Need help with Bufferbloat at 400/100 speed

[Snipzy]

I upgraded my speed connection to fiber with 500 dl and 250 upload but overall it gets 400 / 100, i have bufferbloat problems in download only.
My router is an AC66u-B1 with merlin installed, i've done the test with and without adaptive qos, tried FlexQoS but nothing, currently is disabled.
Also tried with chrome and firefox.

 

[geobernd]

Your router can't handle FlexQoS with 400/100. But the bigger question is:
Do you have any real use issues beyond the test showing 100ms download Bufferbloat? If not then just ignore it...

 

[Tech9]

Your router is >10 years old hardware. Disable all TrendMicro components for a better full ISP speed chance and don't look at online bufferbloat tests. If you have no real use issues you are fighting with non-existing problem.

 

[Snipzy]

Your router can't handle FlexQoS with 400/100. But the bigger question is:
Do you have any real use issues beyond the test showing 100ms download Bufferbloat? If not then just ignore it...

I mean i do gaming and when all the network is on use, i have a lot of lag spikes, before fiber i had docsis and it worked really good but ofc it was a lower speed

 

[Snipzy]

Your router is >10 years old hardware. Disable all TrendMicro components for a better full ISP speed chance and don't look at online bufferbloat tests. If you have no real use issues you are fighting with non-existing problem.

Lol that did a lot, the dl speed increased to 500 and latency down to 60ms but i assume you recommend changing to a newer model which one you sugest?

 

[Tech9]

Just stop looking at this bufferbloat site. You don't have bufferbloat with normal use. The site stresses your ISP line and shows you worst possible condition. If the router does what you expect it to do - continue using it. I don't recommend replacements for no reason. What is your reason to upgrade?

 

[Snipzy]

Just stop looking at this bufferbloat site. You don't have bufferbloat with normal use. The site stresses your ISP line and shows you worst possible condition. If the router does what you expect it to do - continue using it. I don't recommend replacements for no reason. What is your reason to upgrade?

Wow i didn't know that about those sites, then i don't have a reason to upgrade, thx for the help

 

[Rich Brown]

@Snipzy - let me summarize what I'm hearing here:

1. You say you have lag when you're gaming - so you DO have a problem.
2. The Bufferbloat test sites simply make the bufferbloat repeatable. (It likely happens all the time, although briefly, anytime someone (someone else?) downloads or uploads data, reads email, browses the web, etc. when you're gaming.)
3. Other posters say your router is too slow to handle 500/250

A possible contrarian solution: Try dropping your speed from your ISP, and tuning the adaptive QoS or FlexQoS. I will wager that:

a) you will work better because you're not maxing out the router, and it'll be able to shape the line
b) you'll save money with a lower cost plan from your ISP
c) you won't need a new router
d) you won't notice any difference in day-to-day operation from the lower speeds

I'd be curious to hear what happens...

 

[dtaht]

Thx rich for weighing in. I agree with your suggestion, in that it seems likely your router is struggling to keep up. (I gave up and switched to x86 openwrt boxes).

BUT: If there was some way to address the folk saying "You don´t have bufferbloat with normal use" definitively once and for all?
NORMAL USE has TCP "SLOW START" in it. Please read about that in wikipedia or see the explanation here:

*Every connection that lasts long enough* will hit the buffer until it runs out of space and gets a drop, then the rate will be cut by 1/3 (tcp cubic) and then start to grow again (¨Congestion avoidance¨), until it buffers and gets another drop. So a single tcp connection that lasts long enough will induce latency on the link, and on slow start, especially, jitter.

Thereś a key caveat here - "lasts long enough", and here is where people draw the conclusion - most web traffic has a bunch of short flows that in sum rarely exceed 20Mbits/sec at any given instant. At a 400Mbit level it will take seconds of a big download (say, a game) to finally hit this behavior. Or, one or more of the other devices on your network might do you in in this way. Microbursts exist, too...

Secondly, the uplink here is amazing (0?), and I am not sure why it is that good!

