FYI/A: Bufferbloat 101

[xiaofan]

Best Routers For bufferbloat mitigation for 2021

Sent from LG-H930 using GAGT

Basically at 1Gbps, the benefits of QoS and SQM become questionable and you need very powerful router to handle this. UDM Pro can only handle 800Mbpd SQM as per the article.

 

[uncle_josh]

Yes, understand SQM is quite cpu intensive. However I have seen bufferbloat cause the entire home network to lag. So this is a must have feature for my consideration for my next router.

Basically at 1Gbps, the benefits of QoS and SQM become questionable and you need very powerful router to handle this. UDM Pro can only handle 800Mbpd SQM as per the article.

Sent from LG-H930 using GAGT

 

[badsector]

 

[xiaofan]

Yes, understand SQM is quite cpu intensive. However I have seen bufferbloat cause the entire home network to lag. So this is a must have feature for my consideration for my next router.

That is of course good if you want to implement SQM.

Rather than considering expensive ones like UDM Pro, probably x86 based mini PC running pfsense or openwrt will be more cost effective to run SQM.

 

[badsector]

with openwrt SQM.

 

[xiaofan]

And it seems to me there are many factors affecting the buffebloat score. The ISP issued ONT/ONR, the router, the client device and the OS running on the client device, wifi or wired connection, all play a part.

And we can see many people can achieve good A/A+ grade without implementation of SQM in the router.

Then again, there are more important factors like wireless coverage which seems to be the number one issue in this forum.

Then the ISP routing is also important as it greatly affects the latency for international sites. I think that is much more important than bufferbloat which does not really come into the picture for most of the international sites (below 100Mbps).

So in the end, to me SQM is only nice to have. If you have a powerful router which can run SQM at fast rate, by all means enable the feature. But for most of the people, just live with whatever the bufferbloat scores you can achieve.

 

[hwzlite]

So in the end, to me SQM is only nice to have. If you have a powerful router which can run SQM at fast rate, by all means enable the feature. But for most of the people, just live with whatever the bufferbloat scores you can achieve.

The goal of the bufferbloat project was to “hold latencies low or constant, no matter how much bandwidth you have.”


The impact of bufferbloat can be profound and subtle.
Profound: Especially on those asymmetrical xDSL/Cable with low line speed.
Once line speed saturated, all hell breaks loose on latencies.
Subtle: On Gbps fiber, spike/jitter latencies ensured once line speed saturated.

Only obsessive gamer or netG33k will make every ms/rtt counts on taming latency-sensitive applications

fyi: My Past rants during my days on cable

 

[badsector]

off-peak

 

[TanKianW]

The goal of the bufferbloat project was to “hold latencies low or constant, no matter how much bandwidth you have.”


The impact of bufferbloat can be profound and subtle.
Profound: Especially on those asymmetrical xDSL/Cable with low line speed.
Once line speed saturated, all hell breaks loose on latencies.
Subtle: On Gbps fiber, spike/jitter latencies ensured once line speed saturated.

Only obsessive gamer or netG33k will make every ms/rtt counts on taming latency-sensitive applications

fyi: My Past rants during my days on cable

Wah, so long never visit VRzone until I click on your link...now like semi-dead forum......

 

[xiaofan]

with openwrt SQM.

off-peak

Nice. I assume your openwrt installation is on an x86 mini PC.

It will be interesting to compare the results without SQM on your openwrt installation. I tend to think it will still be A+ with your x86 based router and M1 500Mbps subscription.

 

[xiaofan]

The impact of bufferbloat can be profound and subtle.
...
Subtle: On Gbps fiber, spike/jitter latencies ensured once line speed saturated.

Only obsessive gamer or netG33k will make every ms/rtt counts on taming latency-sensitive applications

fyi: My Past rants during my days on cable

The thing is that you seldom saturate the line speed on 1Gbps Fibre, other than speedtest or connecting to some local download servers. Gaming will certainly not saturate the gigabit Fibre connection, or even 500Mbps connection.

