Introduction

• Campus networks, which consist of locally connected users in a building or group of buildings

• Wide-area networks (WANs), which connect campuses together

• Remote connections, which link branch offices and single users (mobile users and/or telecommuters) to a local campus or the Internet

Despite improvements in equipment performance and media capabilities, internetwork design is becoming more difficult. The trend is toward increasingly complex environments involving multiple media, multiple protocols, and interconnection to networks outside any single organization's dominion of control. Carefully designing internetworks can reduce the hardships associated with growth as a networking environment evolves.

This chapter provides an overview of the technologies available today to design internetworks. Discussions are divided into the following general topics:

The distinct characteristic of a campus environment is that the company that owns the campus network usually owns the physical wires deployed in the campus. The campus network topology is primarily LAN technology connecting all the end systems within the building. Campus networks generally use LAN technologies, such as Ethernet, Token Ring, Fiber Distributed Data Interface (FDDI), Fast Ethernet, Gigabit Ethernet, and Asynchronous Transfer Mode (ATM).

A large campus with groups of buildings can also use WAN technology to connect the buildings. Although the wiring and protocols of a campus might be based on WAN technology, they do not share the WAN constraint of the high cost of bandwidth. After the wire is installed, bandwidth is inexpensive because the company owns the wires and there is no recurring cost to a service provider. However, upgrading the physical wiring can be expensive.

Traditionally, Layer 2 switching has been provided by LAN switches, and Layer 3 networking has been provided by routers. Increasingly, these two networking functions are being integrated into common platforms. For example, multilayer switches that provide Layer 2 and 3 functionality are now appearing in the marketplace.

Network designers are now designing campus networks by purchasing separate equipment types (for example, routers, Ethernet switches, and ATM switches) and then linking them together. Although individual purchase decisions might seem harmless, network designers must not forget that the entire network forms an internetwork.

WAN links are connected by switches, which are devices that relay information through the WAN and dictate the service provided by the WAN. WAN communication is often called a service because the network provider often charges users for the services provided by the WAN (called tariffs). WAN services are provided through the following three primary switching technologies:

• Circuit switching

• Packet switching

• Cell switching

• Necessity to connect to remote sites

• Growing need for users to have remote access to their networks

• Explosive growth of the corporate intranets

• Increased use of enterprise servers

Because there are so many remote single users or sites, the aggregate WAN bandwidth cost is proportionally more important in remote connections than in WAN connections. Given that the three-year cost of a network is nonequipment expenses, the WAN media rental charge from a service provider is the largest cost component of a remote network. Unlike WAN connections, smaller sites or single users seldom need to connect 24 hours a day.

Consequently, network designers typically choose between dial-up and dedicated WAN options for remote connections. Remote connections generally run at speeds of 128 Kbps or lower. A network designer might also employ bridges in a remote site for their ease of implementation, simple topology, and low traffic requirements.

To date, local- and wide-area communications have remained logically separate. In the LAN, bandwidth is free and connectivity is limited only by hardware and implementation costs. The LAN has carried data only. In the WAN, bandwidth has been the overriding cost, and such delay-sensitive traffic as voice has remained separate from data. New applications and the economics of supporting them, however, are forcing these conventions to change.

A well-designed internetwork can help to balance these objectives. When properly implemented, the network infrastructure can optimize application availability and allow the cost-effective use of existing network resources.

The goal is to minimize cost based on these elements while delivering service that does not compromise established availability requirements. You face two primary concerns: availability and cost. These issues are essentially at odds. Any increase in availability must generally be reflected as an increase in cost. As a result, you must weigh the relative importance of resource availability and overall cost carefully.

You can assess user requirements in a number of ways. The more involved your users are in the process, the more likely that your evaluation will be accurate. In general, you can use the following methods to obtain this information:

• User community profiles—Outline what different user groups require. This is the first step in determining internetwork requirements. Although many users have roughly the same requirements of an electronic mail system, engineering groups using XWindows terminals and Sun workstations in an NFS environment have different needs from PC users sharing print servers in a finance department.

• Interviews, focus groups, and surveys—Build a baseline for implementing an internetwork. Understand that some groups might require access to common servers. Others might want to allow external access to specific internal computing resources. Certain organizations might require IS support systems to be managed in a particular way according to some external standard. The least formal method of obtaining information is to conduct interviews with key user groups. Focus groups can also be used to gather information and generate discussion among different organizations with similar (or dissimilar) interests. Finally, formal surveys can be used to get a statistically valid reading of user sentiment regarding a particular service level or proposed internetworking architecture.

• Human factors tests—The most expensive, time-consuming, and possibly revealing method is to conduct a test involving representative users in a lab environment. This is most applicable when evaluating response time requirements. As an example, you might set up working systems and have users perform normal remote host activities from the lab network. By evaluating user reactions to variations in host responsiveness, you can create benchmark thresholds for acceptable performance.

The negative side is that IGRP is a proprietary routing protocol. In contrast, the integrated Intermediate System-to Intermediate System (IS-IS) protocol is an open internetworking alternative that also provides a fast converging routing environment; however, implementing an open routing protocol can potentially result in greater multiple-vendor configuration complexity.

The decisions that you make have far-ranging effects on your overall internetwork design. Assume that you decide to implement integrated IS-IS instead of IGRP. In doing this, you gain a measure of interoperability; however, you lose some functionality. For instance, you cannot load balance traffic over unequal parallel paths. Similarly, some modems provide a high level of proprietary diagnostic capabilities, but require that all modems throughout a network be of the same vendor type to fully exploit proprietary diagnostics.

Previous internetworking (and networking) investments and expectations for future requirements have considerable influence over your choice of implementations. You need to consider installed internetworking and networking equipment; applications running (or to be run) on the network; traffic patterns; physical location of sites, hosts, and users; rate of growth of the user community; and both physical and logical network layout.

If the tested internetwork's characteristics are close to the new internetwork, you can try extrapolating to the new internetwork's number of users, applications, and topology. This is a best-guess approach to traffic estimation given the unavailability of tools to characterize detailed traffic behavior.

In addition to passive monitoring of an existing network, you can measure activity and traffic generated by a known number of users attached to a representative test network and then extrapolate findings to your anticipated population.

One problem with modeling workloads on networks is that it is difficult to accurately pinpoint traffic load and network device performance as functions of the number of users, type of application, and geographical location. This is especially true without a real network in place. Consider the following factors that influence the dynamics of the network: