Why is WiFi performance only 60-70% of what they claim?

[firesong]

Found a series of old but still relevant short reads for those who don't want to read the long Duckware. I still recommend the Duckware for better understanding because it's pretty comprehensive.

Some quick reasons why:

1. Interference - Shared medium/airspace. Neighbour crosstalk matters.
2. Headers (Ethernet, TCP overheads of 5%)
3. Beacon Frames - always sent at the lowest data rates
4. Retransmissions - because sometimes packets get transmitted wrongly
5. WiFi is not Full-Duplex - it's Half. Transmissions happen one way at a time - acknowledgements are breaks in transmissions.

The two short reads:
https://www.cwnp.com/wi-fi-overhead-part-1-sources-of-overhead/

https://www.cwnp.com/wi-fi-overhead-part-2-solutions-to-overhead

 

[firesong]

Of course, debunking the myth of large router numbers is important - essentially knowing your theoretical 2x2 hardware throughputs at 80/160/320Mhz will matter.

2.4Ghz:
802.11n 20 MHz - 300Mbps, around 200Mbps in ideal conditions.

5Ghz:
802.11ac 80MHz - 866 Mbps, arouns 500-600 Mbps in ideal conditions.
802.11ax (Wifi 6) 80MHz - 1200 Mbps, around 800-900 Mbps in ideal conditions.
802.11ax (Wifi 6) 160MHz - 2400 Mbps, around 1400-1600 Mbps in ideal conditions.

Just remember one thing though, doubling the channel width also increases the noise floor by +3dB, so it reduces the reliability of the connection.
If you are using DFS channels, also know that those are automatically lower-power by design, so it does increase your risk of retransmission because of interference, essentially reducing network performance.

 

[xiaofan]

Adding on top of the above.
802.11be (Wifi 7) 320MHz/4096QAM - 5764 Mbps, around 3200 - 4000 Mbps in ideal conditions for the 6GHz band based on currently available testing results. But it may be even faster with MLO.

We may have to wait for Windows 24H2 to see real performance of Intel BE200/BE201 and other PC WiFi adapters. Traditionally they are a bit faster than the mobile phone WiFi performance.

 

[xiaofan]

> 802.11ax (Wifi 6) 160MHz/1024QAM - 2400 Mbps, around 1400-1600 Mbps in ideal conditions.

It is actually possible to approach 1.8Gbps based on the test results I have seen.

Non standard
802.11ax (Wifi 6) 160MHz/4096QAM - 2882 Mbps, but still around 1400-1800 Mbps in typical ideal conditions, as it is more difficult to get 4096 QAM to work in the 5GHz band than 1024 QAM.

 

[hwzlite]

And adding to the pain, the WIFI has historically been pretty abysmal at latency...

The lag sources in WIFI are by no means limited to bufferbloat, but buried deep in multiple layers of queues in
the wireless networking stack , which pretty sums up quoting from "Airtime Based Queue Limit for FQ-CoDel in Wireless Interfaces" by Google Engineers implemented on Ath10k based Google WIFI :


"In order to support the wireless protocol’s MAC layer operation such as frame
aggregation, subframe retries and power saving mode, there are multiple layers of queues in
the wireless networking stack: in mac80211, in the host driver, and in firmware for some
architectures. Unless queues in all layers are properly managed, they suffer from long latency
due to bloated queues in one or more layers"

"The newer 802.11ac chipset trend to offload a large portion of data processing tasks,
such as transmit scheduling and frame aggregation, to the firmware running in the
microprocessor inside the wireless chipset. As a result, deep queue often builds up in firmware,
and this additional layer of unmanaged queue become the new source of the bufferbloat
problem"


Check out my rant on fq_codel/ATF/AQL optimized MediaTek mt76 mac80211 driver on a quiet neighborhood's DFS channel 64 @80Mhz, with Line-Of-Sight, sub-5ms lantency pings

 

[xiaofan]

And adding to the pain, the WIFI has historically been pretty abysmal at latency...

The lag sources in WIFI are by no means limited to bufferbloat, but buried deep in multiple layers of queues in
the wireless networking stack , which pretty sums up quoting from "Airtime Based Queue Limit for FQ-CoDel in Wireless Interfaces" by Google Engineers implemented on Ath10k based Google WIFI :


"In order to support the wireless protocol’s MAC layer operation such as frame
aggregation, subframe retries and power saving mode, there are multiple layers of queues in
the wireless networking stack: in mac80211, in the host driver, and in firmware for some
architectures. Unless queues in all layers are properly managed, they suffer from long latency
due to bloated queues in one or more layers"

"The newer 802.11ac chipset trend to offload a large portion of data processing tasks,
such as transmit scheduling and frame aggregation, to the firmware running in the
microprocessor inside the wireless chipset. As a result, deep queue often builds up in firmware,
and this additional layer of unmanaged queue become the new source of the bufferbloat
problem"


Check out my rant on fq_codel/ATF/AQL optimized MediaTek mt76 mac80211 driver on a quiet neighborhood's DFS channel 64 @80Mhz, with Line-Of-Sight, sub-5ms lantency pings

Good point about latency, it is better to use Wired connection for latency senstive applications, like Gaming.

