How Do You Gauge The Real Speed Of Wi-Fi?
Note: The following article is written from a vendor-neutral standpoint. All limitations discussed are not specific to any one vendor and occur across all platforms.
I previously answered questions on How many devices can my access point support? and What is the range of Wi-Fi?. As a continuation on my FAQ-style series on enterprise Wi-Fi, I am going to address this question: What is the expected throughput I can get using Wi-Fi?
The reason for covering this subject is two-fold:
- It comes as a shock to many Wi-Fi investors when recorded speeds don’t match advertised speeds.
- NBASE-T; you’re being told it is required for 802.11ac Wave 2 (I disagree).
Note: Technical elaboration is provided throughout this article (highlighted with * or †). Understanding of this subtext is not mandatory and is provided only for reference.
- Unlike wired networking, wireless performance is impacted by signal attenuation.
- Wi-Fi communication is half-duplex whereas wired is full-duplex.
- Wireless transmission incurs more overheard than wired, affecting performance.
- Wireless data rates are determined by the lowest common denominator – an AP and client do not always support the same capabilities.
- Data rates vary based on deployment. There is no ‘one size fits all’ deployment methodology; some are geared towards capacity, others towards coverage.
- Wave 2 does not require NBASEs-T.
Wired vs Wireless
I want to start this article by comparing the two primary methods to form a computer network. The first being Ethernet or 802.3 and the other being Wi-Fi or 802.11. When someone purchases a network switch capable of 1Gbps and they plug a 1Gbps cable into that switch, they expect to receive exactly 1Gbps of throughput, no more, no less. Why is it when you buy a Wi-Fi device capable of 1.3Gbps, you see throughput ranging from 100Mbps to 500Mbps (for example)? There are many reasons why, but we are going to start by identifying the key differences between the wired and wireless mediums.
Full Duplex vs Half Duplex
Those that have been in the networking industry for a while will remember the day of the networking hub. The hub was originally designed to link multiple computers together to form a wired computer network. The hub had an Achilles heel; it was half-duplex (clients could not transmit and receive the same time). The more clients added to a hub, the slower the overall network became – every device had to take its turn. In practice this meant the 1Gbps capacity of a Hub was divided between the clients connected to it.
Later, the network switch was introduced. This device was capable of performing rapid packet-switched communication between each endpoint/device – say hello to full-duplex communication! Now, a 10 port 1Gbps switch can transmit at 10x 1Gbps (connecting clients does not directly reduce performance).
Unfortunately, Wi-Fi communication is half-duplex. Why? Quite simply, the wireless medium is shared, impossible to control, and subject to collisions.
Wireless transmission is made possible by utilizing Distributed Coordination Function (DCF). Simplified, DCF maximizes throughput by reducing collisions. This is achieved by invoking an exponential back-off algorithm whereby the client attempts to transmit in randomized intervals. Before attempted transmission, the client scans the wireless channel in use and measures RF energy on that channel.
If other transmissions are detected, the back-off timer is paused and only resumed when the transmission window appears to be clear. Once the timer has expired and the channel remains clear the client will transmit a ‘request to send’ (RTS), announcing it wants to send information. The receiving client responds with a ‘clear to send’ (CTS), equating to ‘I’m ready to receive’. Finally, the data is transmitted. This mechanism is replicated across all clients attempting to transmit across the same channel.
The result somewhat resembles a lottery whereby clients ‘win’ their time-slot to transmit, leading to collision avoidance (CA). * This mechanism is less efficient than full-duplex switching, equating to comparatively less Goodput.
Additionally, similar to a hub, APs share throughput capacity amongst connected clients on the same channel; 1.3Gbps capacity of an AP radio will be divided between clients.
*This is why Wi-Fi uses CSMA/CA as opposed to CSMA/CD used by wired networks.
Now you understand the air is a shared medium, it is time to cover attenuation. Unlike the wired medium, wireless is subject to many forms of attenuation, namely: reflection, refraction, diffraction, absorption, scattering, FSPL etc. If you’ve already read my article What is the range of Wi-Fi? feel free to skip to ‘Overhead’.
Turn on a speaker and position it next to you. The sound is clear. Now, put a pillow between your ear and the speaker. The sound is absorbed and the quality of the signal is degraded. Wi-Fi signals encounter exactly the same type of attenuation. The most common form of absorption is water. Can you think something that contains lots of water? You! People are one of the most common sources of signal absorption, unfortunately, this affects Wi-Fi.
Reflections occur when a signal strikes a smooth object and bounces at the same degree the object was struck. You might think signal reflection sounds bad (and it used to be). Thankfully, with the introduction of 802.11n Wi-Fi, signal reflection (known as Multipath) is actually a good thing. 802.11n radio receivers are capable of translating multiple signals arriving at different times (caused by reflection), reducing the impact that reflections once had and turning them into a performance gain. Unfortunately, reflections still negatively impact wireless range.
