This is part 2 of "Dropping legacy 802.11 and 802.11b support from your infrastructure".
To show the differences in speed that you get with 802.11n and legacy 802.11 and 802.11b, I setup a lab that consisted of an a/b/g/n dual radio Cisco AP (1252 model) running on a 5500 series controller. I setup an 11n client (Early 2009 model Mac Book Pro) and an 11b client (Intel 2915ABG), both running Jperf (front end to iperf) to a Linux based iperf server. The AP was gig connected to a Cisco 3750, as was the iperf server. The AP was configured to only support the 11n MCS1-15 data rates and the 1mbps data rate for the 802.11 testing and the 6mb data rate when I moved to testing 11g. (1 and 6 were never enabled at the same time).
The 11n client was kept close to the access point (10' away) and the 11b/g client was kept about 40' away and across 3 walls with drywall/insulation, but well within the range to have a perfect 1mb and 6mb connection. iPerf tests were averaged over 100 second interval tests. Legacy Beamforming/Clientlink was disabled on the AP.
The 11n client on 2.4ghz and being the only associated client to the access point had an average throughput of 88.2Mbit/sec at a connected data rate of 144Mbit (MCS-15).
When the 11b client joins the network but sits idle, the 11n client slows to an average of 75.2Mbit/sec.
When the legacy 802.11 client starts iperf, while set to RTS-Self, the 11n client slows to 19Mbit/sec.
When the legacy 802.11 client starts iperf, while set to RTS/CTS, the 11n client slows to 6.3Mbit/sec.
This would indicate if you have multiple 11n users copying files, and utilizing the 11n network rather heavily at maximum throughput, a single legacy client or client off on the edge of the cell would cause serious impact to their throughput. Now we'll look at how 802.11g impacts the 11n client.
When the 11g client joins the network at a 6mb data rate but sits idle, the 11n client slows from 88.2Mbit/sec to 83.4Mbit/sec, or 8.2Mbit/sec higher than an 11b client.
When the 11g client starts iperf the 11n client slows to 48.6Mbit/sec, or 42.3Mbit/sec faster than 11b at 1mb. (For the record, the 11g client does about 2Mbit/sec while the 11n client is active, and about 4Mbit/sec when the 11n client is absent, while on the 6mb data rate).
As you can see, disabling 802.11 data rates (1, 2) and 802.11b rates (5.5, 11) and leaving 11g rates (6, 9, 12+) enabled increases the throughput of the 11n clients considerably. If you can't run in a complete 802.11n greenfield mode, wouldn't you much rather have your clients sharing 48.6Mbit/sec while an 11g client hangs out at the edge of the cell downloading files, than sharing 19Mbit or even 6.3Mbit while a legacy 802.11 client hangs on at -89 signal for dear life?
If you want to step back a bit and just consider the spectrum utilization, here’s a quick image of the spectrum analyzer when a legacy 802.11 client on channel 1 connected a data rate of 1mb is downloading at an average throughput of 0.4Megabit/sec. You can see the duty cycle is nearly 100%.
The next image is the same view, but of a client connected at 802.11n data rate 144mb, downloading at an average throughput of 88.2Megabit/sec. The duty cycle is nearly the same with a little higher usage to the sides of the center channel, but pushing 220x the throughput of the 1mb data rate in the same RF space. This may be common sense to some, but the pictures may help visualize it for others.
For comparison view, this next image starts off showing the 11g 6mb data rate running iperf and then joined by 11n 144mb data rate client running iperf.
So with a little real world testing and some numbers, you can picture that lone legacy client you have stuck on someone’s desk or in some corner conference room, or in a nook behind some robotic arm assembling your cars, having a serious impact on everyone else sharing the same cell. If you can stop supporting legacy 802.11 protocols, it would be good for everyone else on your network. It’s not unreasonable in some environments to expect your guests to have a wireless card made in the last 4+ years installed on their laptop, or a Wi-Fi enabled phone to be more modern. After all, 802.11g was ratified almost 7 years ago. There are always odd ball cases though, for instance the Nintendo Wii supports 802.11g, but the driver is so poorly written that it still wants to see the 1mb data rate enabled before it’ll associate at all, so for home users with Wii’s, you’re stuck unless you want to get the Ethernet adapter for it. For places that are absolutely trying to have the most flexible network (Like Universities), you may consider making 5Ghz your advertised "fast network" (and possibly use band steering) and everyone else fights for the 2.4Ghz space.
Erik Parker is a wireless network engineer for a Fortune 500 e-commerce company based in the United States. Erik was previously a wireless engineer at Toyota and consulting network engineer for International Network Services (Now BT-INS) prior to that. He has experience with Routing, Switching, Wireless, Security, and Linux systems engineering. His primary focus is on wireless infrastructure, 802.11 protocol analysis, RF, and mobility. Erik's hobbies include arm-chair electronics using Arduino, Parallax, and nearly anything else you can hook random sensors into. Erik has maintained his CISSP designation since 2002, has spoken at multiple Gartner mobility summits, and continues to be active in the wireless community.
This blog represents the personal views of the author and does not necessarily represent the views of his employer.