Insight from Google Fiber - WiFi Performance

[sfx2000]

[edit - I've removed the kitten reference - was cute, but not relevant to the content - I've also merged a post below with this one for clarity purposes, and edited for context -- sfx]

The author here was part of Google Fiber, and has some interesting stats across all things WiFi - Mesh, mixed mode, multiple AP's, beamforming, etc... Google Fiber, like many other ISP's, instrument their Customer Premises Equipment - mostly to gather performance statistics so that they can find opportunities to improve the level of service - they mostly don't care about your email or surfing habits - which is probably a topic for another day - but yes, most do collect data for performance reasons...

Anyways, excellent info and data here, very insightful - and it puts to rest many of the "urban myths" about WiFi, esp. in high density areas...

Below are the slide decks - pdf format, and very information dense -

• http://apenwarr.ca/diary/wifi-consumer-mesh-apenwarr-201706.pdf
• http://apenwarr.ca/diary/wifi-data-apenwarr-201602.pdf

Summarizing the interesting points - best to go thru both slide decks...

In General

• Most folks will see about 8 overlapping AP's in 2.4GHz - not a bad thing
• Performance impact with many nearby AP's - not really an issue until 50 or more neighboring AP's
• Single Stream clients are very popular, although 2-stream clients are getting more common
• Many clients are still single band 11n 2.4Ghz, but 5GHz is getting more common
• Many WLAN's in the home space - about 20 clients - number of clients associated doesn't really matter
• 2.4GHz is still relevant - physics has a play here...
• MDU's (apartments/condos) are generally better performing than single family homes, range is key
• Speed Testing across WiFI to WAN - generally unreliable
• More Spatial Streams - better - a 4-stream AP will outperform a 2 or 3 stream AP in every case
  • even with 1 or 2 stream clients - more radios on the AP add gain, and 4-stream AP's are always 4-stream, so precoding helps here - RF diversity gain for the extra radio, and coding gain from MIMO
• 2.4G performs well until RSSI is around -75
• 5G best place for RSSI is around -55
  • Between -75 to -45 dBm - performance is generally similar - there is a cliff that is hit, and perf drops rapidly after that point
• Close range to the AP is not necessarily a good thing - min range is actually a real thing
  • I'm thinking this is a saturation thing that impacts Eb/No - clipping in the analog domain
• Power - more power at the AP, generally not good - power is noise to adjacent networks, more power doesn't help clients - most WLAN's are going to be client driven, not AP driven
  • Note to @thiggins - should probably look at more clients in future reviews - you touched on this some time back with different client implementations with laptops (i'm missing the link, but it's there on the main site)
• Channel Selection - goes back to power control mentioned above - Google has done some interesting things with minstrel with QCA chipsets
  • Interesting note - going back to 2.4GHz and Wide Channels - the 1/6/11 plan might be not be the best - author is actually suggesting 1/5/11 due to channel minimums on 11g/11n - which might mean that 1/4/8/11 might be validated in a very dense environment

Getting to devices - AP's, Clients, Extenders

• Most generally work - but across different expectations - what works well for the 80 percent, well 20 percent might have problems - so change to a different device...
  • swap the device, solve the problem, at the risk of other problems, but generally most devices just work
• Extenders generally don't work, unless they do - depends on that 20 percent above...
• Bugs/Issues have gain - so if issues are with the root AP, they will be amplified with addtional AP's

Band Steering for Clients

• Split SSID's generally perform poorly compared to common SSID
• Common SSID's perform well - but does depend on client implementation
• Band Steering can help, but Common SSID is a good approach for clients that do not support BandSteering
• Apple generally does well in dual band common SSID, other vendors do as well
• Ping-Ponging can be a problem in Dual Band Common SSID

AirTime...