But we may not need to be overly worried about bufferbloat, especially since the test server plays a big part and the wireless test results may not be really applicable for typical use cases.

Reference:
https://forums.hardwarezone.com.sg/threads/fyi-a-bufferbloat-101.6427979/page-11#post-150213993

imho I wouldnt be overly concerned over bufferbloat.
too many factors will affect the result.
Further to that, in our local context, not many ppl can overwhelm a 1gbps connection under normal usage

wavefoam test servers may not be able to saturate our 1gbps connection.
if your wifi speed is lower than your internet speed, you wont be able to saturate the connection either.

Most consumer routers are not powerful enough to support 1gbps QOS as it isn't hardware accelerated.

 

[eighthours]

On this topic, I am using M1 500Mbps plan, WiFi via a single Asus Router GT-AX6000. My DNL speed is 237.55 Mbps and UPL is 413.56 Mbps. Any reason for this huge difference?

 

[xonix]

On this topic, I am using M1 500Mbps plan, WiFi via a single Asus Router GT-AX6000. My DNL speed is 237.55 Mbps and UPL is 413.56 Mbps. Any reason for this huge difference?

Test server and timing ? And is this base on a single test or an average of multiple tests ?

The server you are testing against and during peak or non peak hours will make a difference sometimes.

 

[Henry Ng]

yes, the maximum i can get on wifi 6 is 1600mbps only. Nothing more.

 

[xiaofan]

In the end, Duckware "Understanding Wi-Fi 4/5/6/6E/7 (802.11 n/ac/ax/be)" is still a good read.
https://www.wiisfi.com/

Or the MCSI Index table here
https://mcsindex.com/

 

[xiaofan]

Then if you are into Maths, you can watch this Youtube video to understand the formula to calculate the table.

 

[BradenHeat]

so the key takeaway : all the house technie who needs to troubleshoot or recommend router at certain prices, can point to this to the normies, and tell them either read themselves or stop asking BEST OF THE BEST lulz

and why some end clients [ laptops/ devices/ mobile phones] cannot hit more is due to restraints [ good ] to not burn your device from the 4x4 max speed and then hog the airwaves with a mere 1mbps gaming or streaming nia

 

[alex22]

so the key takeaway : all the house technie who needs to troubleshoot or recommend router at certain prices, can point to this to the normies, and tell them either read themselves or stop asking BEST OF THE BEST lulz

and why some end clients [ laptops/ devices/ mobile phones] cannot hit more is due to restraints [ good ] to not burn your device from the 4x4 max speed and then hog the airwaves with a mere 1mbps gaming or streaming nia

And always marketing needs to have the best looking numbers to sell a product while being backed by science plus the methodology used; via the superscript characters and footnotes just to stay on the legal side of things.

It's a nice balance

 

[xiaofan]

6GHz will be an interesting thing to watch.

Enterprise also uses 6GHz now (more on WiFi 6E now, WiFi 7 will probably only really catching up in 2025/2026)


More about 6GHz in Europe (Singapore is pretty much in line with continental Europe).
https://blogs.arubanetworks.com/solutions/wi-fi-6e-in-europe-frequently-asked-questions/
https://www.qorvo.com/design-hub/blog/connectivity-q-and-a-whats-next-wi-fi-6-6e-in-european-union

+++++++++++++++
It is also interesting to see that large Enterprise Deployment is quite conservative in terms of channel bandwidth used.
https://www.ekahau.com/blog/channel-planning-best-practices-for-better-wi-fi/

Summary
We know we covered a ton in this blog and hope it is a wealth of information for you as you consider your channel plans. Want the TLDR? Here is a short summary of some recommendations for each frequency band:

2.4 GHz – Use 20 MHz wide channels only and use channels 1, 6 & 11.

5 GHz – Use 20 MHz wide channels in highly congested networks. If your environment allows, use the widest channel you can without causing contention on your network. Use all the available 5 GHz channels for your region that your client devices support.

6 GHz – If the region you are planning for has access to 1200 MHz of 6 GHz spectrum, use 80 MHz wide channels. For anywhere else that has less available 6 GHz spectrum at around 500 MHz – use 20 or 40 MHz wide channels.

And remember, avoid using mixed channel widths in 5 and 6 GHz, as this can be a big cause of Wi-Fi performance degradation!

 

[BradenHeat]

And always marketing needs to have the best looking numbers to sell a product while being backed by science plus the methodology used; via the superscript characters and footnotes just to stay on the legal side of things.

It's a nice balance

Whelp, at least in general most of the networking has been simplified and not being overly complex, to deter people from trying to understand. Boy oh boy do i hate topic that involved the 7 layers of nonsense, go one big round without summary the best course of action.

