Still more implications for innovation, improved performance, and why site surveys continue to be mostly a waste of time
Interop Las Vegas 2010 is now a wrap, and I must say I was impressed with both the level of attendance (up markedly over last year) and the level of debate. As is often the situation, the best discussions occur outside the meeting rooms, and such was certainly the case here. But the debate in the sessions was also intense (if always polite) from time to time, and an analogy really struck me – modern enterprise-class wireless LANs have an awful lot in common with modern jet aircraft, starting with the number of microprocessors involved, and extending well into control systems that are now essential to the mission and capable of far more than network operators or pilots could ever do on their own.
Next time you fly, try to sit at a window near the wing, and just glance, every once in a while, at the control surfaces moving slowly, if almost randomly, as the plane maintains smooth and level flight or responds to turbulence. That’s dozens of microprocessors at work, doing their job to maximize your comfort (and likely fuel efficiency as well) without the pilot being directly involved at all. And when the pilot is involved, say, initiating a turn to the left, again those dozens of microprocessors go to work, carefully adjust those same control surfaces in more of an interpretive than a directly-responsive fashion. The pilot says what to do; the control system determines how. Flying a modern aircraft would be impossible without this technology.
The same is true of modern enterprise-class WLANs as well. I think it’s safe to say that manual configuration of anything more than a few APs would be futile. And then consider that 802.11n, as was discussed in Lisa Phifer’s excellent session on upgrading to .11n in the enterprise, exhibits radio propagation characteristics that are a long way from those convenient circles of coverage that we always use. In fact, these patterns look more like those comic-book explosions, with spikes all over the place, and may even have areas of coverage separated by clear gaps! And, to make matters worse, these patterns can and usually do change from moment to moment, creating all kinds of opportunities for self-jamming and degraded performance. The control plane of a given implementation thus has a lot of work to do, and constantly. Yes, we’re still debating the centralized vs. distributed control argument, along with the benefits of single-channel architectures and direct-forwarding mechanisms. Add in innovations like beamforming and Cisco’s CleanAir, and you have the analogy I’m presenting – forget about manual configuration, or even trying to understand, let alone manually optimize, with the system is doing at any given moment in time. We can only evaluate the performance of an implementation by its results under given sets of conditions – and, unless we can carefully construct a benchmark, that means checking those management logs weekly or so to see what additions or changes might be helpful. I expect control planes to eventually make suggestions here as well (i.e., “add another AP in Sue’s office”). I think it’s safe to say that the data plane should be mostly distributed, the management plane absolutely centralized, and the control plane – well, this is where the opportunity is now.
And also, once again, forget about the formal, expensive site survey. These are, and I must disagree here with the opinions of at least one speaker in the conference, even more useless than before, for the reasons noted above – you’ll just end up optimizing nothing. I still recommend a pre-installation RF sweep for interference, and simple propagation studies in the case of pathological buildings. But if the control plane can’t optimize for a given and changing set of operational circumstances, it may be time for either a firmware upgrade or a detailed heart-to-heart with your vendor.
