I had the privilege of chairing the keynote panel at Interop this year, and I made an almost-offhand comment about how site surveys, often used as part of the planning and installation process for wireless LANs, are mostly obsolete.

I've held this position for a very long time (since at least 2004), and I will explain myself below, but I was surprised at the number of calls and e-mails I got on this subject. A few were from installers who still thought I was trying to ruin their businesses with fraudulent advice. But I also got a couple from a WLAN manufacturer that were easily resolved by clearing up the definition of exactly what a site survey entails.

I can't believe this topic still generates so much interest. A site survey, if you've not done one before, is the process of determining the "optimal" location for an access point (AP). The general process is to set up and power on an AP, but without connecting it to the backbone network. One then walks around with a notebook computer or handheld running site-survey software and notes the readings of signal strength at various locations. A little back-end work, and voilá, we know where the APs should go. Except for one small detail: we really don't - at least when relying on the above labor-intensive and often-expensive exercise.

Surveys are a relic
The problem with site surveys is that they're a relic from the era when WLAN APs were very expensive (more than $2,000 for 1 Mbit/s to 2 Mbit/s), and we had to therefore ration them, optimising for coverage over capacity. This would be a terrible idea today, when APs cost a few hundred dollars and offer 54 Mbit/s and (with MIMO) even more.

The strategy today should focus on capacity, not coverage. So a labour-intensive activity (and engineering labour) that optimises for the wrong variable and bases its recommendations on a snapshot in time not even involving real traffic or actual workloads or throughput provisioning requirements in a given location (number of users and the bandwidth their applications need) is borderline, um, crazy.

It's also a waste - in most cases, and in my humble opinion - of both time and money. I occasionally recommend site surveys for pathological buildings (mostly hospitals and some hotels) or post-installation to understand pathological behavior, but that's it.

But let's not confuse site survey with pre-installation planning, which is absolutely required. I recommend the following steps:

  • Make a quick survey of the building if it's anything other than the common open/modular-office configuration, especially if the building is older, just to identify any coarse-grained RF propagation issues (dense walls, etc.).
  • Survey users to determine the number of people in a given location and their throughput requirements, and plan AP placements accordingly.
  • Perform an initial installation, and see what kind of coverage and throughput are available. You can subsequently add any fill-in APs required to compensate for errors in placement because of propagation-related issues, or simply not fully understanding user throughput and coverage requirements (hey, it happens; the question is, how much money to spend to deal with this problem).

And it really is impossible to determine the number and placement of APs upfront, because that's just the way radio works. It's statistical, nondeterministic and often counterintuitive.

Modern RF spectrum management systems, built in to essentially all current enterprise-class WLAN systems, will automatically do the best they can to set channel assignments, transmit power levels and otherwise adjust to reality, even as it changes. And that's just what we want - more automation, less labor and an optimal (if incrementally obtained) result.

But keep in mind that needs will indeed change over time. Plan on adding infrastructure as users and their throughput demands evolve. In that respect, wireless is no different from wire.

One of the big issues, by the way, sometimes used to justify a site survey is interference, which really can be a big problem. Or is it? I'll look into interference issues next week.

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