Thirdly, 100ms is really quite decent for downlink bufferbloat, and I might even agree with the commentors that it is unlikely to be much of a problem, except that I do not trust this test to be accurate at this speed! In a single queued system, 30ms would be preferred, or with a FQ´d system 0 is feasible. I keep hoping more will ask their ISPs to install something like LibreQos.io´s CAKE implementation on their downstream shapers to achieve that, (confession, I am part of that project, and the basic software is free, and a cheap Xeon can shape 10k subscribers).

Yes, you can also shape on your own the ingress but it does require heftier CPE, and is nowhere near as effective as the ISP doing it.

In the many years since the bufferbloat effort started, we have had many advances across the board - device updates, for example, tend to use a much gentler protocol than cubic, netflix does better pacing, as does all of linux, but for the best latency and jitter, sqm at the ISP -> down and cake -> ISP up remains the penultimate never worry about bloat/jitter/latency ever again thing.

 

[sfx2000]

I upgraded my speed connection to fiber with 500 dl and 250 upload but overall it gets 400 / 100, i have bufferbloat problems in download only.
My router is an AC66u-B1 with merlin installed, i've done the test with and without adaptive qos, tried FlexQoS but nothing, currently is disabled.

Probably time to get a new router/AP... the AC66U is going to have problems with a 500/250 connection with the typical QoS solutions - you're compute bound here

as @Tech9 mentions - disabling the trendmicro and other options likely will help, and enabling NAT acceleration (aka CTF.ko) will help...

 

[sfx2000]

Thirdly, 100ms is really quite decent for downlink bufferbloat, and I might even agree with the commentors that it is unlikely to be much of a problem, except that I do not trust this test to be accurate at this speed! In a single queued system, 30ms would be preferred, or with a FQ´d system 0 is feasible. I keep hoping more will ask their ISPs to install something like LibreQos.io´s CAKE implementation on their downstream shapers to achieve that, (confession, I am part of that project, and the basic software is free, and a cheap Xeon can shape 10k subscribers).

Actually the 100ms downlink is a bit high for a 500/250 connection...

QCA Cypress and Hawkeye can do better without any problem - I would think that the Broadcom solutions for 80211ax would be similar...

QCA Dakota isn't far behind for a sub-1Gbit/Sec connection...

 

[dave14305]

It would be nice if Asuswrt would adopt fq_codel as the default qdisc (when no other QoS is enabled) the same way OpenWrt did some time ago. Still using pfifo.

 

[Snipzy]

@Snipzy - let me summarize what I'm hearing here:

1. You say you have lag when you're gaming - so you DO have a problem.
2. The Bufferbloat test sites simply make the bufferbloat repeatable. (It likely happens all the time, although briefly, anytime someone (someone else?) downloads or uploads data, reads email, browses the web, etc. when you're gaming.)
3. Other posters say your router is too slow to handle 500/250

A possible contrarian solution: Try dropping your speed from your ISP, and tuning the adaptive QoS or FlexQoS. I will wager that:

a) you will work better because you're not maxing out the router, and it'll be able to shape the line
b) you'll save money with a lower cost plan from your ISP
c) you won't need a new router
d) you won't notice any difference in day-to-day operation from the lower speeds

I'd be curious to hear what happens...

I have problems with gaming when all of the devices are used like tvs, multiple phones, tablets etc… and my latency in a game like rocket league increases by 30ms and sometimes up to 100 but with loss of packages.
I haven't test it with the recommendations of trend micro disabled but overall is looking good

 

[Snipzy]

Thx rich for weighing in. I agree with your suggestion, in that it seems likely your router is struggling to keep up. (I gave up and switched to x86 openwrt boxes).

BUT: If there was some way to address the folk saying "You don´t have bufferbloat with normal use" definitively once and for all?
NORMAL USE has TCP "SLOW START" in it. Please read about that in wikipedia or see the explanation here:

*Every connection that lasts long enough* will hit the buffer until it runs out of space and gets a drop, then the rate will be cut by 1/3 (tcp cubic) and then start to grow again (¨Congestion avoidance¨), until it buffers and gets another drop. So a single tcp connection that lasts long enough will induce latency on the link, and on slow start, especially, jitter.

