Whenever the word "performance" is mentioned, at least in a networking context, the first thought to come to mind is, of course, throughput. This is only natural – demands for throughput only grow over time, and we're always trying to get just a little bit more out of our network infrastructures regardless. When we talk about performance, however, wireless introduces two key concerns, as follows:
- Rate vs. range: Unlike the case with wire, the performance of wireless is highly variable, a result not just of network loading but also of the prevailing radio environment (which can and does vary, even from moment to moment), the motion of wireless clients and even otherwise-unrelated people and objects, building construction, and the natural fading of radio signals that increases with the distance between transmitter and receiver. Wireless LANs are designed to compensate for these elements as best they can, but radio will always be an, um, "statistical" medium, nevertheless. No matter what, more distance between access points or between users and the WLAN infrastructure is seldom a good idea.
Tips for benchmarking wireless LANs
- The key to running valid benchmarks is to keep the test conditions as constant as possible between runs. Change only one variable at a time, whether it's the distance between clients and access points, router settings, or benchmark type and duration.
- Whichever benchmarking tool you choose (which could be as simple as a file copy operation or a common application), use multiple runs and average them so as to minimize the effect of transient interference and other radio-wave propagation issues.
- If possible, place mobile computers on turntables in order to minimize the effect of antenna orientation. Be sure to turn any power-conservation features off in this case, as you'll be running on batteries.
- If possible, monitor for interference, from other WLANs as well as other devices operating in the unlicensed bands. Pick an otherwise unoccupied radio channel.
- Time-boundedness: Normally, we allow for a certain degree of headroom in the design of any communications network that is going to be moving time-bounded traffic like voice telephony and streaming video. This is to provision enough capacity to handle time-bounded traffic up to a given limit in terms of number of calls, number of streams, voice or image quality and so on. And on wire, anyway, this is fairly easy to do. We call this technique over-provisioning, and it's why voice calls (usually!) work on the Internet – there's almost always enough capacity so that delay through the network is minimal. Wireless, for the reasons noted above, cannot easily do this, however. Sometimes the quality of the radio channel is good, and sometimes, well, it isn't. Note that upper-layer protocols are always important for time-bounded traffic, and these are identical on wired and wireless LANs.
These factors can make the design of any enterprise-class wireless LAN a challenge. Note also that benchmarking can be very complex on wireless-LAN systems, making comparisons between vendors difficult. For more on this, see Farpoint Group Technical Note 2006-314.1, Benchmarking Wireless LANs: Recommended Practice.
The obvious solution to the wireless performance problem is to improve the performance of the physical layer (PHY) – the radio itself. And that's what the new IEEE 802.11n standard is all about, using advanced radio techniques and particularly the use of multiple transmitters and receivers (and multiple antennas) on each end of the connection. 802.11n-based products will offer between four and six times the effective throughput of 802.11a and .11g, or provide greater throughput at a given range. And the reliability of .11n-based connections is also improved, meaning that time-bounded performance will benefit from the additional capacity inherent in 802.11n.
In the meantime, I recommend using dense deployments, over-provisioning your environment by packing access points a little closer together. No matter what, though, WLAN performance will be improving significantly over the next few years and will as a consequence allow wireless LANs to become the primary -- and even default -- access in commercial environments both large and small.
About the author: Craig Mathias is a principal with Farpoint Group, an advisory firm based in Ashland, Mass., specializing in wireless networking and mobile computing. The firm works with manufacturers, enterprises, carriers, government, and the financial community on all aspects of wireless and mobile. He can be reached at firstname.lastname@example.org.