We tend to spend a lot of time on the specifics of radios when talking about wireless products, systems and services of any form. Sure, radio is tough, and the techniques we often apply to improve the performance of radio can be obscure. But equal attention needs to be paid to how the radios are connected to the rest of the network. This is the role of wireless LAN system architecture.

You might have thought that by this point, a "right" way to design a WLAN architecture would have been discovered. But you would be wrong. Indeed, it appears that diversity in this area is increasing and innovation is far from over.

You also might have thought that we would have converged on a standard WLAN architecture, because that's what happened with the wired LAN. We went from thick Ethernet to thin Ethernet, and from there to hubs and switches - and the switch has carried us architecturally as Ethernet progressed from 10 Mbit/s to 1 Gbit/s. Sure, WLAN switches are also the most visible, but here we see much greater diversity than we do in the wired world.

But let's roll things back a little and start with the root of WLAN architectures, the access point, or AP. In concept, the AP is simply a bridge between some number of wireless stations and a wired backbone network. In execution, there has never been any industry-wide agreement as to what functionality should be implemented in the AP.

Thin is in
Traditional (sometimes called "fat") APs work as stand-alone bridges. It was quickly realised that such an approach would prove unworkable in large configurations because centralised management, security and cross-subnet roaming needed to be implemented. This was often done via third-party gateways, which are still available from Bluesocket, Cisco Systems (for its Aironet series of products) and others. But the core trend has been toward "thin" APs connected to wireless switches. These switches work much like other Ethernet switches but contain the necessary management and control logic required in a wireless LAN.

A variant of this architecture implements the switch as more of an appliance and allows APs to be connected to intermediary switches and thus potentially to be very widely distributed. Regardless, the switched architecture is now the dominant model for enterprise-class WLANs.

But what kind of thin?
But even here we see a great deal of diversity. One can make the argument that moving all traffic from all APs through a single switch will result in potential performance bottlenecks and that similarly, centralising all security processing will also be detrimental to throughput. It must be noted here that there is obviously a difference between architecture and implementation, and both must click to get a good result. But it seems as though such arguments as those above are in fact resulting in a greater diversity of WLAN architectures, rather than the convergence we might have expected by now.

For example, Meru Networks and Extricom avoid the use of a traditional cellular deployment of APs, which involves APs near one another operating on orthogonal (nonoverlapping) radio channels. Instead, they put all APs on the same radio channel and claim benefits of enhanced reliability through antenna (really radio) diversity and greater overall throughput.

And there's another major approach, offered by Meru in its Radio Switch product line and by Xirrus. It's called the wireless LAN array. The idea is to concentrate a great deal of potential throughput in a single location via a package that might be described as a super access point or perhaps a collapsed-backbone switch with up to 16 (in Xirrus' case) integrated APs.

In the end, of course, most buyers aren't going to select a product based on its architecture, but rather on its features, performance and, of course, price. So ultimately, we don't compare architectures, but rather specific implementations of them. Still, the vendors vigorously defend their architectural choices. Our job, then, is to decide how to compare the resulting performance - another big challenge that I'll cover shortly.

Craig J. Mathias is a principal at Farpoint Group, an advisory firm specialising in wireless networking and mobile computing. He can be reached at [email protected] This article appeared in Computerworld.