FYI/A: Bufferbloat 101

[xiaofan]

The point I was trying to make is that SQM/QoS measures are not so useful for wireless clients when the signal is weak. The more important thing when the signal is weak is to add extra AP (if there is LAN port near by) or repeater (not as effective as AP).

SQM/QoS will still be useful if the CPU can handle the task without sacrificing much on line rate. Typical consumer grade wireless router can not handle 1Gbps SQM/QoS effectively. That is why when people only gets 300Mbps or 600Mbps from wired connection from their Asus routers, the first thing is to check whether they have turn QoS Feature ON or not. But if use a router with powerful CPU, like using OpenWRT on an Intel N100 CPU based mini PC, then there should be no issues using SQM/QoS.

There could also be effective SQM/QoS implementations which does not tax the consumer router CPU so much but I do not know which are the better ones for consumer routers.

For those who want to ask about mesh with Ethernet Backhaul or Wireless Backhaul.
1) Mesh with Ethernet Backhaul --> glorified AP with some roaming helping technologies, like 802.11 k/v/r or EasyMesh or proprietory technology like Asus AImesh, Netgear Orbi, TP-Link Deco and Linksys Velop. Small Enterprise and Enterprise Grade APs may have even better technology to help roaming.

2) Mesh with Wireless Backhaul --> glorified repeater with some roaming helping technologies, like 802.11 k/v/r or EasyMesh or proprietory technology like Asus AImesh, Netgear Orbi, TP-Link Deco and Linksys Velop. Small Enterprise and Enterprise Grade APs may have even better technology to help roaming.

 

[hwzlite]

Still in your posted results, both download speed and upload speed are low, this shows a weak router as well. And I know that is because you are using an old Linksys EA8100 v1 router.

This is expected, can't compare apples (802.11ac) vs oranges (802.11ax/be) class due to QAM modulation right?


Btw here are the Kitchen's toilet results to further demo without the benefit of fq_codel/ATF/AQL optimized MediaTek mt76 mac80211 driver, unfortunately only ATF & AQL can be disabled :

https://www.waveform.com/tools/bufferbloat?test-id=3411464a-f6a8-4eb3-a328-1f21b22e6a1f
https://www.waveform.com/tools/bufferbloat?test-id=61e53750-6c07-4a36-af47-708f2b1484c3
https://www.waveform.com/tools/bufferbloat?test-id=f060132d-28d9-4e19-bd3d-d322eda6c9be

Can see the latency difference with an increase of ~40ms download and ~180ms upload .

 

[xiaofan]

Btw here are the Kitchen's toilet results to further demo without the benefit of fq_codel/ATF/AQL optimized MediaTek mt76 mac80211 driver, unfortunately only ATF & AQL can be disabled :

https://www.waveform.com/tools/bufferbloat?test-id=3411464a-f6a8-4eb3-a328-1f21b22e6a1f
https://www.waveform.com/tools/bufferbloat?test-id=61e53750-6c07-4a36-af47-708f2b1484c3
https://www.waveform.com/tools/bufferbloat?test-id=f060132d-28d9-4e19-bd3d-d322eda6c9be

Can see the latency difference with an increase of ~40ms download and ~180ms upload .

Yes the difference is quite significant, even thought the results are still "F".

That being said, I am still not convinced about the benefits of the SQM stuff when you have a faster and more stable wireless connection. Basically my view is to fix the wireless connection first (eg: to get a better wireless router, and/or deploy APs or repeaters or mesh nodes) and not trying to use SQM to try to conpensate for the poor wireless signal.

 

[hwzlite]

Ofcos all these tests are not about compensating for the poor wireless signal or other physical limitations in the first place.

The fundamental of Smart Queue Management (SQM) is all about minimizing latency, even when Internet connection is fully saturated.

 

[xiaofan]

Ofcos all these tests are not about compensating for the poor wireless signal or other physical limitations in the first place.

The fundamental of Smart Queue Management (SQM) is all about minimizing latency, even when Internet connection is fully saturated.

The keyword is "fully saturated" which is never really a real use case for most users other than carrying out SpeedTest or waveform.com bufferbloat test.

So I will still maintain that most of the users in Singapore (500Mbps, 1Gbps and above) do not need to use SQM. And it does more harm than good to use the poor QoS settings in the consumer routers anyway.