Refraction is when the direction of a signal changes as it passes through an obstacle. The best way to visualize this is a light prism. During this process, the strength of the original signal is reduced and subsequently so is the potential range.
Light on air–plexi surface in this experiment undergoes refraction (lower ray) and reflection (upper ray). – Wikipedia.org
Scattering occurs when the originating signal strikes an uneven surface, the resulting weaker signals are reflected in multiple directions. The weaker signals will have a reduced range of the original signal.
This occurs when a signal bends around an object or spreads through an opening. This often occurs in an environment with lots of pillars. The most relatable analogy of diffraction is water. If you push water through a gap, the waves will spread through and outward (see below for an illustration).
An illustration showing the different types of diffraction – https://web2.ph.utexas.edu/~coker2/index.files/diffraction303l.htm
Free Space Path Loss (FSPL)
Signal attenuates over distance. In the same respect that light travels so far before dimming, microwave signals used in Wi-Fi travel so far before the power of said signal is degraded and fades over distance from the source.
At this point you may be thinking ‘The wireless medium is a messy place to transmit data, how do we pull it off’? To guarantee reliable and successful transmissions, safeguards are implemented. These safeguards result in overhead and these overheads eat away at effective Goodput. If we were to use a car as an analogy, a cooling radiator adds weight to a car, inevitably reducing top speed. Without the radiator however, the car would overheat and become undrivable! Similarly, overheads (better said, ‘efficiency mechanisms’) are absolutely necessary and enable efficient wireless communication.
Unfortunately, the result of overheads is reduced throughput when compared to the theoretical maximum. To put this into perspective, the more attenuated the signal, the lesser throughput/data rate. Avoiding a technical deep dive, let’s just say these efficiency mechanisms reduce Goodput by about 50% when compared to the theoretical maximums. Find out more more here.
Another overhead comes in the form of security mechanisms. The wireless medium is unbounded, meaning we have no control over exactly where signals propagate and who can potentially listen to them. This is contrary to a wired medium where the data only travels along the cable (bounded medium). Due to the uncontrolled nature of Wi-Fi, we need additional measures to guarantee only intended recipients receive data and it is not tampered with during transit. Extra security mechanisms facilitate data confidentiality, integrity and authentication. Again, these mechanisms impact effective user goodput.
It Takes Two To Tango
We have spent the first half of this article highlighting the differences between wired and wireless networks but one thing they share in common is that transmission speed is determined by the lowest common denominator. For example, if you have a 1Gbps switch and you connect a client capable of 100Mbps, you’ll only get 100Mbps for that client. Likewise, with Wi-Fi, if you have an 802.11ac AP capable of 1.3Gbps but you connect an iPhone 6S capable of 867Mbps, you will never hit 1.3Gbps for that client! Not every 802.11ac client is capable of 1.3Gbps (we’ll cover this more later).
No Two Deployments Are the same
Are you deploying for coverage or capacity? Depending on which methodology you decide on, varying degrees of performance will ensue.
Let’s look at the requirements to achieve 1.3Gbps 802.11ac speeds:
1. You must deploy 5GHz (802.11ac is 5GHz only)
In the 5GHz spectrum there are 19 available channels in Europe and 25 in the US (varies across regions).** In theory, this means you can deploy 25 APs within radio frequency (RF) propagation range of each other before encountering co-channel interference (CCI). Unfortunately, 16 of those channels require DFS. That leaves 9 non-DFS channels (less in other regions)! ***
What is DFS? DFS is a technology that allows APs to coexist with radar systems. Radar can operate on the same frequency as your AP and this is bad for radar (interference). As such, if your APs detect radar on the channel they are currently occupying, they will switch to a different channel. Although this protects radar communication it is potentially damaging to your Wi-Fi because your APs will be forced change channel, causing a drop in communication (albeit temporary). Depending on your environment, deploying across the DFS frequency space may or may not be a problem.
2. You must deploy 80MHz (non-overlapping) channels
Remember those 9 channels? Those channels are 20MHz wide. We need 80MHz to achieve the top data rate. This means a single AP must combine the channel space of 4x 20MHz to support 1.3Gbps. The resulting limitation is 2 APs per RF propagation area before risking CCI.
** Further 5GHz channels may become available in the future under UNII-2B and UNII-4.
*** Even with DFS enabled, you only have 6 non-overlapping 80MHz channels available. Consequently, an engineer should rarely recommend deploying 80MHz channels as network scalability will be extremely limited.
3. The client must be very close to the AP
In order to achieve the highest data rates, the client must be very close to the AP with strong received signal strength (RSSI) and high signal-to-noise ratio (SNR) (‘Noise’ refers to background interference present across the same frequency). Essentially, unless you are sat directly underneath the AP with no interference, you won’t hit high data rates. †
OS X El Capitan Wireless Diagnostics Tool showing RSSI and Noise.
