Predicting Wi-Fi Performance: Can It Be Done?
Whenever I plan and configure an enterprise WLAN system, I always advise that an additional 10% to 20% of the capital budget be reserved for expansion and filling in holes in coverage or capacity that couldn't be anticipated during the planning process.
The fundamental uncertainty (not, as some occasionally refer, unreliability; the air doesn't break) associated with a statistical medium like wireless practically dictates this strategy, disconcerting though it may occasionally be. But wouldn't it be great if we could more accurately (if not precisely) predict radio performance, coverage, throughput and other factors in advance for any given installation?
The standard technique for Wi-Fi planning and deployment is the site survey, a decidedly practical approach to dealing with the uncertainty inherent in wireless. The basic idea is to turn on an access point (AP), place it somewhere in the area to be covered and measure the signal strength at various locations. It's a time-consuming and often expensive process, as professionals are involved, and it unfortunately has a serious flaw: site surveys only consider coverage, and not capacity.
We'll come back to this point. But, for the moment, wouldn't it be great to replace this trial-and-error with something a little more analytical, and perhaps even less expensive in the bargain?
A number of WLAN system vendors do include fairly robust planning tools in their repertoire. The general procedure for using these tools is to import an AutoCAD (.dxf) or similar file of each floor of one's building into the planning tool (note: this drawing will be used later for management and operations anyway). Then you fill in the parameters that the particular tool requires.
These usually include such elements as building construction properties, so that radio behavior with respect to wall, doors and other physical elements can be estimated, and an indication of the type and amount of traffic in a given area. The math involved here is, of course, extremely complex, and, again, we're still dealing with statistics. While these tools will usually produce very pretty output (including bills of material and other information for the installers), showing exactly where APs should be placed, it is fair to ask how accurate this radio simulation really is.
I've found that accuracy depends upon the skill of the engineer (yes, a professional is still required), but it can be pretty good with a little time and careful entry of data. I still think adjustments to the configuration will always be required, but an RF planning exercise can cut down on the amount of error that otherwise always exists.
Have a look, just for example, at Motorola's LANPlanner and Trapeze's RingMaster, which are good examples of this capability. But also have a look at Meru Network's Wireless Virtual reality app, which obtains performance data in an operating network.
Planning, after all, isn't enough; you need to use available management tools to determine what's really working -- and what still needs attention. Keep in mind, again, that RF planning is only as good as the data fed into the system. I've found that absolutely every detail relating to RF propagation, user traffic loads (and time-boundedness requirements), and location-specific demand can't always be determined in advance. Plus, these variables can and will change over time.
And, finally, it's increasingly becoming a good idea to use a spectrum analyzer to do an "RF sweep" of a given area, looking for potentially harmful interference; planning tools can't take this into account. Some WLAN products will eventually build this functionality in, eliminating (or, at least, managing) this additional variable.
I've yet to do a side-by side comparison of different installation techniques, but I'm itching to do so -- that's the only way we'll know whether RF planning, site surveys or simple trial and error yield the best and/or most cost-effective results. But no matter how much up-front simulation and analysis you do, plan on stepwise, incremental refinements and on buying additional equipment over time. Hey, it's radio, and no simulation or analysis will ever be perfect.