• AirTime Fairness is a good thing - a slow client can kill everyone without AirTime Fairness
• Wide Channels in 2.4GHz - not as evil as most folks think - the secondary channel in 11n must check for clear air before transmitting, and then if possible, it will - get the bits across and clear the air afterwords...
  • Secondary Channel is only used if it can be - this is in 802.11n Specs
• 11ac - 80MHz seems to be the sweet spot for range/performance (my data concurs) - 40Mhz is ok, 20Mhz not much benefit, superwide 160 or 80+80 - at the moment, no - but something to consider in 2020 - noise floor is the limit there, similar to extended modulation rates (Turbo/Nitro QAM - might work in the lab/conducted, but in the real world, noise rules)

Beamforming

• 11n - basically doesn't work due to standards ambiguity
  • might consider disabling this on the AP
• 11ac - much better - required in any event, and a must for MU
  • TxBF support required for AP, optional for client
• Can help, but it's complicated - regulatory issues restrict EIRP with beamforming
  • EIRP is limited to a max value, period - so might not be a benefit
  • Beamforming incurs overhead that might be better put to airtime instead
• More AP's at lower power are preferred to a single AP with big power

MU-MIMO

• telling that the author only sees the use case of TV set-top boxes
  • editorial - I agree - MU is good for static client locations and multicast/broadcast traffic as it stands with 11ac Wave 2
  • MU can suffer because of EIRP issues noted above with Beamforming
  • MU also incurs overhead, more than SU beamforming
  • Content is important - some traffic is better suited, some isn't because of overhead

MESH

• Generally good
• Better than AP's + Repeaters due to routing at the dot11 layer
• More nodes - there is a limit - more nodes can be a problem with instability in firmware - if a crashed node is in the mesh (or a node with bad tables), the mesh can have problems overall
• Location of nodes is critical - far enough to extend coverage, too close, clients might stick and not jump
  • Editorial Comment - this is my observation as well with 11n/11g efforts back in the 2010/2011 timeframe
  • Editorial Comment 2 - this aligns with experience on LTE with "smart antennas" and beam forming - better to have many small cells than one big one...

 

[80211WiGuy]

More Spatial Streams - better - a 4-stream AP will outperform a 2 or 3 stream AP in every case

• even with 1 or 2 stream clients - more radios on the AP add gain, and 4-stream AP's are always 4-stream, so precoding helps here - RF diversity gain for the extra radio, and coding gain from MIMO

I'm curious about this one, I wasn't aware that SOHO WiFi vendors were actually implementing MRC. I'm probably a little dated on this though.

 

[sfx2000]

I'm curious about this one, I wasn't aware that SOHO WiFi vendors were actually implementing MRC. I'm probably a little dated on this though.

MRC is more of a client thing - and pretty much anything with 11ac does it (it's transparent as it's handled at the baseband level, along with Rx diversity)

 

[sfx2000]

Sharing some thoughts and insight - and this goes back to a couple of articles on the main site...

https://www.smallnetbuilder.com/bas...751-snb-answer-guy-how-many-ssids-is-too-many

(it's a little known secret that I wrote the first article back in 2015)

and Tim's excellent follow up...

https://www.smallnetbuilder.com/basics/wireless-basics/33180-how-to-fix-wi-fi-roaming

Here's some basics of WiFi, and how devices pick and choose, and some realities of the WiFi channels in general...

The WiFi noise floor is around -90 dB - Good Wifi needs 20 dB, although with 802.11n, we can get by with good performance with 10dB.

Look at a crowded channel with a tool like Inssider or similar, you'll find a lot of WLAN SSID's around in the -70 to -80 dBm

And that's ok - even on the same channel as your current AP from the client perspective, as long as you have that 10 to 20 dB RSSI at the client.

It's a given that RSSI at the client is the trigger for Roaming, and that trigger for many is -65/-68 in traffic, to about 10 dB less in idle state... this is due to a couple of things, but basically, the beacon frame is transmitted at BPSK, even in 11n/11ac - BPSK has a huge amount of coding gain, and is used for channel estimation by the client.

One thing is very certain with WiFi Roaming, different SSID's will prevent roaming - as the candidate AP if a different SSID, means it's a different network, so most clients are not going to jump until the current SSID is lost.

This is where we have the "sticky client" problem - no reason to jump - RSSI is one thing, and the SSID/ESSID is the other.

So what if one is in a crowded WLAN neighborhood?

Good question - as apartments and the enterprise are rich with AP's, and this raises the noise floor - sometimes above the roaming triggers by many clients.

