Ethernet NICs: Are they really a commodity?

A search of network gear supplier CDW’s Web site shows that Fast Ethernet PCI NICs can be had for as low as $11 (Linksys) and upwards of $70 (Intel) – with Fast Ethernet server adapters in the $80 to $100 range (more, typically, if outfitted with a fiber-optic interface). So what are the differences?

At Circuit City last week, a glance to one side stopped me dead in my tracks – it was a Fast Ethernet PCI network interface card for $4 and change. When a LAN interface costs about the same as a combo meal at Wendy’s, you have to wonder – is the NIC just a faceless, nameless commodity?

Like pork bellies, soybeans, wheat and light crude, is buying NICs all just about volume and price? And, with more and more PC and notebook vendors using integrated “LAN on Motherboard” approaches, you’ve got to ask yourself whether there is even a point in being concerned about the differences. Is anyone really going to change his PC preference because of the LAN interface? Not likely.

Still, there are significant differences among NICs. And, while these differences might mean little to home consumers, they can have a big impact on overall manageability and total cost of ownership in corporations.

A search of network gear supplier CDW’s Web site shows that Fast Ethernet PCI NICs can be had for as low as $11 (Linksys) and upwards of $70 (Intel) – with Fast Ethernet server adapters in the $80 to $100 range (more, typically, if outfitted with a fiber-optic interface). So what are the differences?

A set of features important to both client and server NICs is “manageability.” It’s not hard to find NICs that label themselves “managed,” but it is often a challenge to determine what exactly vendors mean by it. It would seem that some of them define managed as being “not NOT managed.” That is, once they implement any function that can be construed as letting some element of the NIC be managed (like assigning a locally administered address), voilà – they have a managed NIC.

But really, it means having features like “Wake on LAN” (WoL) or support of the pre-execution environment (PXE). WoL lets the NIC trigger a system boot on receipt of a specific packet. PXE allows for special management software to load and execute before the Windows boot loader gets control of the CPU.

Less esoteric and more useful on a day-to-day basis are management features that help isolate and debug connectivity problems. While many of these functions are available via arcane command-line interfaces, having them easily accessible via a GUI will simplify the job for the help desk and let some users fix their own problems.

Vital signs are essential. These displays tell the user about the addressing and operational status of the NIC. If the NIC doesn’t have an IP address, for example, you often need to look no further to solve your problem. Features that allow “loopback” at Layer 2 and Layer 3 (i.e., “ping”) also can come in handy.

Server adapters offer up an even broader array of options based, unsurprisingly, around performance.

The more subtle performance features, though, are destined to help as well – and here it is all about “offload.” Even “commodity-class” server NICs are offering some basic offload of checksum calculation, TCP segmentation, IPSec termination and the like. Formal and informal testing of these features by The Tolly Group have shown that these translate into demonstrable benefits especially when pushing through Fast Ethernet to Gigabit Ethernet levels.

So for enterprise nets, don’t be lulled into thinking that all NICs are created equal.