Typical smoke screen from boomers who need to hold on to power. like some major group.


now its still back to hardware + software + coding + A.I optimization on the fly.

Existing tweaks are there, but its really not opened up and explored fully yet.

we can have a 10Gbps Fiber home, but if the international is only measly 2.5, and the flow is on average 1.2gbps fiber to other countries,


its still a waste of resource and learning time for all.

 

[Mach3.2]

Typical smoke screen from boomers who need to hold on to power. like some major group.

I think calling it smoke screen is a little disingenuous. A bit like statistics where you can beat the numbers until it show what you want people to see, so it's really up to the end user to be discerning of all the marketing speak.

All those theory are the backbone of networking, a lot of consumer products hide the low level implementations and present things to the end user using simple GUI, but behind the scenes, all those theory still apply.

That said, you don't have to learn all those theory just to setup your network at home.
See it as learning how to fish, vs buying the fish from the fisherman.

You can still eat fish if you buy it from the supermarket, no need go fish yourself innit?

 

[xiaofan]

we can have a 10Gbps Fiber home, but if the international is only measly 2.5, and the flow is on average 1.2gbps fiber to other countries,

its still a waste of resource and learning time for all.

Technology just moves and I will not say it is a waste of resouce and learning time.

10Gbps Fibre and the corresponding 10G capable WiFi 7 router will take some time to be the main stream (say in 2027/2028 in Singapore). At that time maybe there will be no more 1Gbps plan and the minimum will be 2.5Gbps plan.

In the end, we do not need to be in the cutting edge. I am not going to subscribe to 10Gbps plan now. I am not going to buy a 10G capable true WiFi 7 router either. Rather I will wait then both become the main stream.

Still that does not prevent me from learning more about 10G and WiFi 7.

 

[xiaofan]

I think calling it smoke screen is a little disingenuous. A bit like statistics where you can beat the numbers until it show what you want people to see, so it's really up to the end user to be discerning of all the marketing speak.

All those theory are the backbone of networking, a lot of consumer products hide the low level implementations and present things to the end user using simple GUI, but behind the scenes, all those theory still apply.

That said, you don't have to learn all those theory just to setup your network at home.
See it as learning how to fish, vs buying the fish from the fisherman.

You can still eat fish if you buy it from the supermarket, no need go fish yourself innit?

Well said!

Actually I have seen quite some people who are quite technical yet they do not mind using Singtel ONR and Singtel provided mesh solution, even though I tell them that both are lousy. Reason is simple, it is good enough for their simple use case at home. Their technical interests are not in home networking but rather other aspects.

Then some of them may listen to my advice and change to a different ISP and get a better Asus router. But that is it. They will not tinker much with the Asus router just set and forget. Again, Their technical interests are not in home networking but rather other aspects.

 

[Henry Ng]

Whelp, at least in general most of the networking has been simplified and not being overly complex, to deter people from trying to understand. Boy oh boy do i hate topic that involved the 7 layers of nonsense, go one big round without summary the best course of action.

Typical smoke screen from boomers who need to hold on to power. like some major group.


now its still back to hardware + software + coding + A.I optimization on the fly.

Existing tweaks are there, but its really not opened up and explored fully yet.

we can have a 10Gbps Fiber home, but if the international is only measly 2.5, and the flow is on average 1.2gbps fiber to other countries,


its still a waste of resource and learning time for all.

Well to be honest, we do have the capability to connect to other countries at up to 10Gbps. However, most websites are commercial websites or those websites are own by some businesses or private companies. Most such businesses will want to have maximum profit so they will not spend too much to upgrade to support 10Gbps speed and some files on their website may be quite small and do not need to have 10Gbps download speed at all. Our download speed may be still at 1Gbps most of the time. Then you may wonder why Henry sign up 10Gbps Homehub+. I just like their 4k UHD TV contents like EPL and their bundled Apps and the ability to use wifi that is beyond 1Gbps which is not possible if i am on 1Gbps plan. Now i have the freedom to surf any where within my home with my laptop. I am happy.

 

[BradenHeat]

I think calling it smoke screen is a little disingenuous. A bit like statistics where you can beat the numbers until it show what you want people to see, so it's really up to the end user to be discerning of all the marketing speak.

All those theory are the backbone of networking, a lot of consumer products hide the low level implementations and present things to the end user using simple GUI, but behind the scenes, all those theory still apply.

That said, you don't have to learn all those theory just to setup your network at home.
See it as learning how to fish, vs buying the fish from the fisherman.

You can still eat fish if you buy it from the supermarket, no need go fish yourself innit?

not calling those numbers or hard to comprehend knowledge smokescreen, but rather how it was deemed [ thou shall need CCNA to get into holy networking basics ]

doesnt help when my lecturers at that time were focused more on whens the next school break rather than actual solid information, when asked or enquired.

its because i had to fish, as you put it, im here semi-half-or not so much knowledge on most wireless and network, but i can pull and put together pretty decent suggestions after failure after failure [ linksys cs 1.6 days, yea thats how i got into network ]