops is the topic of chapter 6. Those who are familiar with TCP/IP will find the reading very straightforward, but the author also introduces Weighted Random Early Detection (WRED) and Flow WRED, which permit different RED parameters based on packet precedence, and a scheme for penalizing flows that attempt to utilize more than their allocated queue lengths. The author also discusses Selective Packet Discard (SPD), which assists in the differentiation of `priority' traffic from `normal' traffic. An interesting case study in how to prevent `smurf attacks' by using SPD is included in this chapter. Comment | | (Report this)