 

[xiaofan]

Basically I do not believe the following sentence from waveform.com is true for typical users in Singapore.

I believe for most of the users here in Singapore, "If so, poor wireless signal may be to blame".

https://www.waveform.com/tools/bufferbloat
Do your video or audio calls sometimes stutter? Does your web browsing slow down? Do video games lag?
If so, bufferbloat may be to blame.

 

[xiaofan]

The test itself may still be useful though to see how one's network performs. But I believe it is more meaningful to carry out wired tests and less on wireles tests.

 

[hwzlite]

The keyword is "fully saturated" which is never really a real use case for most users other than carrying out SpeedTest or waveform.com bufferbloat test.

Use case: Tell those who BitTorrent alot of swamping bandwidth.
10Gbps is fantastic as of today, but will back to square one once all apps or network services bloated to new levels

 

[xiaofan]

Use case: Tell those who BitTorrent alot of swamping bandwidth.
10Gbps is fantastic as of today, but will back to square one once all apps or network services bloated to new levels

That is a minority.

BTW, I support ISP's policy to throttle BT traffic. For example, Singtel roughly limits the BT traffic speed to 400Mbps and to me that should be fine.

 

[hwzlite]

That is a minority.

BTW, I support ISP's policy to throttle BT traffic. For example, Singtel roughly limits the BT traffic speed to 400Mbps and to me that should be fine.

There is even another minority that go for those unlimited VPN to bypass it.
For me I just go slow-but-faster-then-throttled in TOR network

 

[hwzlite]

Update on my WIRELESS adventure into using fq_codel/ATF/AQL optimized MediaTek mt76 mac80211 driver on Linksys EA8100(OpenWRT) , configured as Dumb AP, and tweaked with "aql_txq_limit = 1500" setting :

Here are the wireless benchmarks (M1 500Mbps, DFS channel 64 @80Mhz, Wall = old HDB type):
Line-Of-Sight, ~6m:
https://www.waveform.com/tools/bufferbloat?test-id=ef0c1814-5699-463b-957c-0b057da1e06b (A+)
https://www.waveform.com/tools/bufferbloat?test-id=9669484d-82ab-4561-acf3-e040a15ea052 (A+)
https://www.waveform.com/tools/bufferbloat?test-id=9e60560f-cc01-4ede-b17d-63d7b3fc6065 (A+)

Block by 1x Wall , ~8m:
https://www.waveform.com/tools/bufferbloat?test-id=653bcd05-58c3-4f4d-9d15-0c97a004f7fd (A)
https://www.waveform.com/tools/bufferbloat?test-id=9b69cf66-f6d0-4305-9b38-f5fbf7b98dc7 (A)
https://www.waveform.com/tools/bufferbloat?test-id=336b7e85-4913-4367-b8a3-4c3910fee3d5 (A)

Block by 2x Walls, ~9m:
https://www.waveform.com/tools/bufferbloat?test-id=6ea67cfd-1681-4f21-974a-246094a7181a (B)
https://www.waveform.com/tools/bufferbloat?test-id=54d11a51-bd0c-4a49-9016-2a3341b0f569 (B)
https://www.waveform.com/tools/bufferbloat?test-id=5a0d52fb-0192-41cb-8931-fbe87f422938 (C)


Ref: How OpenWrt Vanquishes Bufferbloat

Getting a little scientific with visualization from Heat Map generated by WiFi Analyzer and Surveyor :

(-25 dBm = DumbAP location)

By Agaration, can derive that anything <55dBM signal strength get "A/A+" rating.

Worse "F" hits are from the Kitchen's toilet, largely bogged down by upload latencies:
https://www.waveform.com/tools/bufferbloat?test-id=4eea6f46-8bed-4717-9e1a-d4ad89e54973
https://www.waveform.com/tools/bufferbloat?test-id=019de526-26bb-4f53-a236-ea5cfc1ca869
https://www.waveform.com/tools/bufferbloat?test-id=872038ef-9cce-4884-80aa-586e80ffa56a

Weekend-MidnightRun-Update:

Ideally for best results, both AP and client should have TXQS/ATF/AQL supported mac80211 drivers working in both ways.

