Network Computingwww.networkcomputing.com

During the migration to routed networks, you may recall that almost every machine in the network needed to have its IP address changed. The flat, bridged network didn't care which IP address was where. With a routed network, your IP addresses need to be split into subn ets and then assigned to individual router ports across the network.

Adding switches to the fray can "flatten" your network and change the routing topology. Let's look at Network Computing's lab at the University of Wisconsin Computer-Aided Engineering (CAE) Center. The network was part of the 144.92.x.x address space, subnetted at 25 bits. It accommodated up to 128 IP addresses per routed segment, including the FDDI backbone, which only housed routers and servers prior to migrating to switches.

To decrease latency and increase bandwidth to the servers, CAE added switches that bridge from Ethernet to FDDI. With this addition, any machine that was switched required an IP address that was part of the backbone's subnet. CAE quickly ran out of the 128 IP addresses (144.92.4.0 to 144.92.4.127) assigned to the backbone and needed another solution.

Instead of adding arbitrary IP subnets to the backbone, CAE requested a la rge block of contiguous IP subnets from the campus assignment center. Currently, the FDDI backbone and all machines on switches use 144.92.240.0 subnetted at 22 bits, allowing 512 machines on the backbone's subnet (144.92.240.0 to 144.92. 43.255). If you don't have Class B address space, you can get consecutive Class Cs (for example, 208.210.8.0 to 208.210.11.0 would have a "supernet" mask of 255.255.252.0). Or, if your Internet service provider (ISP) supports Classless Inter-Domain Routing (CIDR) addressing, you can obtain an arbitrary block of consecutive addresses (for more on CIDR, see www.lab.unisource. ch/services/internet/direct/cidr.html).

Keep in mind that this solution requires that if you use variable length subnet masks (VLSMs), you must use either static routing or a routing protocol (like OSPF or EIGRP) that supports VLSM. Distance vector protocols like RIP and IGRP do not support VLSM. In the final solution, the routed, Ethernet ports still have the 25-bit subnet mask, while the backbone ha s widened to 22 bits. As the network migrates to more switched ports, CAE will need to consider adding more addresses to the backbone.

As you move in this direction, consider VLAN functionality. With an entirely switched network, you may have broadcast traffic--previously suppressed by routers--saturating your network. Using VLANs to contain broadcast domains will lighten that load and give you added flexibility in network topology. (For more on IP addressing and subnetting, see Network Computing's Network Design Manual at www. NetworkComputing.com/netdesign/outline.htm.)

Chris Lewis is vice president of international operations at ILX Systems. He is currently working in Europe. He can be reached at chrisl@ilx.com.