YouTube Buffering and Slow Loading on BT Full Fibre 900 – Possible Routing/Peering Issue

[CarlH]

Messing with DNS does not help either you will end up at rr?---sn-cu-cimy.googlevideo.com whatever you try.

[tyencyu]

@CarlH 

Yeah it's absolutely not a DNS problem

The GGC node you end up at as regional, it seems that a chunk of the North West / N.Wales / Midlands share the same regional node, and there's congestion because of a botched parameter forcing connection speeds down at ingest... So I'm told. 

[Kamek]

I filtered for videoplayback and the request is going to rr5---sn-cu-cime7. The URL also has mn=sn-cu-cime7,sn-ajaig5-5a and the ip parameter is IPv6, starting 2a00:23c7, so it looks like my YouTube video traffic is going via IPv6 to the same sn-cu-cime7 cache cluster.

[tyencyu]

@Kamek  Awesome stuff

So yeah you're on the same ***.---sn-cu-cime7 node that I think most of us in the general region are on, and then if you turn on a VPN you should see it swap to an entirely different node.

I believe it's the initcwnd configuration that's invalid, which I think can be obtained via packet capture but I don't have wireshark currently.

[Kamek]

And just to confirm, I'm right in the middle of being North West/Midlands, so it's all adding up to a problem for that region (plus North Wales, apparently), as far as I'm concerned anyway.

[Kimberlin]

@tyencyu  Undoubtedly then why I'm completely unaffected. I'm no where near the node locations.

[tyencyu]

@Kimberlin Sounds about right to me, also explains why friends deeeewn south also don't save the same issues

[Steve_M]

Thanks for all the investigations and feedback so far. I'm feeding it into the Networks engineers

[Kimberlin]

@tyencyu  I think it's spoken 'daaaan souf' which would indicate somewhere in the Saaaaauuufff East region.

I'm no where near there, ooohhh arrggg ooohhhh arrrggg!

[tyencyu]

Thank you @Steve_M , Much appreciated 🙏

It might be something they can put in monitoring for, as I understand it, it's a bit of a blindspot because as far as reachability is concerned, there's technically not an issue, the node is perfectly reachable, but the service is highly degraded as a result and seems to have slipped monotoring