Buffer bloat test and sharing results

Forums › New Zealand Broadband › Buffer bloat test and sharing results

cddt

1965 posts

Uber Geek
+1 received by user: 1904

#318022 8-Dec-2024 09:35

I recently came across the "buffer bloat" test: https://www.waveform.com/tools/bufferbloat

Basically it seems to test how well your router priorities latency-sensitive applications when the link is saturated.

I'm interested in any thoughts on this test, is it useful? Could be more useful than a speed test in some situations.

I'm also interested to see other people's results with different routers. My aging HG659 scores a B.

More reading: https://www.bufferbloat.net/projects/

My referral links: BigPipe | Mercury

1 | 2

3818 posts

Uber Geek
+1 received by user: 1366

Lifetime subscriber

#3317943 8-Dec-2024 09:42

A grade for Mikrotik CCR1009-8G-1S-1S+.

Spark Max Fibre using Mikrotik CCR1009-8G-1S-1S+, CRS125-24G-1S, Unifi UAP, U6-Pro, UAP-AC-M-Pro, Apple TV 4K (2022), Apple TV 4K (2017), iPad Air 1st gen, iPad Air 4th gen, iPhone 13, SkyNZ3151 (the white box). If it doesn't move then it's data cabled.

Behodar

11094 posts

Uber Geek
+1 received by user: 6071

Trusted

Lifetime subscriber

#3317945 8-Dec-2024 10:15

Synology RT6600ax: A

Jase2985

13730 posts

Uber Geek
+1 received by user: 6202

ID Verified

Lifetime subscriber

#3317946 8-Dec-2024 10:17

Spyware:

A grade for Mikrotik CCR1009-8G-1S-1S+.

ditto for a Grandstream GWN7000

richms

29098 posts

Uber Geek
+1 received by user: 10208

Trusted

Lifetime subscriber

#3317951 8-Dec-2024 10:41

UXG pro got me an A, but I don't think the computer could really saturate it in the test as it was only giving 870 and 380ish for the speeds and the CPU was pegged at 100% on some cores

Richard rich.ms

SpartanVXL

1498 posts

Uber Geek
+1 received by user: 666

#3317953 8-Dec-2024 10:57

Doesn’t the router need to implement a queue? Fq codel or cake iirc.

This isn’t really needed if you have a lot of bandwidth, and since fibre came around most people don’t worry about it. It has overhead on routers too which means better cpu/hw accel is needed.

k1w1k1d

1711 posts

Uber Geek
+1 received by user: 1305

#3317954 8-Dec-2024 11:04

ASUS RT-AX3000 scored an A.

HP

Shop now for HP laptops and other devices (affiliate link).

CYaBro

4708 posts

Uber Geek
+1 received by user: 1182

ID Verified

Trusted

#3317955 8-Dec-2024 11:10

Grade A for UDM SE.

Opinions are my own and not the views of my employer.

freitasm

BDFL - Memuneh
80646 posts

Uber Geek
+1 received by user: 41030

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#3317960 8-Dec-2024 11:53

I came across this test page a couple of months ago. I remember testing and thinking this will be pointless with New Zealand broadband. It makes a lot of sense in the USA with their not so great broadband plans and low bandwidth availability.

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

cddt

1965 posts

Uber Geek
+1 received by user: 1904

#3317963 8-Dec-2024 12:04

freitasm:

I came across this test page a couple of months ago. I remember testing and thinking this will be pointless with New Zealand broadband. It makes a lot of sense in the USA with their not so great broadband plans and low bandwidth availability.

Not sure I agree. The test is measuring what happens to latency-sensitive applications when the connection is saturated by other applications - and a connection can be saturated whether its 3, 30, or 300 Mbps.

My referral links: BigPipe | Mercury

freitasm

BDFL - Memuneh
80646 posts

Uber Geek
+1 received by user: 41030

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#3317964 8-Dec-2024 12:13

Yes, sure. But then, if you are running latency-sensitive applications, you would pause your 2000 seeding torrents during that so important job interview video call, right?