Yes, and it goes really towards the RSSI observed by the client that determines performance. If one can get that 10dB or better RSSI, you'll have good performance, if you can't, then it's just physics - as the noise floor starts to overwhelm the channel, and for engineers that do this stuff - we call it noise limited...

Which goes to one of the more cogent statements about WiFi... it's the client that is most important in relation to the AP's

The rule of thumb for decent WiFi coverage in most homes in the US - and that means wood-frame construction with sheetrock - 1500 sq ft for 2.4GHz, and 750 sq ft for 5GHz.

Punching thru walls - sheetrock is around 2.5dB hit on the link - plaster/lath (which is common in older homes) is around 5dB, and brick/cement is similar if not a bit more - depends on metal in the wall, but generally about a 5dB hit.

Most complaints about WiFi performance really is about distance to the AP itself... and performance there is is a 10LogR relationship - it's not linear, it's a log function.

Since WiFi is largely client driven, as they're the weaker hand - more power at the AP doesn't solve things, and higher gain antennas only flatten the pattern, which in multi-level homes makes things worse rather than better...

One of the better AP's I've found, and folks will give me a hard time about this, is the Airport Extreme AC - mostly because of the antenna pattern - it's a low gain antenna on that device, but it still pushes out a decent amount of coverage... it's "good" zone is actually pretty decent, esp in multiple level floor plans, and it works very well with mobiles and tablets, as they are also low antenna gain.

 

[dfarning]

Absolutely fascinating section on airtime fairness and latency control. I always wondered why even a moderately small upload would cause a significant reduction in network performance while the network could smoothly handle multiple gigabyte downloads gracefully.

Another bit that gets the mental juices flowing is the section thT Moore's applies to WiFi. Processing power on Wi-Fi devices is still increasing exponentially because digital signal processing can be highly parallelized.

 

[WiFiNemesis]

Useful info, thanks.

Interesting about Airport. I parted ways about three routers ago (and still questing for the "ultimate" wifi solution), but looks like that product truly has reached the end of the road.

 

[kab]

Great info, thank you. My big takeaway (immediate application) has been the multiple SSID discussions. I originally split the 2.4Ghz bands for "slow" (a/b) clients vs. faster (g/n) - way, way back it seemed to actually make a difference.

I'm a new member - Hi all! Great site.

Working my way through all the reference articles as I prep a home network overhaul. My DIR-655 w/ 8 yr old ROMs just ain't cutting it any more. Hey, I retired from high-tech and did a 4 yr purge - no cell phone, no new tech toys, lots and lots of biking and walking. Now reading this stuff is fun again!

 

[dfarning]

Welcome, I am also pretty new here. Was activate in the NSLU2 era.

I am not sure if I am an outlier, but the overall customer (ie family) satisfaction of our home/office network has been significantly improved by dealing with the slow clients rather than focusing on the highest speed.

 

[thiggins]

Iam not sure if I am an outlier, but the overall customer (ie family) satisfaction of our home/office network has been significantly improved by dealing with the slow clients rather than focusing on the highest speed.

Bingo!

 

[coxhaus]

Before I switched to totally 5 GHz and turned off 2.4GHz. I had 23 APs around me using inssiders and Cisco's neighborhood software which caused lag in my 2.4 GHz wireless network. So I am thinking 50 is bit high. I was using 2 or 3 Cisco WAP321 access points backhauled on gig Ethernet. I am picky about my wireless access.

 

[sfx2000]

Before I switched to totally 5 GHz and turned off 2.4GHz. I had 23 APs around me using inssiders and Cisco's neighborhood software which caused lag in my 2.4 GHz wireless network. So I am thinking 50 is bit high. I was using 2 or 3 Cisco WAP321 access points backhauled on gig Ethernet. I am picky about my wireless access.

Lot of it depends on the proximity of those neighbors relative to one's own AP and Clients....

One can have 50 neighbors, but if one has that 20 dB of freespace between one's own beacons and the neighbors, one can maintain a good 802.11n traffic session, as the neighbors in that case are essentially just noise - the noise floor does impact range, and a given throughput over that range once hits noise limits. In any event, the rate selection in a WLAN is generally client driven, not AP driven, and even there, many clients are 3 to 6 dB lower in Tx than the AP is. Even in 5GHz, an 802.11ac client might be 16 dBm max Tx compared to an AP that might have 30 dBm in some cases.

