« 802.11ac Boosts Buzz More than Bandwidth | Main | A PUBLIC APOLOGY »

March 12, 2012

Facing Wi-Fi Reality: Vendors Lie (Surprise, Surprise)

RealityIn the age-old battle for one ups-man-ship, vendors continue to mislead potential buyers with theoretical speeds that look good on boxes and data sheets but are never experienced by expectant customers.

Despite what vendors (including us) spew, the speeds you achieve on your Wi-Fi network in the real world will be lower, sometimes much lower, depending on your circumstances, even with emerging dual-band 802.11n APs that can now support multiple spatial streams in each band, 2.4 and 5GHz.

Assuming the use of wider, 40MHz channels in each band (through channel bonding), the use of the narrower 400ns guard interval (GI), and the ideal channel quality (in other words, a great connection) required to support 64-QAM, a theoretical throughput of 300 Mbps per radio is possible.

A guard interval is a small delay that ensures that distinct OFDM symbols don't interfere with one another. Their purpose is to introduce immunity to propagation delays, echoes and reflections, to which digital data is normally very sensitive. And QAM, or quadrature amplitude modulation, is an analog and digital modulation scheme that conveys data by changing some aspect of a carrier signal, or wave, (usually a sinusoid) in response to a data signal.

The IEEE standard defines a 400ns GI option, but increased bit error rates in practice generally disqualify 400ns guard intervals from utility in the real world of live networks. Operation with an 800ns GI is much more common. The more common 800ns GI would effectively lower peak data rate to 270 Mbps.

With these wider channels and smaller gaps between bits, 600Mbps throughput for a dual-band, two stream 802.11n AP is theoretically achievable. But you’ll never get it. Such theoretical throughput numbers, while enticing, are based on a number of big assumptions.


Reality in the Not-So-Perfect World of Wi-Fi

In most Wi-Fi networks today, the 2.4GHz band is forced to do the majority of the heavy lifting. This is because so few clients — especially existing smart mobile devices that are the source of a rapidly growing majority of traffic — support 5GHz.  While this is changing fast, it’s still the reality for IT managers trying to build fast and efficient Wi-Fi networks.

Each channel within a Wi-Fi network typically consumes 20MHz of spectrum or bandwidth. To achieve higher data rates, one technique used by 802.11n is bonding or combining these channels into larger 40MHz lanes (up to 160MHz channels are being touted by 802.11ac).

Since 40MHz operation in 2.4 GHz is rare (and a bad idea), 20MHz channels are to be expected. With a much more practical 800ns guard interval and 20MHz channel, IT managers are now looking at about 130 Mbps of throughput per radio. But it doesn’t stop there.

Inefficiencies and errors that arise in any real, commercial chipset implementation and radio hardware help to bring the peak down into the neighborhood of 110 Mbps depending on the quality of the radio implementation. Note that this is at zero range and in essentially perfect (double isolation-chamber) radio conditions.  

UDP traffic can come close to the 110 Mbps, but most of the traffic in today’s mobile Internet world is TCP. This delivers a performance decrement on the order of 15 to 20%, yielding a new peak throughput of about 90 Mbps.

So far we’ve been talking about the raw symbol rate (also known as baud or modulation rate) in the over-the-air Wi-Fi protocol. A symbol is either a pulse (in digital baseband transmission) or a "tone" representing an integer number of bits. A theoretical definition of a symbol is a waveform. But these symbol rates exclude the effects of Wi-Fi protocol overhead as bits are disaggregated from and reassembled into layer 2 frames and layer 3 packets on both ends of the link - lowering the actual throughput experienced by users.

Next, most laptops sold within the last couple of years come equipped with 802.11n chipsets supporting two, three or even four spatial streams. However the current crop of smart mobile devices, because of power, design, and size constraints, are typically only equipped with single-stream 802.11n implementations, often of relatively low quality. As a result, users will see, at best, about 30 Mbps for common Android and Apple handsets.
 

What About Wi-Fi Interference?

The real world of Wi-Fi is filled with a lot of signal obstructions such as people, furniture, walls, windows, doors, plants, merchandise on shelves, and so on. This is addition to the sources of active interference such as microwave ovens, cordless phones, security systems and interference from neighboring wireless systems.

