Buffer Bloat

[SpellowHouse]

I think that's what it's called, anyway.

I discovered that my upload lag goes out the window then there is a load on the router. By limiting the speed of the connection to about 2/3 maximum, then this problem is averted. This seems out of my control, nothing I can do about it. The problem is that I don't want to limit the router all the time to 2/3 the maximum speed of the connection.
I know that some routers have the ability to solve this issue, but it depends on them detecting the fact that I am playing a game.
I don't know of many routers that actually give any stats or feedback on this, but perhaps more importantly which allow me to configure it, and which reliably detect a game.

 

[SpellowHouse]

Yes. Running the test I do have an issue with upload. Alas my router seems to have an issue that it won't limit upload speeds so I can't do anything about it at all.

 

[SpellowHouse]

There have been whole discussions about this. Search for Dave Taht (VM thread) and related discussions. If you're technical, you can build or flash a router with OpenWRT and practically eliminate bufferbloat with a few clicks. IPFire and OPNSense also allow you to set bufferbloat controls, and probably others, but OpenWRT with cake is the SQM gold standard. Eero Pro 6 can kill bloat basically automatically up to 1Gbps. What Internet connection do you have, and what speed (eg fibre 500Mbps, cable 1Gbps etc)?

Run this test for a basic overview of how your line performs, though Flent's RRUL test is again the gold standard - just more complicated to set up and run if you're new to all this. You can start to read more about bufferbloat and how to combat it here.

I have Zen Fttp 300mbps.
The download latency is good but the upload collapses under load.
The connection is rated as C. Under load, the latency increases by 100ms, which is a really noticeable difference.
A few years back I very specifically bought a router recommended to combat Buffer Bloat, and being non-technical with network stuff, I got a popular model that uses English rather than techno-speak!! Problem is that that router has a bug in the anti-buffer bloat and doesn't work with pppoe and I have been waiting 18 months now for a fix.. still nothing.

 

[SpellowHouse]

That looks good to me. Here's mine for comparison (cake). My downstream is capped a little aggressively (gigabit) but my VM line is woefully useless without it. I'll be ordering FTTP in the coming months, and can't wait to get rid of cable.

OP, as said you need to switch routers. PPPoE is an awkward beast at times; stick to something Linux based with Cake rather than anything *BSD or (preferably) codel based. Edit: You also didn't say which router it is. When you're seeking help/advice/input it's best to tell us everything in one go (ISP, line connection, network equipment, local devices, setup, OS etc etc) rather than drip feed things randomly over the course of a few days.

UGH!!!

This is the result with the new router.....



The amazing thing is that NOTHING is configured in the new router. It seems that the problem was the old router - which I specifically bought becuase it was supposed to minimise things like buffer bloat.

 

[SpellowHouse]

Consumer 'routers' are mostly overpriced useless tat, anyway, tbf. You still didn't mention what was the outgoing and the new router model(s)? It'd be interesting to know for future buy/avoid recommendations for others. I'm glad you got it sorted! Fibre doesn't tend to need much by way of SQM anyway, usually. Your router was probably slowly dying, overheating or similar and the packets were hitting IOWAIT or some such, causing the symptoms you experienced. All's well that ends well. Enjoy (and don't forget to tell us what the routers were/are! ).

I am almost embarrassed to say now it was a netgear XR1000. *cringes* Sorry, I fell for the hype, but in my defence I am a network idiot. The new one is an Asus with Merlin. Not that it needs anything - out of the box it is working way better than the netgear.

 

[SpellowHouse]

@SpellowHouse As stated you won't get away with that on VM and FTTC, I have seen a few people that have posted results the same as yours that have moved to FTTP.
Happy Days!

On the surface I am guessing that the XR1000 has a throughput issue. I didn't spend much time quantifying it, but my guess is that it's around 150Mbps. Anything after that and it starts to introduce the very problem it's supposed to solve. This may be down to the pppoe, I don't know.

I really like DumaOS. It's an incredible interface for a router. However, I am afraid that my own experience of them is that their products are riddled with bugs. I have tried all of their products. It's pretty rare for me to even find a bug in a router, it's not something I usually use all the features of, but I have stumbled from one bug to the next in their routers. It seems utterly incredulous to me that their premium router is guilty of the very thing it is supposed to prevent.
My guess on Netduma is that their testing is failing them. Not only do they release products with tons of bugs, but they seem to tie themselves in knots testing, and so can't get updates out in a reasonable time. I know for a fact they have been aware of problems with pppoe for 18 months and yet still haven't released any new firmware because it's being passed backwards and forwards between their own testing department and Netgears. They just can't seem to get anything released.
I wish them well in future, but I've had it with them now. That was the last one of their products you will find me buying!

