Network Processors

Architecture, Programming, and Implementation

Ran Giladi

• Publisher: Morgan Kaufmann
• Year: 2008
• Print ISBN: 9780123708915
• eBook ISBN: 9780080919591
• Topic: Technology & Engineering
• Subtopic: Electrical Engineering & Telecommunications
• Index: Technology & Engineering

Network processors are the basic building blocks of today's high-speed, high-demand, quality-oriented communication networks. Designing and implementing network processors requires a new programming paradigm and an in-depth understanding of network processing requirements. This book leads the reader through the requirements and the underlying theory of networks, network processing, and network processors. It covers implementation of network processors and intergrates EZchip Microcode Development Environment so that you can gain hands-on experience in writing high-speed networking applications. By the end of the book, the reader will be able to write and test applications on a simulated network processor.- Comprehensive, theoretical, and pracitical coverage of networks and high-speed networking applications- Descirbes contemporary core, metro, and access networks and their processing algorithms- Covers network processor architectures and programming models, enabling readers to assess the optimal network processor typer and configuration for their application- Free download from http://www.cse.bgu.ac.il/npbook includes microcode development tools that provide hands-on experience with programming a network processor

CHAPTER 1 Introduction and Motivation

Network processors (NPs) are chips—programmable devices that can process network packets (up to hundreds of millions of them per second), at wire-speeds of multi-Gbps. Their ability to perform complex and flexible processing on each packet, as well as the fact that they can be programmed and reprogrammed as required, make them a perfect and easy solution for network systems vendors developing packet processing equipment.

Network processors are about a decade old now and they have become a fundamental and critical component in many high-end network systems and demanding network processing environments.

This chapter introduces this relatively new processing paradigm, and provide a high-level perspective of what NPs are, where to use them and why, and conclude with a brief description of the contents of the rest of the book.

1.1 NETWORK PROCESSORS ECOSYSTEM

Telecommunications and data networks have become essential to everything that we do, to our well-being and to all of our requirements. The prevalence of Internet technology, cable TV (CATV), satellite broadcasting, as well as fixed and cellular mobile telephony, tie many and expanding services to a very large population that is growing exponentially. The increasing speed of the communication links has triggered a wide range of high-speed networks, followed by an increasingly broad spectrum of services and applications.

We are witnessing this dramatic growth in communication networks and services; just think of the changes that networks and services have undergone in the past 10 years in the areas of mobile, video, Internet, information availability, TV automation, multimedia, entertainment, online services, shopping, and multiplayer games. You can safely assume that an equivalent jump in technology and services will happen again in the next 5 years or so.

Networks and infrastructures have enabled it all. Most houses, vehicles, pieces of equipment, and people, maintain a communication link to the “network,” a giant octopus with zillions of arms. And the oxygen that runs in its veins, pipes, and trunks are packets and cells; zillions of zillions of them are flying around us and surrounding us at every moment. These packet flows undergo various treatments, processing, and forwarding, in many kinds of network devices. These network devices are systems by themselves, and they keep the octopus, or the “network,” alive. Such network systems include switches, adapters, routers, firewalls, and so on.

Network systems, therefore, face an ever-increasing magnitude of packets they have to handle, while at the same time, the processing of these packets becomes more and more complex. This creates a gigantic performance problem. In order to cope with it, vendors have replaced the traditional general purpose Central Processing Unit (CPU) in the network systems with Application Specific Integrated Circuits (ASICs), which are hardwired processing devices. However, as vendors also face rapid changes in technology, dynamic customer requirements, and a pressure for time-to-market, short developing cycles and lots of revisions have become necessary. All of this has required an innovative approach toward network systems architecture and components.

This is the foundation on which NPs flourish. NPs enjoy the advantages of two worlds—they have the performance capabilities of an ASIC, and the fast, flexible, and easy programmability of a general-purpose CPU. NPs constitute the only option for network systems developers to implement dynamic standardized protocols in performance-demanding environments.

