Can Router override TCP Window Size (do it's own auto tuning?)

[kfp]

And yeah know the link i gave stated Websockets and that it used TCP, but thought the latency was determined by pinging.
I guess it looks at the differences between the current and last packet or something.
Or maybe the ACK response after the Window Size has been reached.

I think it just uses that term because everyone is familiar with it, I did not observe any ICMP Pings when it was doing the test.
Basically, latency can be determined from any protocol, not exclusive to ICMP pings.

As for your results, that's quite interesting. It seems like your ISP is handling bufferbloat well (with Auto Tuning off).
Can you stream something on your phone or another device and do the test again (QoS & Auto Tuning disabled)?
Also, do you get a public IP from your ISP?

 

[zerowalker]

Ah i see, well that makes things more clear

Yeah it's quite well, Download always works fine, Upload is affected by the Auto Tuning.
What i don't like is how slow the start is for the Upload, it's super slow (ramp) compared to the download, which i guess is connected to the auto tuning stuff anyhow as it's agreeing on the size.

I can try from my Iphone, but not sure what you mean with Auto Tuning there, the one i disable is on my PC, so obviously only that one is affected.
Doubt i got any control over Auto Tuning from an Iphone though.

Doing some tests it seems to vary quite a bit, and i am not that into the Wireless Settings, was playing with them yesterday to just decrease the power etc,
as i don't use any wireless device for anything important in terms of latency nor speed most of the time anyway.

 

[kfp]

I can try from my Iphone, but not sure what you mean with Auto Tuning there, the one i disable is on my PC, so obviously only that one is affected.
Doubt i got any control over Auto Tuning from an Iphone though.

I’m mainly just curious how the test would come out if there are other traffic on the network. Test on the same device, not the iPhone; iPhone just used to generate traffic.

Are you getting a public IP from your ISP or not? There are some newer ISPs that don’t own enough IPv4 space for their customer base so they have to NAT them in a local subnet. I’m just wondering if that’s the case and how that would affect bufferbloat.

 

[zerowalker]

Ah you meant like that, so basically run the tests in parallel and see what happens.

Think i should have a public one, have the same IP as long as i don't change NIC or mess with the Network Address in the settings (or change the MAC in the router etc).

 

[kfp]

have the same IP as long as i don't change NIC

That's normal for LAN side IPs, I'm talking about WAN IP.
If you go to http://checkip.amazonaws.com/, and compare that with WAN IP in the web UI of Asus, are they the same?

 

[zerowalker]

That's the IP i was talking about sorry.
Didn't mean the 192.168.x.x, but the standard one. And that one i can kinda change on demand, and it's based on the MAC or NIC Network Address thingy setting (whatever that is).
So i can for example switch between 2 ips back and worth if i want. Not sure how the ISP does it, but i guess it does some signature check and select an IP based on it.

And yeah, the Router show the same IP as that site.
Why wouldn't it do that though, is that when there's a subnet, meaning i would get an ip that looked real, but later it would switch somewhere resulting in a different one?

Heard they did something like that with ppl that had IPv6, they could fall back to IPv4, but it was shares among others or something.

 

[kfp]

And that one i can kinda change on demand, and it's based on the MAC or NIC Network Address thingy setting (whatever that is).
So i can for example switch between 2 ips back and worth if i want. Not sure how the ISP does it, but i guess it does some signature check and select an IP based on it.

I’m confused, can you describe how you change between the two IPs?

And TBH asking all this is probably not improving your latency or anything, I’m just curious. It’s cool if you want to continue this via PM or just call it a day, I’m fine either way.

 

[zerowalker]

Well i suck at explaining haha;P

I can continue on, i mean i learn stuff, even if it's not tied to improving something, knowledge is power!

Skip the "NIC Network Address" setting, quite sure it worked, but no clue how (i just gave som random number and it would give a different IP when connected back to the ISP).
It does NOT work when connected to the Router (which make sense, as the router is the thing the ISP sees).
But it doesn't seem to affect the MAC Address either on my NIC when changing it, so no clue what it even does.


So to the part i know works, if i got my ASUS Router connected to the Internet and i go to WAN where i can set a MAC Address.
If i give it nothing (i guess it uses the Routers own address?) it gives me one IP, if i give it my PCs, it gives me another IP.

If i switch back and worth, i will always get the same IP as far as i know.
So one MAC == one IP basically.

Now it's impossible that's how the ISP does it, cause it wouldn't make sense, but it does play a role in how it gives me the IP at least.

example:

e0:cb:4e:7a:94:e8 gives me 78.67.154.24

blank gives me 78.71.76.243

Hope that made sense xd

 

[kfp]

Ha, it does now. I wonder why turning off Auto Tuning improves bufferbloat that much for you, maybe ISPs are finally doing something right?

Maybe it’s because your line is too fast for your router’s CPU, which model are you using?

 

[zerowalker]

The Auto Tuning affects it both via Router or directly into the wall. (That's how i noticed that was the culprit eventually).

It's RT-AC56U.

The CPU on the WebbUI does max out if i get near 100mbps on either (not sure what happens if both does it at the same time, would be neat to have a speed test for that).
But looking at the "tops" or whatever the command for checking the processes in linux is, it seems it's not using all power,
but it's some "Software Interrupt" process that starts to do stuff, so i guess it blocks much of the normal stuff which results in it showing a super high CPU Usage?

EDIT:

Fun fact about the IP,
my normal IP that i use can't reach DslReports anymore (guess i got blocked from testing so much:s)?).
I couldn't figure out why i wasn't able to get on the site, so i tried changing the IP, and then it works.
But switched back in hope that they will lift it after some time xd

 

[kfp]

OK yea that’s under-powered for 100Mbps connection if you want to use QoS.

What does your phone get for bufferbloat test? (with or without other traffic, doesn’t matter, just curious if your ISP handle network stack on other OSes differently)

The IP ban is most likely automatic and temporary.

 

[zerowalker]

Oh that sucked, good that QoS doesn't seem to give me anything then;P

Well was gonna test that but have forgotten my damn passcode and only got one retry,
so have to wait till i can find it out or reset the device, no clue how i can't figure it out,
i have to write it all the time, but it's automatic so don't really remember the actual numbers, just somewhat -_-.

Will report back once i am able to do the tests.

 

[kvic]

So, to go back to the CTCP stuff, i noticed i had a really bad Bufferbloat, even though i got 100/100 fiber.

With a symmetric 100Mbps connection, you shall not worry about buffer bloat. It only happens when you push towards the limits. That's rare for a fat pipe in a SOHO environment..