Gigabit to the desktop: Foundry get real

By Kevin Tolly
Network World, 06/12/00

"Slide a blade into your . . . [Foundry] switch . . . slip a NIC into your desktops, and stand back. Because in three easy steps, you've turned a 10/100Base-TX trickle into a gigabit tidal wave. Without pulling an inch of wire." Voila. Gigabit to the desktop.

According to a new Foundry Networks ad (NW, May 22, page 39), adding a Foundry switch to your Category 5 network "brings 1000Base-T performance [i.e., gigabit] to the desktop."

Sorry, but it's not that easy.

First, let's rap the knuckles of the ad copy writer. Because if Step Three is "stand back to see your masterpiece in action," Step Four should be "prepare your resume."

Although it's not Foundry's fault (the vendor's switch can process streams at gigabit speeds), the company needs to be reprimanded for raising hopes for desktop throughput well beyond reasonable expectations.

Clearly, this is a case of a marketing department run amok. To my knowledge, this is the first time a vendor has recast Fast Ethernet as a trickle. That alone should tip readers that the hype machine is being cranked up too high.

The biggest impediment to bringing Gigabit Ethernet to the desktop is . . . the desktop. Limitations of the PCI bus, disk access speeds and CPU power keep even a machine optimized for network functions far below the gigabit level. Let's not forget the operating system. The 32-bit Windows operating system isn't known for having a highly efficient TCP stack or network subsystem. Windows NT machines we've tested have typically peaked at a few hundred megabits per second with no disk I/O or processing taking place.

Gigabit to the server is not much more viable than gigabit to the desktop. The dearth of published test results is largely because servers have similar hardware and software constraints as desktops. Throwing CPU and memory at the problem will only get you so far. When will we see Microsoft servers pumping data out at gigabit speeds? Not in this hardware generation.

The server is, by definition, on the receiving end of many connections, and it makes sense to take advantage of whatever extra bandwidth you might get beyond 100M bit/sec. But don't forget that Fast Ethernet link aggregation technologies can also provide fractional gigabit connections at very low prices.

If you take the gigabit route, are you sure your Cat 5 can run gigabit without unacceptably high bit error rates? I'd advise getting that cable tester out.

According to folks in the know, a lot of existing Cat 5 may be well below specifications. Inferior-quality cable or connectors, sloppy jobs assembling connectors, too many "untwists" at the cable ends - even the acts of pulling cable through conduit or tightening cable ties - can magically turn your Cat 5 into Cat 3. Such "in field" downgrades matter little when you run 10M bit/sec or, in many cases, even Fast Ethernet. It's another story when you reach gigabit speeds.