1.2 COMMUNICATION SYSTEMS AND APPLICATIONS

Communication systems are composed of networks and network devices (which are sometimes referred to as network elements, or network systems, as we called them above, since they are computerized, special purpose systems that include both software and hardware).

Communication networks can be separated into three main categories: the core, the aggregation (or metro), and the access networks. Each of these network categories are characterized by different requirements, technologies, and equipments.

In addition, there are traditionally two communication systems we use: telecommunications (telecom) and data communications (datacom). The more established and older network, the telecommunication network, is based on “circuits,” or channels of continuous bit streams, which grew out of its original application to telephony services. The more recent networks, the data communications networks, were initially based on packets of data to carry information among computers. This resulted in two paradigms of network technologies—circuit switching and packet switching.

The main technology in telecommunication systems is the Synchronized Optical Networks/Synchronous Digital Hierarchy, which is used in both the core and the aggregation networks. Radio Access Networks through which most mobile telephony is conducted, as well as the wireline telephony access networks, are usually attached to these networks, using their own technologies. CATV is a parallel network that is traditionally used for TV services.

The main technologies in the datacom systems are the Ethernet and the Internet Protocol (IP). For the last two decades, converged telecom–datacom networks have been subject to vast research, industry implementation trials, and services. This convergence happens in the technology plane as well as in services: recent Telecom networks’ cores are implemented using datacom technologies (such as Ethernet, Multiprotocol Label Switching [MPLS] and IP). Recent trends in converged services, starting with “triple-play” service (or “quadruple-play” for Internet, TV, telephony, and data-oriented services) to Voice over Internet Protocol telephone services and TV over IP, are just a few examples of the transition to a unified, converged network.

The result is that packet networks have become the prevalent technology for communication systems. Network systems are making the transition from circuit-switched based technologies (multiplexers, cross connects, branch exchanges, etc.) to packet-switched based technologies (such as bridges and routers).

Services are also developing in scale and complexity; from plain telephony, TV, and Internet surfing and e-mails, we are now facing High Definition TV, new generations of web and web services, digital libraries, and information availability, gaming, and mobile 3G and 4G services that include information, multimedia, video, and TV, and many other demanding applications. Other adjunct services, such as security, provisioning, reliability, and accounting must be supported by network systems, which may impose additional significant load on them.

As networks become the infrastructure for information, interactive data, real-time data, huge multimedia content transport, and many other services described previously, the technology of networks must cope with various requirements, but primarily that of speed. High-speed networking refers to two aspects of speed: the links’ transmission rates—from multi Mbps (10⁶ bits per second) to multi Gbps (10⁹ bits per second)—and the complexity and speed of the required processing due to the number of networks, addresses, services, traffic flows, and so on.

If we examine the speed of network links over the years, we find a similar but higher growth pattern than that of processing capabilities. In computing, this exponential growth is modeled as doubling roughly twice every 2 years (after Moore’s law);¹ however, in the last decade this growth rate has shrunk, and is roughly at 41% annually (with clock speedups increasing only 29% annually). [116] If we look, for example, at Ethernet bandwidths, we find a ×10⁴ speedup in 27 years (from 10 Mbps approved in 1983 to 100 Gbps expected to be approved in 2010, as shown in Figure 1.1), which is doubling the bandwidth every 24 months. However, if we examine the increase from 100 Mbps (approved in 1995) to 100 Gbps (×10³), it is doubling the bandwidth every 18 months. This pattern of growth is similar to that ...

Table of contents

• Cover image
• Title page
• Table of Contents
• The Morgan Kaufmann Series in Systems on Silicon
• Copyright
• Dedication
• Preface
• Acknowledgments
• Chapter 1: Introduction and Motivation
• PART 1: Networks
• PART 2: Processing
• PART 3: A Network Processor: EZchip
• List of Acronyms
• References
• Index