Internetwork Design and Address Assignment

Designing an Internet-compatible network requires both physical design of the internal internetwork topology as well as infrastructure and/or procedures for assigning addresses to the computers in your organization. Many organizations take an ad hoc approach in these areas, sometimes out of ignorance and sometimes because they are under time pressures. Sooner or later, they usually come to regret it, because either performance degrades or reliability suffers. Careful planning will take some time, but it will make your life easier in the long run.

Selecting a Router

Most of todayís corporate TCP/IP networks are built using multiprotocol routers that are capable of supporting TCP/IP as well as other popular communications protocols over the same physical network infrastructure. Choosing a router vendor is a subjective dec ision usually involving trade-offs between cost and performance. Leading router vendors such as Cisco, Bay and 3Com offer high-performance hardware-oriented systems that provide rich functionality, flexibility and scalability. Other vendors, including Novell, offer software-based routers that add routing functionality to personal computers. While hardware-oriented vendors generally offer better performance and an expanded feature set, producers of software-based routers have come a long way in recent years. Increases in processing power coupled with enhancements in personal computer I/O architectures make these systems a viable alternative for many organizations, particularly those that need to deploy large numbers because of physical design issues. Not only are up-front costs lower, but maintenance costs are also lower because you are dealing with commodity hardware components.

Most ISPs will provide a hardware-based router as part of the installation fee. Alternatively, you may be able to add an interface to one of your existing routers. Given its dominant market share, it should come as no surprise that Cisco is the most popular router used by ISPs. For many sites, this is a good reason to standardize on Cisco, since effective administration of your routers is fundamental to maintaining both reliability and security. However, improvements in router management interfaces may allow you more flexibility in this area.

Designing the Corporate Backbone

In addition to selecting appropriate routers to meet your needs, you need to design a corporate backbone infrastructure that provides adequate performance for Internet access. In designing your internetwork, you will need to consider both raw bit rates as well as latency. Most network designers have a pretty good handle on the design issues revolving around raw bit rates. As will be discussed shortly, a range of performance options is available for Internet connections ranging from slow speed 9,600 -bps serial connections to very high-speed 45-Mbps T3 service. Youíll need to match your bit-rate performance needs to your application requirements. In most cases, for an organization with several hundred users, a 1.5-Mb T1 connection will offer the most ìbang for the buck,î but this is, admittedly, only a rule-of-thumb.

While bit-rates are well understood by most network designers, latency-which is a measure of delay in packet processing-often is not. While closely related to bit rate, latency can also result from improper physical network design or underspecification of network routers. All routers will introduce latency into your network regardless of how fast its interfaces are. In general, the more expensive the router, the lower the latency. Latency can also be introduced into a network by poor design. To minimize latency, a basic design principle is to minimize network diameter, or the average number of routers a packet must pass through to reach its destination.

Flat Addressing or Subnetting?

When interfacing an existing network-particularly one that is geographically dispersed-to the Internet, minimizing network diameter can be a significant challenge. For many companies, a single Internet connection provides services to the entire corporate network. Thus, ensuring adequate end-to-end performance, particularly for highly interactive applications, may require careful selection of the Internet connection site. It may even require a fundamental redesign of the corporate wide area network. For single-site networks, the design of the backbone itself becomes critical. Ideally, the Internet connection should reside at the center of the network, assuming equivalent access is required throughout the network.

If your requirements call for the provision of direct Internet access from all desktop computers on your networ k, assigning IP addresses can be a significant challenge. There are two major elements that should be carefully considered in this regard. First, you will need to decide whether to adopt a flat address model or a subnetted address model. Until fairly recently, a subnetted model was almost always preferred because routers were usually employed to minimize traffic levels of individual network segments. However, recent trends toward increased use of switches make a flat address space slightly more appealing.

Manual vs. Dynamic Addressing

In either case, you will need to develop procedures for assigning addresses to computers. At first, it may be tempting to assign these addresses to individual systems manually. However, most large sites have recognized the flexibility associated with a more dynamic approach to addressing through the use of the bootstrap protocol (BOOTP) or the dynamic host configuration protocol (DHCP). Both of these protocols allow a TCP/IP protocol stack to contact a central server at boot time and have an address assigned from this server. In the event that readdressing becomes necessary, this approach offers significant benefits since all addresses are managed via tables on a central server. The alternative of manually reassigning addresses at each workstation can be very expensive.

BOOTP vs. DHCP

In addition to assigning IP addresses, a BOOTP or DHCP server is also commonly used to provide the client with other information, including the IP subnet mask, the default gateway and the addresses of one or more name servers. DHCP provides a superset of functionality in comparison to BOOTP, including the provision of temporary address ìleasesî for mobile computers and, in the long run, a variety of other network and server configuration information. While DHCP is a more robust protocol, it is also relatively new, so finding clients and servers that support it is more difficult than is the case for BOOTP . But this situation is likely to improve, if only by virtue of the fact that Microsoftís Windows95 TCP/IP stack only supports DHCP.

Implementing a Domain Name Server

The implementation of a Domain Name Server (DNS) is one of the other challenges faced by organizations connecting to the Internet. By mapping logical addresses (for example, www.microsoft.com) to numeric IP addresses, these servers are an essential component in making Internet services more accessible to users. Since implementation of a DNS is one of the blackest of Internet arts, many sites that initially connect to the Internet depend on their ISP for this capability. In the long run, however, most sites find the performance benefits associated with maintaining their own DNS to be worth the cost of implementation and maintenance.

Next

Return to Table of Contents

November 15, 1996