I did some experimentation with UniFi hubs and came to the conclusion that if you can give each device its own WiFi channel that would be ideal -- contention is that bad and often an uncontended channel with otherwise poor characteristics will beat a contended channel that otherwise looks good.
The other bit of advice that is buried in there that no-one wants to hear for residences is the best way to speed up your Wi-Fi is to not use it. You might think it's convenient to have your TV connect to Netflix via WiFi and it is, but it is going to make everything else that really needs the Wi-Fi slower. It's a much better answer to hook up everything on Ethernet that you possibly can than it is to follow the more traveled route of more channels and more congestion with mesh Wi-Fi.
As far as I'm concerned Wifi is a "solution" for lazy or incompetent people outside of mobile devices (who do really need it to be useful).
Network cable is cheap, crimper is cheap, creating a simple network in your house is mostly a trivial matter.
I have helped some people who had many troubles with wifi devices (particularly printers) and when then didn't want to run a cable to solve the problem forever I told them to fuck off. If there is one thing that is certain with wifi, it's that it will break at some point and randomly show poor performance/issues. Anything that doesn't have to be absolutely wireless has to be connected that way, problem solved, forever.
The thing about speed tests causing a bad experience because they hog airtime felt like a non sequitur (since performing them is rare and manual) until I saw this:
> Many ISPs, device manufacturers, and consumers automate periodic, high-intensity speed tests that negatively impact the consumer internet experience as demonstrated
But there’s no support for this claim presented frankly I am skeptical. What WiFi devices are regularly conducting speed tests without being asked?
I don't get what the point of the article is. Is the takeaway that I should lower the channel width in my home? How many WAPs would I need to be running for that to matter? I'd argue it's more important to get everyone to turn down TX power in cases where your neighbors in an apartment building are conflicting. And that's never going to happen, so just conform to the legal limit and your SNR should be fine. Anything that needs to be high performance shouldn't be on wifi anyway.
If you want to spend a really long time optimizing your wifi, this is the resource: https://www.wiisfi.com/
I don't know why I was downvoted.
I was just expressing awe at how many wi-fi devices an average US household has.
I was not implying the information was wrong.
Honestly what's unsaid in a lot of this is that it would be really nice if there were more and wider ISM bands. So much makes use of 900Mhz, 2.4GHz and 5GHz in novel and innovative ways, that if the government and FCC really actually wanted to spark innovation including augmenting wifi performance, they'd stop letting telcos and other questionable interests hoard spectrum and release it as ISM (and no, they shouldn't steal from ham bands to make ISM bands either).
WiFi 6E added 6 GHz spectrum bigger than 5 GHz. The problem is that not that many devices support it. I think it is required in WiFi 7. I think it is shared with other users so devices have to find free space. I think most developed countries have allocated it.
> Many ISPs, device manufacturers, and consumers automate periodic, high-intensity speed tests that negatively impact the consumer internet experience as demonstrated.
Is that actually a thing? Why would any ISP intentionally add unnecessary load to their network?
In the IoT space I really wish an "ESP for power line Ethernet" existed these days.
I have 50+ ESP based devices on WiFi and while low bandwidth (and their own SSID) I really wish there were affordable options that they could be "wired" for comms (since they mostly control mains appliances, but the rules and considerations for mixing data and mains in one package are prohibitively expensive).
For me the only thing that really matters, and globally sucks with WiFi is roaming.
My house is old and has stones walls up to 120cm, including the inner walls, so I have to have access points is nearly all rooms.
I never had a true seamless roaming experience. Today, I have TP-Link Omada and it works better than previous solutions, but it is still not as good as DECT phones for examples.
For example if I watch a twitch stream in my room and go to the kitchen grab something with my tablet or my phone, I have a freeze about 30% of the times, but not very long. Before I sometime had to turn the wifi off and on on my device for it to roam.
I followed all Omada and general WiFi best practice I could find about frequency, overlap... But it is still not fully seamless yes.
I wish the Wi-Fi developers would put some serious effort into improving range and contention. Forgot 40 MHz vs 80 MHz — how about some 5 MHz channels? How about some modulations designed to work at low received power and/or low SNR? How about improving the stack to get better performance when a device has mediocre signal quality to multiple APs at the same time?