Here are the Kitchen's toilet results on FENVI AX1800 External USB client (Mediatek MT7921au) connected to ThinkPad E485 booted from Kali Linux Live Boot USB :

https://www.waveform.com/tools/bufferbloat?test-id=f2918722-7a8b-427a-8689-53bacf0c04f4
https://www.waveform.com/tools/bufferbloat?test-id=aaf4055d-04d2-41c5-abfb-d76a5dd9beff
https://www.waveform.com/tools/bufferbloat?test-id=7fb5a814-9305-4e09-b984-159feecf23ae

All scored "B" with ~36ms upload latency, indeed a significant improvement from "F" with ~530ms latency.


Ref: ""At higher attenuation, the improvement is even more significant; there is more than 10x latency reduction at the 40 dB attenuation point.""

 

[xiaofan]

Interestingly I got the following wireless result this morning using OpenWRT + ZTE BE7200 Pro+ AP. Unfortunately it is still A and not A+.
https://www.waveform.com/tools/bufferbloat?test-id=4ef4aba4-19c0-4ecd-9947-e02270568837

Normal result is also A but not as good as the above.
https://www.waveform.com/tools/bufferbloat?test-id=a590bf16-0d04-4572-9c25-9f472f194e34

BTW, based on previous understanding, the ZTE GPON ONT has some issues with upload latency when under loading. Probably that is the main reason that I can not get A+ (even using wired). Last time @miloaisdino managed to limit the upload speed to 800Mbps to have better result.

 

[lobukong]

The keyword is "fully saturated" which is never really a real use case for most users other than carrying out SpeedTest or waveform.com bufferbloat test.

So I will still maintain that most of the users in Singapore (500Mbps, 1Gbps and above) do not need to use SQM. And it does more harm than good to use the poor QoS settings in the consumer routers anyway.

May I know what harm is caused? only thing I can see is not able to fully saturate the bandwidth. But if that's not a real use case then QoS doesn't really have any harm?
I chose to enable QoS and set max bandwidth to 650M, so that I can always have a A+ results and be sure that I don't have any risk of lag spike for online games.

I'm interested if ASUS routers can also get A+ rating when enable QoS? If the user care about lag more than bandwidth, why not just enable QoS?

 

[xiaofan]

May I know what harm is caused? only thing I can see is not able to fully saturate the bandwidth. But if that's not a real use case then QoS doesn't really have any harm?
I chose to enable QoS and set max bandwidth to 650M, so that I can always have a A+ results and be sure that I don't have any risk of lag spike for online games.

I'm interested if ASUS routers can also get A+ rating when enable QoS? If the user care about lag more than bandwidth, why not just enable QoS?

You can always go with your choice. 650M is not bad anyway.

Many Asus router will max out at about 300+ Mbps once QoS is enabled.

The thing is that the user will not feel the lag even if Waveform.com result is "F" as the test is not valid anyway for most users.

Anyway, this is just my personal opionion. You can always have different opinions.

 

[lobukong]

You can always go with your choice. 650M is not bad anyway.

Many Asus router will max out at about 300+ Mbps once QoS is enabled.

The thing is that the user will not feel the lag even if Waveform.com result is "F" as the test is not valid anyway for most users.

Anyway, this is just my personal opionion.

True, most user may also not able to feel difference between 300M vs 1G without speedtest.
So I think it's a trade off that's for user to decide. For gamers I believe it's worth enabling.

 

[xiaofan]

True, most user may also not able to feel difference between 300M vs 1G without speedtest.
So I think it's a trade off that's for user to decide. For gamers I believe it's worth enabling.

If you look at the discussions, quite some are related to wireless test. I feel that is even less useful than wired bufferbloat test. Gamers should probably use wired anyway.

All in all, everyone agrees latency is important. The argument is whether waveform.com bufferbloat test is really valid or not for typical use case, including gaming where you do not really download lot of things when gaming.

If the result is not valid, then the remedy to use some SQM/QoS measures is doubtful.

Again, the above is just my personal opinion. I may be totally wrong here. But so far none of the explanations have been able to convince me that waveform.com bufferbloat test is really valid for typical use case.

Still it is good to quote the other forum user.

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.

 

[lobukong]

I'm using wired and not referring to wireless, but I think for Wireless to have SQM is still better than not to have.
I asked chatGPT to give an analogy.

Without SQM or QoS: It's like a coffee shop where everyone waits in the same line, no matter how quick their order is. If someone orders a hundred coffees, you'll have to wait, even if you just want a single espresso.