4. The AP and client must support 3 spatial streams
To hit 1.3Gbps, clients must support 3 spatial steams. Unfortunately, you’ll be hard pressed to find any. To name a few: MacBook Pro Retina, iMac 5K and Dell XPS 15.
Conlusion: The Real Speed of Wi-Fi
To get the full speeds of your Wi-Fi you need to use a deployment methodology that allows for little scalability (80MHz channels result in limited channel reuse), your clients must be the most expensive consumer devices available (3x spatial stream support) and those same clients must be metaphorically kissing the AP! (high RSSI and SNR).
Advertised vs Recorded
So what is the real speed you can expect?
Let’s deploy 20MHz channels to achieve: the best scalability, capacity and range ††. What effect does this have on speed? Unfortunately, this knocks our 1.3Gbps down to 288.9Mbps. Did I also mention this is only attainable if your client supports 3 spatial streams? What if it’s only 2 (iPads)? It’s knocked back even further to 173.3Mbps †††.
†† The wider the channel width, the more demanding the MCS requirements. MCS ranges from 0-9. Higher numbers signify the ability to encode data in a more complicated method providing greater data throughput. MCS 0 (the lowest, basic form of modulation/communication) requires 8dB of SNR when using 80MHz channels. Conversely, MCS 0 requires only 2dB of SNR when deploying 20MHz channels (more info here). To put another spin on this, using 80MHz channels requires -85dBm of signal strength to achieve some sort of data transfer. 20MHz channels on the other hand requires only -91dBm – That’s 200% less power! In short, using 20MHz is less demanding and will allow for data transfer at extended ranges.
††† Figures based on MCS index.
Let’s be generous though and deploy 40MHz channels with all clients supporting 3 spatial streams. Ok, so now our maximum speed is 600Mbps. What about attenuation? As discussed, attenuation would make wireless communication impossible if it wasn’t for the use of overheads. Security mechanisms also add overhead. Unfortunately, the typical result of these overheads is about 50% reduction in Goodput when compared to theoretical maximums. Practically, this means our 600Mbps AP is now reduced to 300Mbps.
Of course, providing an arbitrary Goodput figure serves no effective purpose other than to antagonize other wireless engineers who will find a million reasons to disagree and propose an alternative figure! So, in an effort to deflect those comments – please don’t take these figures literally. Instead, understand that:
- Benchmark configuration vs. deployment configuration are two completely different items that are almost never interchangeable.
- The advertised maximum speed is not representative of user Goodput.
- The Half-Duplex nature of Wi-Fi means the AP throughput capacity is shared amongst connected clients on the same channel.
With this in mind, you should have a solid understanding as to why recorded speeds don’t necessarily equal advertised speeds.
At the start of this article I mentioned NBASE-T (2.5Gbps Ethernet). This new technology is aimed to address the increase in speeds that 802.11ac Wave 2 brings. A typical high-end Wave 1 AP for example is usually capable of 1.3Gbps. Considering the preceding points, it’s understandable this will place no realistic strain on a 1Gbps Ethernet backhaul.
But what about a Wave 2 AP? Some vendors are now pushing Wave 2 APs with the following features: dual 5GHz, 4 spatial streams and160MHz channel support. Referencing the aforementioned discussion, you should come to a similar conclusion that NBASE-T is still not necessary. If you’re still unconvinced, let me summarize:
Don’t get me wrong, dual 5GHz is a good thing. It offers more flexibility, capacity and helps support a 5GHz-only future. Having said that, deploying 2x 80MHz channels not only requires enabling DFS but further reduces scalability (channel reuse). Unless you are benchmarking or have a special use case, dual 80MHz on a single AP is a bad idea.
4 Spatial Streams
3 spatial stream clients are hard to come by. I challenge you to find clients capable of 4 spatial streams (it takes two to tango!)
This is the same principle as dual 80MHz, occupying the same frequency space as 160MHz and requiring DFS to be in operation. Again, this offers little scalability and should only be used for benchmarking.
If you understand the preceding three points you should come to the conclusion that NBASE-T is not needed to accommodate 802.11ac Wave 2.
Consequently, the only AP that will realistically increase backhaul load is one capable of Dual 5GHz. Even then, it’s unlikely to exceed 1Gbps over Ethernet.
Wave 3 combined with regulatory bodies (FCC) opening up channel space may deliver a future where NBASE-T or 10BASE-T become relevant to support Wi-Fi. Regardless, the industry has a long way to go and technology continues to improve in the meantime – more reason not to make an unnecessary investment in NBASE-T today.
At this point, you may be thinking ‘what is the point in Wave 2 if it offers no performance benefit?’ It can do, but in a rather unorthodox manner. Wave 2 introduces an efficiency-focused technology called MU-MIMO (already supported by many client devices), addressing the half-duplex nature of Wi-Fi. It’s cool. Read more here.
All Posts In This Series: