Intranet Journal: Intranet Management

Using Dynamic Host Configuration Protocol
Chapter Excerpt from Windows NT TCP/IP Network Administration

By Craig Hunt & Robert Bruce Thompson

Why DHCP is Needed

DHCP has become a practical necessity for large IP networks for two reasons. First, each host in a TCP/IP network must have a unique IP address. This simple fact has caused a tremendous amount of aggravation and extra work for network administrators, and has resulted in more than a few crashed networks. In the early days of TCP/IP networking, there was no automated alternative; you had to assign an IP address manually to each host. Even today, many networks continue to use manual assignment and tracking of IP addresses.

Assigning IP addresses manually is practical only for small networks. As the size and complexity of the network increases, using manual IP address assignment becomes increasingly unworkable. Each time a workstation, server, network printer, router or other host is added or relocated, someone must determine a valid IP address, ensure that that IP address is not already in use by another host, record the assignment of that address, and then finally configure the host manually for that IP address. This process requires expert staff time and is always prone to error. Accidentally duplicating an IP address will at best cause a communication failure on both affected hosts. At worst - if the duplicated IP address belongs to a server, router or other critical network component - the duplicate IP address may cause the entire network to crash. Microsoft TCP always checks to see if its address is a duplicate by issuing an ARP before using the address.

The second motivation for using DHCP is that the perceived shortage of IP network addresses has made it necessary to use IP host addresses more efficiently. Only a few years ago, getting a Class C Network Address (256 IP addresses) was a matter of simply asking InterNIC to assign one to you. Requests for as many as 16 contiguous C blocks were routinely honored by InterNIC without much formality. If you said you needed it, they gave it to you. Even getting a Class B Network Address (256 C blocks, or 65,536 IP addresses) required minimal paperwork and justification.

Nowadays, it's a struggle to get InterNIC to assign even a single Class C Network Address. Getting multiple C blocks assigned requires spending hours or days completing detailed justifications, network plans, and so forth. Getting a B block assigned is almost impossible unless you are applying on behalf of a Fortune 500 corporation, and even then it's not a foregone conclusion.

The large granularity of Network Addresses - a C block is the smallest unit that can be assigned - means that many IP addresses are wasted. Consider a small branch office with a router, a server and 7 workstations. If that branch office is assigned a Class C Network Address, only 9 of the available 256 IP addresses are in use. The remaining IP addresses cannot be used except at that branch office, and so are wasted. In the past, this didn't much matter, because Network Addresses were free and were easily available from a seemingly inexhaustible supply. Some large companies with many small remote sites wasted 90% or more of the many IP addresses assigned to them.

Network addresses are assigned by InterNIC on a first-come, first-served basis, which means that there is absolutely no correlation between Network Address and geographic location. For example, InterNIC assigned to Triad Technology Group, Inc. (Thompson's company, located in Winston-Salem, NC) the Class C Network Address 204.238.30.0. The Network Address immediately preceding that one, 204.238.29.0, is assigned to Warner Brothers Imaging Technologies in Sherman Oaks, CA. The Network Address immediately following that one, 204.238.31.0, is assigned to the Bead Gallery in Juneau, AK.

A side effect of this policy has been the explosive growth of routing tables. Each individually assigned Network Address requires a routing table entry in every router on the backbone. A contiguous block of, say, 16 Class C Network Addresses assigned to the same network requires only a single routing table entry. Those same 16 C blocks, if assigned individually to different companies (and different networks), require 16 individual routing table entries. As of early 1997, the routing tables on the Internet backbone have grown to more than 30 MB

InterNIC strongly encourages you to use IP addresses assigned to you by your Internet Service Provider (ISP) rather than applying directly to InterNIC for your own block of addresses. They do so both to avoid wastage of IP addresses and to slow the growth of routing tables.

However, there is a downside to using addresses provided by your ISP, and you won't hear either InterNIC or your ISP talking much about it. Addresses provided by your ISP belong to the ISP rather than to you. This means they aren't portable. If you decide to change ISP's, you have no option but to recast your IP address assignments network-wide to use the addresses provided by your new ISP. In effect, using addresses provided by an ISP locks you into that ISP.

At first, InterNIC simply recommended that you use ISP-provided IP addresses, but that didn't accomplish much. Most administrators were concerned about address portability, and so simply continued to apply to InterNIC when they needed additional Network Address blocks. Seeing this, and still determined to slow the growth of routing tables, InterNIC next began warning applicants for Network Address blocks that there was no guarantee that individually assigned blocks would be routable in the future.

Apparently, this hasn't worked either, because InterNIC now proposes to charge for directly assigned IP addresses. Under this proposal, any organization to which InterNIC directly assigns a Network Address must pay a $1,000 annual fee, with additional charges assessed based on the number of IP addresses assigned. If this proposal is implemented, you will see the wholesale abandonment of Class C Network Addresses. Almost everyone will use Network Addresses provided by his ISP.

So, what relationship exists between the source of your IP addresses and using DHCP? Simply this. Implementing DHCP on your current network will allow you to recast your IP addressing much more easily when (not if) you find yourself switching to addresses provided by your ISP. If you are using DHCP when the time to recast arrives, you will need to change only the DHCP server configuration and the few static addresses assigned to servers and routers, including the DHCP server. If you are not running DHCP, you will need to change the IP configurations individually for each machine on your network.

[ Next: How DHCP Works]
[ Previous: TOC ]

The Authors

Craig Hunt is an expert on TCP/IP and the author of the best seller TCP/IP Network Administration. Robert Bruce Thompson is an exper on Windows NT and the author of several books on NT including the recently released Windows NT Server 4.0 for the NetWare Administrator.

To purchase, click here...