Testing this was a problem. The switch creates a virtual cluster of servers and distributes requests to the servers based on which server is least heavily loaded. The traffic tends to be very different from the sorts of heavy streams of one-way traffic used to test basic switch functionality. Using software products running on PCs is a typical way of getting the types of traffic necessary to test load-balancing functions, but it would take thousands of PCs to generate sufficient loads to overwhelm one port, let alone test the increase provided by the virtual matrix.

After conferring with engineers from Adtech, Ixia, and Netcom, we were able to use the traffic generators to create traffic more characteristic of a Web site, where a session is initiated by a client, a response is generated by a server (with both client and server simulated by the traffic generator), a connection opened, maintained for a period of time, and then closed. Using these tests, we were able to determine that the virtual matrix dramatically increased the performance of the switch in load-balancing functions.

Although we didn't have enough time and equipment to determine the upper limits of the increase, we verified an increase of more than double the number of connections per second.

While we learned some useful lessons in our test, the bigger question is whether Gigabit Ethernet over copper is a usable, worthwhile technology. The answer is a provisional yes. In another six months, the wrinkles should be ironed out with drivers. There should also be more products from which to choose, thereby lowering costs. Switches should also be available with higher port densities than are available with fiber.

However, the cabling issues remain. A Cat 5 cabling plant is unlikely to work well with gigabit speeds, and refitting the cabling plant to make it work well is likely to be as expensive as putting in fiber. In fact, the Cat 6 standards being touted by some vendors are more expensive to install than fiber.

Fiber is cheaper than it used to be and provides substantial advantages over copper, both in immunity to electromagnetic interference and in an upgrade path to 10 Gigabit Ethernet.

While it's tempting to go for the lower cost of copper and the theoretical ability to use existing wiring, the systems that need gigabit speeds are servers or high-performance workstations being used for rendering or other high-bandwidth applications. Given that these applications are likely to be critical to a business, is it worthwhile to take a chance on degradation in network performance to save a little money on equipment? Probably not.

If things get to the point where Gigabit Ethernet to the desktop is a standard and network planners must contemplate running hundreds of switch ports and NICs, copper would be a good bet.