Apart from DumaOS, I have a long history of buying Asus routers. Although personal experience is different for everyone, Merlin has always been reliable. I have moved back to Merlin now and will stay there (I did buy a Draytek once but found the speak just too technical for me, so I never actually did anything with it cus I didn't understand it!!).

Anwyay, thanks for the help. Everything is sorted now.

 

[SpellowHouse]

Thanks, ah see that it suggests to test hard wired, most of my devices are wifi so im not going to get precise results over wifi.

In passing, it's worth buying a really long ethernet cable for the ocassional test. It's not like they are expensive.

 

[SpellowHouse]

Thanks, ah see that it suggests to test hard wired, most of my devices are wifi so im not going to get precise results over wifi.

Test it anyway. At least it will assure you everything is fine if it passes the test. If it doesn't pass then you can consider a long ethernet cable to take the testing one step further.

 

[SpellowHouse]

Sorry, it (quod erat demonstrandum) means you've proven the point. You're asking why your runs are variable, and then say they were performed over WiFi (which is, inherently, variable and subject to interference).

Exactly this. Wifi is not a predictable thing, unless you are a telecoms company and have access to seriously expensive kit. It is hihgly prone to interference from other sources.

(It still amazes me that my Wifi has trouble picking up my old laptop at 10 feet while BT happily chat to it from two miles away.)

 

[SpellowHouse]

More experimenting today.

I think that the connection is generally a bit poor. Extended testing shows that there does seem to be an issue with bufferbloat, but I am not sure how to fix it.

I tried Adaptive QOS and almost as expected that did little. Cake on the other hand worked immediately and seems to be extremely good. I have never heard of Cake before so I am reserved about it. Does Cake need the bandwidth setting to 10% below maximum? (That's what I currently have it set to).

Well, I set it to about 7% below and that seems to work just fine.

 

[SpellowHouse]

Whats the exact model of your router please?

In fact over the next few hours I tested it again and again, and that result was just a one-off. Most of the time the result was the same as the old router. The one huge difference is that with the old router anti-buffer bloat was not working at all so it made no difference. I suspect this was because my ISP uses pppoe and for some reason the XR1000 really falls apart with pppoe. If I placed a router in front of the XR1000, so the first router was doing the pppoe, amazingly the buffer issues disappeared. However, using two routers is not an option really.
The new router runs Merlin, which is really important. That has Cake, which has completely eliminated the problem. Cake is included in Merlin for all of the supported routers, I believe, which are all Asus.

 

[SpellowHouse]

Who is your ISP?

Zen.

 

[SpellowHouse]

Aha! Been migrated by any chance?

Yes. Why is that?

 

[SpellowHouse]

Because their new network is ****.

Ah!

 

[SpellowHouse]

Because their new network is ****.

Interestingly, I just talked to Zen and they told me pppoe is optional. There is a way to configure the router without selecting it. Am chasing this up....

 

[SpellowHouse]

DHCP or PPPoE won't make any difference if the issue lies within Zen's network.

Do you have a Think Broadband ping monitor enabled?

I guess. It was just interesting that the only reason I junked the XR1000 was it had a bug in pppoe, not it seems there was a way to "not" use pppoe. Anyway, not bothered really, I just got sidetracked.
I'm not too concerned with the poor performance, Cake has taken care of it. Just that it is pretty poor. I doubt somehow that Zen have any performance tolerances for latency??

 

[SpellowHouse]

Nope, they told me up to 70ms to a London server is acceptable for full fibre.

However they state in their terms and conditions that they can only make changes to your service without notifying you if it isn't going to be a signifant change. I argued that tripling my idle latency is significant. I got out and moved to Aquiss on the same line and back to ~6ms from mid 20's at times with Zen post GEA migration, and much more consistent speeds.

The performance without Cake is poor, but with Cake its consistantly ....



You can see I have lost some performance there. I am nolonger achieving 300/50Mbps but that's really not too bad.

I'll look in to Aquiss though..

 

[SpellowHouse]

I have an AC86U and the puter hardwired to the router running the latest Merlin (386.10?) with Cake enabled and I don’t get A+

What am I doing wrong? It’s FTTP 500/500 and I get A with 10-15ms jitter

I don't know how you have Cake configured, but I reduced my bandwidth in the Cake settings until I got 0 0. For me that was a reduction of about 7% or so on both Upload and Download. Cake seemed to take a little more, by itself, but that was probably my fault in the other parts of the configuration. Anyway, it needed it. Using a Zen broadband speed test (which I would expect to be the best possible result) I am getting 274/47 which is really good.