WiFi is essentially noise limited, but even co-channel interference rejection for 802.11n is actually -3dB as per spec - which was a significant change from earlier 802.11b/g rejection specs.

The data in the google fiber report tends to confirm this, as does the AP testing done on SmallNetBuilder.

In my own environment, I do keep 2.4GHz active, but plan the AP's and placement of those AP's where 5GHz is the preferred connection for dual-band clients, and use the 2.4 channels as a backstop and for devices that don't support 2.4 (which I have more than a few)

 

[coxhaus]

I was not able to run wide channels on 2.4GHz.

 

[Easy Rhino]

I thought these slides were great thanks for putting them up.

 

[HTBruceM]

Thank you for taking the time to post a condensed summary!

 

[sfx2000]

I was not able to run wide channels on 2.4GHz.

Depends on the client - that being said, 20MHz channels are probably best for most use cases...

Wide Channels can offer more bandwidth - the opportunity for them is limited - and there is a range impact with wide channels... even if a wide channel is deployed, many clients won't use it.

 

[liamr]

MRC is more of a client thing

 

[Deleted member 51572]

[edit - I've removed the kitten reference - was cute, but not relevant to the content - I've also merged a post below with this one for clarity purposes, and edited for context -- sfx]

The author here was part of Google Fiber, and has some interesting stats across all things WiFi - Mesh, mixed mode, multiple AP's, beamforming, etc... Google Fiber, like many other ISP's, instrument their Customer Premises Equipment - mostly to gather performance statistics so that they can find opportunities to improve the level of service - they mostly don't care about your email or surfing habits - which is probably a topic for another day - but yes, most do collect data for performance reasons...

Anyways, excellent info and data here, very insightful - and it puts to rest many of the "urban myths" about WiFi, esp. in high density areas...

Below are the slide decks - pdf format, and very information dense -

• http://apenwarr.ca/diary/wifi-consumer-mesh-apenwarr-201706.pdf
• http://apenwarr.ca/diary/wifi-data-apenwarr-201602.pdf

Summarizing the interesting points - best to go thru both slide decks...

In General

• Most folks will see about 8 overlapping AP's in 2.4GHz - not a bad thing
• Performance impact with many nearby AP's - not really an issue until 50 or more neighboring AP's
• Single Stream clients are very popular, although 2-stream clients are getting more common
• Many clients are still single band 11n 2.4Ghz, but 5GHz is getting more common
• Many WLAN's in the home space - about 20 clients - number of clients associated doesn't really matter
• 2.4GHz is still relevant - physics has a play here...
• MDU's (apartments/condos) are generally better performing than single family homes, range is key
• Speed Testing across WiFI to WAN - generally unreliable
• More Spatial Streams - better - a 4-stream AP will outperform a 2 or 3 stream AP in every case
  • even with 1 or 2 stream clients - more radios on the AP add gain, and 4-stream AP's are always 4-stream, so precoding helps here - RF diversity gain for the extra radio, and coding gain from MIMO
• 2.4G performs well until RSSI is around -75
• 5G best place for RSSI is around -55
  • Between -75 to -45 dBm - performance is generally similar - there is a cliff that is hit, and perf drops rapidly after that point
• Close range to the AP is not necessarily a good thing - min range is actually a real thing
  • I'm thinking this is a saturation thing that impacts Eb/No - clipping in the analog domain
• Power - more power at the AP, generally not good - power is noise to adjacent networks, more power doesn't help clients - most WLAN's are going to be client driven, not AP driven
  • Note to @thiggins - should probably look at more clients in future reviews - you touched on this some time back with different client implementations with laptops (i'm missing the link, but it's there on the main site)
• Channel Selection - goes back to power control mentioned above - Google has done some interesting things with minstrel with QCA chipsets
  • Interesting note - going back to 2.4GHz and Wide Channels - the 1/6/11 plan might be not be the best - author is actually suggesting 1/5/11 due to channel minimums on 11g/11n - which might mean that 1/4/8/11 might be validated in a very dense environment

Getting to devices - AP's, Clients, Extenders

• Most generally work - but across different expectations - what works well for the 80 percent, well 20 percent might have problems - so change to a different device...
  • swap the device, solve the problem, at the risk of other problems, but generally most devices just work
• Extenders generally don't work, unless they do - depends on that 20 percent above...
• Bugs/Issues have gain - so if issues are with the root AP, they will be amplified with addtional AP's

Band Steering for Clients

• Split SSID's generally perform poorly compared to common SSID
• Common SSID's perform well - but does depend on client implementation
• Band Steering can help, but Common SSID is a good approach for clients that do not support BandSteering
• Apple generally does well in dual band common SSID, other vendors do as well
• Ping-Ponging can be a problem in Dual Band Common SSID

AirTime...