There are are these cool new features like MLO, but maybe devices could mostly use narrow channels and only use more RF bandwidth when they actually need it.
A households bandwidth use is quite a bit different to a business. While a household may have a lot of devices most of them are doing very little at any given time, but the primary device in use requires the best speed possible. In a business however there are a lot of primary devices and not a lot of idle little devices and as such fairness and reliability dominate the needs as does getting the frequencies maxed out for coverage and total bandwidth available.
Wifi 8 will probably be another standard homes can skip. Like wifi 6 it is going to bring little that they need to utilise their fibre home connnections well across their home.
> Because consumers have been conditioned to understand only raw speed as a metric of Wi-Fi quality and not more important indicators of internet experience such as responsiveness and reliability.
Whie the two are not the same, they are not exactly separable.
You will not get good Internet speed out of a flaky network, because the interrupted flow of acknowledgements, and the need to retransmit lost segments, will not only itself impact the performance directly, but also trigger congestion-reducing algorithms.
Most users are not aware whether they are getting good speed most of the time, if they are only browsing the web, because of the latencies of the load times of complex pages. Individual video streams are not enough to stress the system either. You have to be running downloads (e.g. torrents) to have a better sense of that.
The flakiness of web page loads and insufficient load caused by streams can conceal both: some good amount of unreliability and poor throughput.
I'm surprised, at least for businesses, small cell wifi is not a thing. For example, if you walk into an office building everyone seems to have a physical phone on their desk that is hard wired. What if that is also a small cell AP. Like a personal AP. Using automation and central provisioning and analytics can make this doable. Yeah handoff and roaming has to be seamless and quick but it doesn't feel that hard, no? If so this would be pretty neat and would solve the contention issue in the air.
Their `networkQuality` implementation is on the CLI for any Mac recently updated. It's pretty interesting and I've found it to be very good at predicting which networks will be theoretically fast, but feel unreliable and laggy, and which ones will feel snappy and fast. It measures Round-trips Per Minute under idle and load condition. It's a much better predictor of how fast casual browsing will be than a speed test.
Moving into a house for the first time since before college this year, I only just learned about Wi-Fi channel width this week. Apparently the mesh routers I ended up picking several months ago had a default width of 160 MHz, but only go as low as 80 MHz, so that's what I ended up switching to. Anecdotally it has seemed to be somewhat more reliable, but maybe in the long run finding something that can go even lower might be worth it because we do still notice some stutter occasionally that would be nice to reduce even if the theoretical max throughout was a bit lower.
This is a clear case of "you get what you measure". Measuring speed is so easy, everybody can do it, and do it all the time. No wonder that providers optimize for speed. But it also works the other way around. We have developed a focus on speed as it was the only thing that mattered.
I have worked with networks for many years, and users blaming all sorts of issues on the network is a classic, so of course in their minds they need more speed and more bandwidth. But improvements only makes sense up to some point. After that it is just psychological.
This is such a great write-up it highlights a truth that’s been hiding in plain sight for years: we’ve optimized Wi-Fi for headline speeds, not human experience.
Emphasis on throughput reminds one of the "megapixel wars" of early digital photography a simple, clear-cut figure that completely misrepresents actual quality. Responsiveness and reliability are the actual measures that control day to day satisfaction, but they are harder to define and won't neatly go on a store shelf.
What is fascinating here is that speed tests themselves actively degrade the network performance. It's like taking your pulse by first dashing a lap in sprinting mode. Whether we'll see more router makers or ISPs start offering "responsiveness scores" instead of speed numbers once the consumer pays attention to latency and airtime contention remains to be seen.
At any rate, this post nails the broader cultural problem in networking: the industry still chases awe inspiring numbers instead of better experiences.
Is this still true with OFDM and subchannels or whatever it's called?
Also MIMO.
And don't think it's relevant to compare what to do in a large space with what one should do at home. The requirements are entirely different.
In a large space with many users I'd use small channels and many access points. I want it to work good enough for everyone to have calls, and have good aggregate throughput.
