Dude, Where’s My Throughput?

It’s been a long time since our last post. We have been very busy working on the next wave of products, and we are very excited about what's coming.

Ever since our launch of Wireless-AC access points, our integrators have been deploying thousands of networks throughout the world, and we couldn't be more thankful for your trust and partnership. Talking to numerous integrators and listening to our support calls, there is one major concern we noticed. When it comes to Wi-Fi speeds, there is a major misunderstanding between what technology manufacturers advertise and what consumer expect. Unfortunately, the integrator is at forefront of this dilemma daily. So here's an attempt to clear things up for you, the integrator, so you can set expectations when talking to consumers.

Two facts we can't ignore: Internet Service Providers (ISPs) have increased their speed offerings to consumers, and consumers' expectations have increased dramatically when it comes to their Wi-Fi experience. Almost every major networking equipment manufacturer has a product or two that boasts phenomenal Wi-Fi speeds: “1200Mbps,” “1750Mbps,” or “come here for 2600Mbps!” But when the integrator sells those Wi-Fi speeds to the consumer, reality hits after deployment. Speeds are not near what has been advertised, dead spots still exist, and buffering is the kiss of death when the consumer paid for 300Mbps+ from the ISP. To explain this mismatch, we will examine two issues: the numbers game, and clean Internet.

Networking equipment manufacturers (ourselves included) have fallen into what we call "the numbers game." There is no better way to explain this than a real-world scenario. Let's take a 3x3:3 Wireless-AC access point, for example. The advertised speed for this product is 1750Mbps. Pretty great when you have 450Mbps coming from the ISP, right? Before you get excited, let's break down how manufacturers got the 1750Mbps number. The math is simple: we have 3 streams on 2.4GHz and 3 streams on 5GHz radio interfaces. Per 802.11ac standard, on 2.4GHz radio, when a device establishes a connection with the access point using 40MHz channel bonding (MCS7), then the stream maximum throughput is 150Mbps. We have 3 streams on 2.4GHz, so this equals 450Mbps on 2.4GHz radio. Similarly, (again per 802.11ac standard) when a device establishes a connection with the access point using 5GHz radio and 80MHz channel bonding (MCS9), then the stream maximum throughput is 433.3Mbps (5GHz has more throughput than 2.4GHz). We have 3 streams as well on 5GHz, so this equals 1300Mbps on 5GHz. Now since the access point is dual-band concurrent, combine the maximum speeds on both interfaces and voila! We have 450Mbps + 1300Mbps = 1750Mbps!

Combined Throughput

 

The main problem with this math - while technically correct - is that it's not realistic. Let's take a different view from what’s to be expected in reality. Starting with 2.4GHz interface, since the spectrum is very congested and the number of non-overlapping channels on the 2.4GHz spectrum is limited to 3 channels each of which is 20MHz wide, there are no devices that negotiate 40MHz channel bonding on 2.4GHz. The probability of frame retransmits is very high when using 40MHz on 2.4GHz radio due to noise or interference. So achieving 40MHz channel bonding on 2.4GHz ­– while technically possible and tested in labs – is virtually impossible in real-life. Devices opt for 20MHz channel bonding only, which results in the case of 3 streams to a maximum of 195Mbps (remember it was 450Mbps in the ideal state). In addition, Wi-Fi as a technology is a shared medium. In other words, it's a half-duplex medium. In other other words, at one point in time only one device is talking: either the access point to a device or a device to the access point. This means the maximum 195Mbps is effectively a maximum of 97Mbps realized by the device (it has to listen half of the time assuming no one else is talking to the access point). The fun doesn't stop here, this 97Mbps is assuming the wireless medium is used 100% of time to transmit content (e.g. YouTube video stream), but that's not the case. Since it's a shared medium, Wi-Fi is notorious for having management traffic overhead in order to control who is talking at one point of time and who gets to talk next. This management overhead increases dramatically by three factors: number of SSIDs advertised on an access point, number of access points in the location, and the number of Wi-Fi devices in the location. The more of these three factors, the more coordination and management messages need to be sent around occupying valuable airtime that could be used to transmit actual traffic. Also, interference causes faulty frames to be received, which triggers retransmits and slow down mechanism, lowering the effective throughput. In addition, the maximum throughput assumes clear line of sight and a very close distance between the device and the access point. The further the device moves away from the access point, the lower the throughput gets. The same logic and limitations go to 5GHz radio.

So what does that mean to the effective real-life throughput? In order to answer it, we have to flip the perspective of the above figure to be from the end device trying to connect to an access point. All the  aspects listed below dramatically affect the throughput of the device:

  1. Radio Interface: Big difference between 2.4GHz and 5GHz from a nominal speed perspective
  2. Number of streams: It's a two-way street between the device and the access point. Most mobile devices have 2 streams only.
  3. Channel Bonding: If the radio interface is 2.4GHz, 20MHz is the realistic channel bandwidth. If 5GHz, we start with 80MHz but can automatically fall back to 40MHz if the environment is noisy
  4. Half-Duplex Nature: By design, Wi-Fi is half duplex, so whatever the maximum negotiated throughput, it's cut in half because of this fact.
  5. Wi-Fi Overhead: The more access points and SSIDs exist in a location, the less effective the wireless medium is.
  6. Wireless Contention: The more Wi-Fi devices trying to connect to an access point, the longer a certain device waits before it gets a chance to talk.
  7. Distance: Also known as signal strength, which happens naturally as the signal degrades with distance as well as obstacles in the location.
  8. Interference: All sources of interference severely affect the effective throughput.
  9. Legacy Devices: Because of the half-duplex nature, if there is an old device talking at 54Mbps on the network, all other devices will wait longer for the old device to finish talking.
  10. TCP Acknowledgements: If the connection is TCP, it adds a lot of overhead to the medium because every packet needs to be acknowledged by the receiver.
  11. Retransmits: If one bit of a frame is corrupted due to collision, the entire frame needs to be retransmitted

Funnel

This is a list of things that eat from the nominal throughput and vary widely by each location or time frame. This results in rules of thumb throughput figures rather than concrete numbers, more of an average than a measurement. We find that a good 2.4GHz environment provides an effective throughput between 20Mbps-40Mbps, while a good 5GHz environment provides 150Mbps-400Mbps effective throughput. These are big deltas from the advertised maximum theoretical numbers by networking equipment manufacturers, which are again, technically not wrong.

So the next time a customer asks you: I just signed up for 300Mbps internet, will I get 300Mbps across the house via Wi-Fi? The answer is: technically yes, but realistically not likely.

 

Next, we visit the pipe going into the location and explore the other question: Am I really getting 300Mbps of Internet?