My point is that the ISP routing and wireless connection are contributing more to latency. So for gamers, use wired connection, and choose better ISPs like Myrepublic and M1. And use a relatively good router. After that then you need to worry about bufferbloat.

 

[hwzlite]

The thing is that you seldom saturate the line speed on 1Gbps Fibre, other than speedtest or connecting to some local download servers. Gaming will certainly not saturate the gigabit Fibre connection, or even 500Mbps connection.

My point is that the ISP routing and wireless connection are contributing more to latency. So for gamers, use wired connection, and choose better ISPs like Myrepublic and M1. And use a relatively good router. After that then you need to worry about bufferbloat.

Yes if you assuming single user
But.... Typical scenario - 500Mps connection shared with several users :
User1 - fragging FPS game
User2 - Torrenting without app limiting download/upload speeds which potentially eat up all bandwidth.
User3 - same as above
User* - ...video streaming, FTP, normal file download....etc you get the idea.

 

[hwzlite]

Wah, so long never visit VRzone until I click on your link...now like semi-dead forum......

LOL, i have even forgotten my login pw there

 

[xiaofan]

Yes if you assuming single user
But.... Typical scenario - 500Mps connection shared with several users :
User1 - fragging FPS game
User2 - Torrenting without app limiting download/upload speeds which potentially eat up all bandwidth.
User3 - same as above
User* - ...video streaming, FTP, normal file download....etc you get the idea.

Good point.

I do not use torrent (other than to test how SingTel throttles torrent, using Ubuntu Linux ISO) myself so that is one use case I did not consider. Indeed I can get around 240-320Mbps and two users will indeed saturate a 500Mbps connection.

 

[badsector]

Nice. I assume your openwrt installation is on an x86 mini PC.

It will be interesting to compare the results without SQM on your openwrt installation. I tend to think it will still be A+ with your x86 based router and M1 500Mbps subscription.

yes. install on x64 mini pc.
dun think it's worth wasting time tuning it to get a A+
i've disabled SQM since actual usage there's no improvement at all

 

[xiaofan]

yes. install on x64 mini pc.
dun think it's worth wasting time tuning it to get a A+
i've disabled SQM since actual usage there's no improvement at all

Thanks for the confirmation.

 

[badsector]

Thanks for the confirmation.

at least i have no issues playing netflix on 3 TVs at home tonight
4x webcam...constantly uploading to xiaoyi-css-sg-15d.oss-ap-southeast-1.aliyuncs.com
without visitors.. i have abt 16 devices only

i'm not a heavy user

 

[hwzlite]

at least i have no issues playing netflix on 3 TVs at home tonight
4x webcam...constantly uploading to xiaoyi-css-sg-15d.oss-ap-southeast-1.aliyuncs.com
without visitors.. i have abt 16 devices only

i'm not a heavy user

4x webcam? You have a mini-fortress there

 

[xiaofan]

Read a bit more, now some consumer routers can do vendor specific QoS at near 1Gbps, like Xiaomi AX3600, Asus RT-AC86U, AX88U and AX86U, etc.

However they may not really implement SQM and the vendor specific QoS may not be as good as more proper SQM implementation.

And for the question of whether the link can get saturated or not, yes sometimes the link can be saturated momentarily and then bufferbloat can really cause problems.

So in the end, if you really care about this issue, get an x64 based mini PC and run openwrt/pfsense and run proper SQM implementation.

 

[snowwolf13]

I find that www.dslreports.com/speedtest may not able to correctly reflect or identify the bufferfloat as the laptop or browser could be a factor for the spike in ping.

Did a test by performing ping from another machine and running www.dslreports.com/speedtest, when there were ping spikes in the speedtest, there were none to little change in the ping. Even when I ran both the www.dslreports.com/speedtest and Ookla speedtest client at the same time to max out the bandwidth, www.dslreports.com/speedtest showed over 100ms but the ping only increased by 1ms.