• AirTime Fairness is a good thing - a slow client can kill everyone without AirTime Fairness
• Wide Channels in 2.4GHz - not as evil as most folks think - the secondary channel in 11n must check for clear air before transmitting, and then if possible, it will - get the bits across and clear the air afterwords...
  • Secondary Channel is only used if it can be - this is in 802.11n Specs
• 11ac - 80MHz seems to be the sweet spot for range/performance (my data concurs) - 40Mhz is ok, 20Mhz not much benefit, superwide 160 or 80+80 - at the moment, no - but something to consider in 2020 - noise floor is the limit there, similar to extended modulation rates (Turbo/Nitro QAM - might work in the lab/conducted, but in the real world, noise rules)

Beamforming

• 11n - basically doesn't work due to standards ambiguity
  • might consider disabling this on the AP
• 11ac - much better - required in any event, and a must for MU
  • TxBF support required for AP, optional for client
• Can help, but it's complicated - regulatory issues restrict EIRP with beamforming
  • EIRP is limited to a max value, period - so might not be a benefit
  • Beamforming incurs overhead that might be better put to airtime instead
• More AP's at lower power are preferred to a single AP with big power

MU-MIMO

• telling that the author only sees the use case of TV set-top boxes
  • editorial - I agree - MU is good for static client locations and multicast/broadcast traffic as it stands with 11ac Wave 2
  • MU can suffer because of EIRP issues noted above with Beamforming
  • MU also incurs overhead, more than SU beamforming
  • Content is important - some traffic is better suited, some isn't because of overhead

MESH

• Generally good
• Better than AP's + Repeaters due to routing at the dot11 layer
• More nodes - there is a limit - more nodes can be a problem with instability in firmware - if a crashed node is in the mesh (or a node with bad tables), the mesh can have problems overall
• Location of nodes is critical - far enough to extend coverage, too close, clients might stick and not jump
  • Editorial Comment - this is my observation as well with 11n/11g efforts back in the 2010/2011 timeframe
  • Editorial Comment 2 - this aligns with experience on LTE with "smart antennas" and beam forming - better to have many small cells than one big one...

interesting reading, thanks.

regarding channel selection, adjacent vs co-channels on the 2.4GHz band, it is perfectly acceptable to use overlapping (adjacent) channels. 802.11 was explicitly designed for this. the widespread myth about only using channels 1, 6 or 11 is misplaced, and simply not good advice anymore, especially in the crowded 2.4GHz spectrum.

 

[sfx2000]

especially in the crowded 2.4GHz spectrum

And 2.4 (along with 5Ghz) is getting more and more crowded...

quick scan in my local suburban neighborhood...

(attached is a CSV file, change the extension and import into Excel or similar)

 

[thiggins]

@techcafe As your linked sources state, operating off-channel from 1/6/11 lowers signal-to-noise, which potentially reduces bandwidth in all the surrounding channels. This is not something I could recommend doing.

If you're seeing better throughput by using an off-channel, it's likely because the adjacent 1/6/11 aren't that busy.

On the other hand, equipment manufacturers apparently don't adhere to the 1/6/11 rule. I've seen routers come up on channel 4 (or others) when left to auto select a channel.

 

[L&LD]

@thiggins that is my experience too. Much higher peaks when other networks are not busy, more complaints when they are. When the router was set outside the 1/6/11 recommendations.

After setting them to the best for their environment, the customers initially lament the lower peak speeds, but they can't complain about the consistency of their Wi-Fi and network anymore.

Being throttled by one band of 'other' routers in the neighborhood is always better (for everyone) than being throttled by two bands full of them.