Because Wi-Fi is a shared medium, adding lots of clients, all fighting for access to the spectrum from the same AP, brings yet another hit to performance. Basically users must wait to transmit and receive data as other clients are using the Wi-Fi network. If a given client experiences packet loss due to interference or other environmental problems, it must retransmit packets causing others to wait to get on the air thereby lowering performance again.

Co-channel interference is yet another performance degrading problem. With a limited number of non-overlapping channels available in the 2.4GHz band (3) and the use of omni-directional antenna systems that send and receive transmissions in all directions, APs that can hear each other will wait to put clients on the network until the channel is available.  

It is commonly believed that within the 5GHz band, with 21 non-overlapping 20MHz channels, this problem will go away, or at least be mitigated somewhat. And on the face of it there is plenty of spectrum to use in the 5GHz band. But to achieve higher data rates, 802.11n channel bonding (as well as spatial multiplexing) is needed to increase client speeds.

5GHz-chart

When you eliminate DFS channels, the mechanism that ensures that WLANs won't interfere with commercial or military radar, or weather radar equipment, all of a sudden, after bonding 20MHz channels into wider 40MHz channels, there are only four non-overlapping channels available that are not affected by DFS.


Death with Distance

Another performance killer is distance. As with any telecommunications technology, Wi-Fi throughput falls off with distance, as transmit power dissipates in the medium. Dimensioning a Wi-Fi network requires an assumption about how far away the average client will be from the AP, and discounting the peak aggregate throughput accordingly. Directional antenna systems that focus RF energy toward clients to increase signal strength can help mitigate this problem. Stronger signals mean higher data rates for clients as well as less co-channel interference.

Ultimately, understanding these Wi-Fi performance inhibitors is essential to getting what you pay for. There is no one, single panacea to solving these problems. New, adaptive antenna arrays can help but so can advanced functions such as airtime fairness, band steering, client load balancing, and automatic channel selection. When combined together, these techniques improve the end user experience greatly- but never to the level that vendors will promise.

Being careful to clearly understand the difference between throughput claims from vendors and what’s achievable in the real world is fundamental to building a fast and reliable Wi-Fi network. Equally important is being careful to take advantage of the best technology available to improve the odds your Wi-Fi service will deliver what your customers want:  better connections to the Internet and your network in more places.

The best way to ensure that you are going to achieve the utmost performance from perspective vendors is to test. Don’t test it in a lab either as that won’t replicate the real world. Test it in your most challenging locations with as many connected devices as you can muster. This should give you a clear picture of what a given Wi-Fi system really delivers. Only then will you be confident in your purchasing decision. 

 

Comments

Rob S

Fantastic article guys. This is what we like to tell customers to set expectations... It's refreshing to hear it direct from the vendor.
Thanks :)

Paulo Raponi

The big problem is "how to" test this scenarios.

A doubt. The own switch controller is responsible for setting options such as Channel bonding and Guard interval?


regards,
Paulo Raponi

Rachel

Hey,
Thanks! Great post you have written on "Facing Wi-Fi Reality: Vendors Lie (Surprise, Surprise)". Really I can say that your post is very informative, I'll come across your blog again when you will update it with new.
Thanks,
Rachel
http://www.mobilemark.com/

John S

Finally a article dealing with realities of WiFi. Its nice to finally see someone write about the expectations and realities of WiFi.
I think the manufactures of routers do a disservice to buyers by advertising speeds that are typically not achievable in real world operation. People live in neighborhoods or apartments where other routers are also operating nearby. That triggers a good neighbor policy with routers to reduce bandwidth to prevent interference on the 2.4Ghz band. Everything from baby monitors to security cameras to microwaves operate within the 2.4ghz band. Please do not think the 5Ghz will save the day either as the higher the band the shorter the range. Now we will see a new set of standards in 2014 that again promises more throughput and better range. All at a costs and a lack of supported hardware so far. Yes, even today I see a lot of devices only supporting 2.4Ghz and not even the 5Ghz band. Wireless will always have its limitations even though they are trying to improve the speed and range. The end results will be mixed for users. I always consider WiFi more of a convenience over wired then actually being better. I find it very hard to believe I can ever experience constant speeds in WiFi that would exceed wired. They promise but I have not experienced it.

Post a comment

Comments are moderated, and will not appear on this weblog until the author has approved them.

If you have a TypeKey or TypePad account, please Sign In.