This is important: what system functions should go where in a wireless LAN architecture? Particular focus on how data moves and where the control for that and other functions resides will ultimately determine the success of particular applications in given installations. And we don't yet have the data we need to know for sure.
Motorola today announced the latest advance in its WLAN architecture, dubbed WiNG5. In a nutshell, the key additions here are an extension to the “adaptive AP” model the firm had previously been applying, as well as a control plane that can move from a WLAN controller to an AP and back again. They even did a demo at the announcement of 84 notebooks displaying independent unicast data streams, winning (on the spot, in fact) a place in the Guinness Book of World Records. This is an interesting implementation – distributed data (via direct forwarding from an AP without having to send the data through a controller), centralized management (and it would be difficult to argue for any other approach here), and either a centralized or a distributed control plane. Centralized has the benefit of a single entity knowing all WLAN network state for optimal resource-allocation decisions, and distributed works best in remote/branch-office (and similar) applications where state needs to be evaluated locally but a full controller might be overkill.
OK, so the question is how to determine the value in this particular strategy, and the key to this is in understanding the value contributions of the three functional planes I noted above. I think it’s impossible to argue with the value of centralized management. It would be ridiculous to return to having to manage each functional unit separately. Indeed, I see wired-network management, wireless-LAN management, and wireless-LAN assurance (which includes a broad range of functions, most notably IDS/IPS, interference detection/classification/localization/ remediation, compliance analysis, among many others) merging, creating a unified management strategy that will pay for itself in reduced operating expense. Motorola already has a great deal of this via the AirDefense product line. Add multi-vendor support here, and that’s as good as it gets apart from additional specific feature enhancements – so much is possible here that the mind boggles, but there’s no question in mine as to what will happen here.
In analyzing the other two planes, though, things get more complex. The direct-forwarding data plane, in which data goes directly from an AP to its destination rather than necessarily though a controller, seems pretty intuitive, especially where time-bounded traffic is involved. Eliminating the network capacity bottlenecks between AP and controller, and/or via multiple (and sufficiently powerful) controllers, however, there is no necessary difference in performance. But the idea of minimizing the data capacity of the controller itself is sound, and this approach will be increasingly influential. Simpler and most cost-effective upgrades and additions should also be an obvious benefit over time.
That leaves the control plane, and Motorola has with today’s announcement staked out an interesting position. The control plane is the OS of a WLAN, directing the data plane in concert with the policies defined by the management plane. It’s here that we’ve seen the greatest divergence in thought, with everything from very robust controllers supporting centralized data to entirely distributed control with no physical controller. Who’s right?
While I suspect that Motorola will be able to broadly demonstrate real advantage from WiNG5, we need two pieces of knowledge, apart from customer testimonials, which are also valuable, to know for sure. The first is realistic analytical models of WLAN architectures, mathematical simulations of the performance of an implementation of a particular architecture. This technique was used for years in computing – until the microprocessor shifted the economics to the point that there was no return from a detailed analysis of this type. The other is results of apples-to-apples empirical testing – real-world benchmarks of large-scale deployments. Unfortunately, neither of these is entirely feasible today, although the former could make a dandy PhD thesis for someone, assuming the cooperation and support of the WLAN system vendors. We can only hope.
In the meantime, the debate will continue – and it’s getting more complex. I’m convinced, though, that we’ll have both the models and enough empirical evidence over the next few years that such won’t always be the case.