With SQM or QoS: It's like having a fast track for espresso orders, so you don't wait behind the big orders, ensuring you get your coffee quickly.

For a 1G connection: Imagine the coffee shop can serve coffee extremely fast, almost instantly. Even if someone orders a hundred coffees, your espresso will be ready quickly. However, the fast track still helps during the busiest times, ensuring your order is always prioritized.

 

[xiaofan]

I'm using wired and not referring to wireless, but I think for Wireless to have SQM is still better than not to have.
I asked chatGPT to give an analogy.

Without SQM or QoS: It's like a coffee shop where everyone waits in the same line, no matter how quick their order is. If someone orders a hundred coffees, you'll have to wait, even if you just want a single espresso.

With SQM or QoS: It's like having a fast track for espresso orders, so you don't wait behind the big orders, ensuring you get your coffee quickly.

For a 1G connection: Imagine the coffee shop can serve coffee extremely fast, almost instantly. Even if someone orders a hundred coffees, your espresso will be ready quickly. However, the fast track still helps during the busiest times, ensuring your order is always prioritized.

Your analogy does not apply here, no matter it is wired or wireless (good wireless), unless you are gaming or carry out video conferencing while downloading a lot of stuff at the same time (eg: those who keep BT running to the max all the time).

You own the coffee shop (or coffee machine) in this case, you do not do big orders at home, because you can not drink so fast. You never saturate the capability of the coffee machine or the coffee shop. So you do not need to queue.

Anyway, it is good that you find SQM/QoS useful in your cases. I find it useless in my use cases.

Edit to add: when the wireless is bad, then yes there will be issues. But the fix is not the SQM/QoS stuff, rather you have to fix the wireless issues, by using a better wireless router, deploying extra mesh node, or AP, etc.

 

[lobukong]

Your analogy does not apply here, no matter it is wired or wireless (good wireless), unless you are gaming or carry out video conferencing while downloading a lot of stuff at the same time (eg: those who keep BT running to the max all the time).

You own the coffee shop (or coffee machine) in this case, you do not do big orders at home, because you can not drink so fast. You never saturate the capability of the coffee machine or the coffee shop. So you do not need to queue.

Anyway, it is good that you find SQM/QoS useful in your cases. I find it useless in my use cases.

Edit to add: when the wireless is bad, then yes there will be issues. But the fix is not the SQM/QoS stuff, rather you have to fix the wireless issues, by using a better wireless router, deploying extra mesh node, or AP, etc.

Yes, of course you may not need it. But wouldn't it be nice if you could BT, download, gaming at the same time and be sure that gaming is not affected? (After reading the comments I remembered indeed I used to stop BT tasks before gaming. Then probably forgot to start them after gaming.)

It's not too uncommon to do all three at once, for a gamer that has a second screen and a NAS. If there are more than one person at home, it's even more common.

Also the SQM algorithms have some recent improvements, for MikroTik it's only available not too long ago in 2022ish. (https://forum.mikrotik.com/viewtopic.php?t=188420) Since I have spare CPU and good algorithm, it seems like a good choice to use for me.
I'm not sure whether ASUS has cake or fq-codel algorithm built in already. If not, probably that's the reason you think it's doing more harm than good.

 

[xiaofan]

Yes, of course you may not need it. But wouldn't it be nice if you could BT, download, gaming at the same time and be sure that gaming is not affected? (After reading the comments I remembered indeed I used to stop BT tasks before gaming. Then probably forgot to start them after gaming.)

It's not too uncommon to do all three at once, for a gamer that has a second screen and a NAS. If there are more than one person at home, it's even more common.

Also the SQM algorithms have some recent improvements, for MikroTik it's only available not too long ago in 2022ish. (https://forum.mikrotik.com/viewtopic.php?t=188420) Since I have spare CPU and good algorithm, it seems like a good choice to use for me.
I'm not sure whether ASUS has cake or fq-codel algorithm built in already. If not, probably that's the reason you think it's doing more harm than good.

Good, you belong to the users who need SQM.

It is interesting that you say it is "not too uncommon to do all three at once". To me it is really rare. And I believe 99% (or 99.9%) or more of the Singapore Internet users do not do all three at the same time.

BTW, I have mentioned many times I do not do BT and I support some ISPs' policy to throttle the BT traffice. I use Singtel and Starhub myself and both throttle BT traffic.