Wow, quite complex description worth a good diagram :D

Regards,

Grzegorz

From: Eric Kuhnke <eric.kuhnke@gmail.com>
Date: Friday 2021-04-23 at 17:23
To: Thomas Schäfer <tschaefer@t-online.de>
Cc: "ripe-atlas@ripe.net" <ripe-atlas@ripe.net>
Subject: Re: [atlas] Satellite based "last mile" and Atlas probes

MTU being only 1280 can't be right, my connection looks like a normal 1500 MTU... 

The v4 IP shown in the measurement which is only 10ms from something, I do not think the ICMP echo is faked, that's the terrestrial gateway side of the cgnat. Which in the examples of my own starlink terminal and several others, is some item of Google equipment located near a major city's IX point. The "public" side of the cgnat for my terminal is less than a millisecond from core stuff in Seattle.

Starlink terminals which exit to the Internet through a gateway in Chicago, have global-routing-table facing sides of their cgnat only 1 to 2ms away from several ISPs' looking glasses at 350 E. Cermak and other IX points in Chicago. But the end to end latency from a terminal is more like an absolute minimum of 16ms, more often an average of 23ms.

On Fri, Apr 23, 2021 at 8:09 AM Thomas Schäfer <tschaefer@t-online.de> wrote:

Am Freitag, 23. April 2021, 11:18:17 CEST schrieb Tim Chown:

>
> Is that a local choice, or does Starlink not carry IPv6?

A very interesting setup was yesterday documented by awlnx via twitter.

https://twitter.com/awlnx/status/1385263694649700353

She got it working. Some things I can remember:

inbound is open, she presented an IPv6-only webserver,
at least yesterday it was functional:
https://atlas.ripe.net/measurements/29792632/#general

there are some strange things: mtu is only 1280
icmp echo request answers are faked

Regards,
Thomas