In a two bed home I'd use large channels and probably only one AP. Peak single device speed is MUCH more important than aggregate speed.
And in a home it matters much more what channels are being busyed by neighbors.
For latency, of course, there is only wired. Even with few devices.
35 comments
[ 4.5 ms ] story [ 58.9 ms ] threadThe other bit of advice that is buried in there that no-one wants to hear for residences is the best way to speed up your Wi-Fi is to not use it. You might think it's convenient to have your TV connect to Netflix via WiFi and it is, but it is going to make everything else that really needs the Wi-Fi slower. It's a much better answer to hook up everything on Ethernet that you possibly can than it is to follow the more traveled route of more channels and more congestion with mesh Wi-Fi.
I have helped some people who had many troubles with wifi devices (particularly printers) and when then didn't want to run a cable to solve the problem forever I told them to fuck off. If there is one thing that is certain with wifi, it's that it will break at some point and randomly show poor performance/issues. Anything that doesn't have to be absolutely wireless has to be connected that way, problem solved, forever.
If you want to spend a really long time optimizing your wifi, this is the resource: https://www.wiisfi.com/
Is that actually a thing? Why would any ISP intentionally add unnecessary load to their network?
I have 50+ ESP based devices on WiFi and while low bandwidth (and their own SSID) I really wish there were affordable options that they could be "wired" for comms (since they mostly control mains appliances, but the rules and considerations for mixing data and mains in one package are prohibitively expensive).
My house is old and has stones walls up to 120cm, including the inner walls, so I have to have access points is nearly all rooms.
I never had a true seamless roaming experience. Today, I have TP-Link Omada and it works better than previous solutions, but it is still not as good as DECT phones for examples.
For example if I watch a twitch stream in my room and go to the kitchen grab something with my tablet or my phone, I have a freeze about 30% of the times, but not very long. Before I sometime had to turn the wifi off and on on my device for it to roam.
I followed all Omada and general WiFi best practice I could find about frequency, overlap... But it is still not fully seamless yes.
I wonder how many of those could be wired.
So yeah, I do think speed is more important.
Responsiveness doesn’t matter that often and when it does, plugging in Ethernet takes it out of the equation.
There are are these cool new features like MLO, but maybe devices could mostly use narrow channels and only use more RF bandwidth when they actually need it.
* https://en.wikipedia.org/wiki/IEEE_802.11bn
So other considerations are being considered.
Wifi 8 will probably be another standard homes can skip. Like wifi 6 it is going to bring little that they need to utilise their fibre home connnections well across their home.
Whie the two are not the same, they are not exactly separable.
You will not get good Internet speed out of a flaky network, because the interrupted flow of acknowledgements, and the need to retransmit lost segments, will not only itself impact the performance directly, but also trigger congestion-reducing algorithms.
Most users are not aware whether they are getting good speed most of the time, if they are only browsing the web, because of the latencies of the load times of complex pages. Individual video streams are not enough to stress the system either. You have to be running downloads (e.g. torrents) to have a better sense of that.
The flakiness of web page loads and insufficient load caused by streams can conceal both: some good amount of unreliability and poor throughput.
Their `networkQuality` implementation is on the CLI for any Mac recently updated. It's pretty interesting and I've found it to be very good at predicting which networks will be theoretically fast, but feel unreliable and laggy, and which ones will feel snappy and fast. It measures Round-trips Per Minute under idle and load condition. It's a much better predictor of how fast casual browsing will be than a speed test.
I have worked with networks for many years, and users blaming all sorts of issues on the network is a classic, so of course in their minds they need more speed and more bandwidth. But improvements only makes sense up to some point. After that it is just psychological.
Also MIMO.
And don't think it's relevant to compare what to do in a large space with what one should do at home. The requirements are entirely different.
In a large space with many users I'd use small channels and many access points. I want it to work good enough for everyone to have calls, and have good aggregate throughput.
In a two bed home I'd use large channels and probably only one AP. Peak single device speed is MUCH more important than aggregate speed.
And in a home it matters much more what channels are being busyed by neighbors.
For latency, of course, there is only wired. Even with few devices.