My point is that different things impact performance: asymmetrical broadband speeds, low-spec routers, Wi-Fi interference causing collisions and retries, bad adapter driver software implementation, etc. And this is only inside your network.

Was the test performed while nothing else on the network was running? Or was the network under load? This is the kind of thing your OS won't know. The load was artificially created by that specific test computer, but what if something else is already running on another device?

This test might indicate your router is crap - or is it your computer? - and even changing it may not solve problems without a hard look at everything going on - not only hardware and software but usage patterns too.

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

Mikek

112 posts

Master Geek
+1 received by user: 42

#3317966 8-Dec-2024 12:18

I Feel like the download did the damage using the connection but upload was not pushed enough to really test the jitter.

AliExpress

Shop now on AliExpress (affiliate link).

cddt

1965 posts

Uber Geek
+1 received by user: 1904

#3317977 8-Dec-2024 13:50

freitasm:

Yes, sure. But then, if you are running latency-sensitive applications, you would pause your 2000 seeding torrents during that so important job interview video call, right?

That's right, in a single-user scenario it's easy to manage the overall connection usage so this type of buffering never becomes a problem. I still think it could be relevant in scenarios when there are several users, e.g. a flat with multiple people who might be saturating the connection occasionally (or frequently) with large downloads (e.g. game installs or updates are often tens or hundreds of gigabytes now; linux isos as you mentioned; even OS updates can temporarily saturate a connection).

My referral links: BigPipe | Mercury

SpartanVXL

1498 posts

Uber Geek
+1 received by user: 666

#3318070 8-Dec-2024 18:07

Like i mentioned earlier, if you router needs to implement queues then it may reduce overall bandwidth due to processor overhead. Most consumer routers don’t bother as they’re just enough to provide 1Gb speeds without queues.

So it becomes a choice between max speed or queues to deal with bufferbloat but less bandwidth.

djtOtago

1181 posts

Uber Geek
+1 received by user: 605

#3318077 8-Dec-2024 18:51

A Grade Deco XE75. Test was done over WiFi
https://www.waveform.com/tools/bufferbloat?test-id=8c7531fa-7191-4e07-8462-b7d5556199ac

KiwiSurfer

1722 posts

Uber Geek
+1 received by user: 993

ID Verified

Lifetime subscriber

#3318078 8-Dec-2024 19:01

freitasm:

I came across this test page a couple of months ago. I remember testing and thinking this will be pointless with New Zealand broadband. It makes a lot of sense in the USA with their not so great broadband plans and low bandwidth availability.

I've come across bufferbloat issues on 300/100 connections in the past so it can happen.

One obivous symptom for bufferbloat for me is dropped frames on Zoom/Teams/etc calls. In my experience these can be pretty common on certain routers where there is several other connections in flight. Even as simple as computer A downloading a large file via Bitorrent and computer B on a Zoom call can trigger bufferbloat -- but have also seen it triggered in networks where there's a lot of active connected devices with lots of different kinds of activity.

You're right that in theoy it shouldn't be a big issue, but in reality some of the lesser router struggle with this. The worst offenders include some of the older RSP supplied gear e.g. the Netcomm router Slingshot used to supply some years ago bufferbloated so bad I quickly switched that out for a different router. I acknowledge current RSP gear are much better now which is great to se.

With my Edgerouter and its Smart Queue feature enabled I rarely see dropped frames on Zoom/Teams calls related to issues on my end (for example I rarely see everyone freeze at once -- if you ever see everyone freeze that is pretty strong indicator that the issue is on your end).

Interestingly I get an A+ with my Edgerouter -- 7ms latency with only +3ms when download active and +0ms (yep zero) when upload active. That tallies with my experience, I can be downloading a steam game at line speed, and if I jump into a Zoom/Teams/etc call I see zero dropped frames in Teams/Zoom/etc even with the download going full blast. Its brillant.

On the other hand a good way to see Bufferfloat for real is to tether to a phone connected any of the mobile networks here -- it is quite bad. 2degrees on my S21 for example is 51ms unloaded, +1712ms download active, but interestingly +0 upload active resulting in an F score. So it is still a relevant issue here in NZ -- but more of an issue on wireless networks perhaps.

1 | 2