Thereś a key caveat here - "lasts long enough", and here is where people draw the conclusion - most web traffic has a bunch of short flows that in sum rarely exceed 20Mbits/sec at any given instant. At a 400Mbit level it will take seconds of a big download (say, a game) to finally hit this behavior. Or, one or more of the other devices on your network might do you in in this way. Microbursts exist, too...

Secondly, the uplink here is amazing (0?), and I am not sure why it is that good!

Thirdly, 100ms is really quite decent for downlink bufferbloat, and I might even agree with the commentors that it is unlikely to be much of a problem, except that I do not trust this test to be accurate at this speed! In a single queued system, 30ms would be preferred, or with a FQ´d system 0 is feasible. I keep hoping more will ask their ISPs to install something like LibreQos.io´s CAKE implementation on their downstream shapers to achieve that, (confession, I am part of that project, and the basic software is free, and a cheap Xeon can shape 10k subscribers).

Yes, you can also shape on your own the ingress but it does require heftier CPE, and is nowhere near as effective as the ISP doing it.

In the many years since the bufferbloat effort started, we have had many advances across the board - device updates, for example, tend to use a much gentler protocol than cubic, netflix does better pacing, as does all of linux, but for the best latency and jitter, sqm at the ISP -> down and cake -> ISP up remains the penultimate never worry about bloat/jitter/latency ever again thing.

I have an old laptop that is no longer used but it only has one rj45 port with 100mbps speed when i had the slower connection i used an usb to ethernet and installed openwrt but the speed was awful.
Idk if buying a mini pcie to 2 ethernet ports would be good and give it a try or maybe the laptop is the problem as it has 1.5 ghz dual core

 

[L&LD]

It's not you, it's the laptop.

 

[Snipzy]

It's not you, it's the laptop.

Oh, I heard that a raspberry pi 4 would be a good solution but then i heard again one problem is the usb adapter

 

[Wistuplu]

before fiber i had docsis and it worked really good but ofc it was a lower speed

Try dropping your speed from your ISP, and tuning the adaptive QoS or FlexQoS. I will wager that:

a) you will work better because you're not maxing out the router, and it'll be able to shape the line

Hello Snipzy,

I have nowhere near the knowledge of dthat and most around here.
But let me suggest a cheap and easy experiment.

You report that it worked really good with docsis, at a lower "speed" (throughput).
So, setup Cake on your router.
As speeds for download/upload, choose the speeds of your old docsis plan. (BTW, what was it ?)
Let us know if you still see "a lot of lag spikes" that you are seeing today (probably not).
And how the house is generally reacting to the change (or not). I mean real life experience after a few days. Not some test result from a website.

It is somehow similar to the idea of "Try dropping your speed from your ISP", but no phone call or contract changes needed before you can start experimenting.

And of course, after a few days, you can also test with higher speeds. Maybe 200/90. Maybe 250/95. Maybe 300/98, as long as you're not maxing out your router shaping capabilities (I guess the waveform site results can indirectly inform you about how the router can cope with higher throughput).

dthat, thanks a million for the videos, website, evangelizing and all your concrete steps to improve this.

 

[coxhaus]

I have been saying for years don't worry about QoS and just buy a faster connection. Of course, you need to buy faster equipment to run the faster connection speed.

Use an older low wattage PC with Pfsense which will work and is pretty much unlimited on speed today as a newer PC will drive 10 gig which should handle any home connection or small business.
No real trick here for higher speed. QoS was needed back in the old slow DSL days when you could not get faster connections. DS3 and OC3 circuits were too expensive back in the old days for home use.

PS
Personally, I think it is a dumb idea to drop your speed to run QoS. It will increase your latency to use QoS.

 

[RMinNJ]

"..I upgraded my speed connection to fiber with 500 dl and 250 upload but overall it gets 400 / 100, i,,"

When I upgraded to gigabit my old i3 intell cpu with a gigabit interface could not do full gigabit.. Getting a faster laptop I could. I then found my Tp-Link gigabit switches could not do full gigabit. Definition get a better laptop for 500 Mbit service.

 

[L&LD]

See post 15.