[Starlink] musk on mean latency

[Starlink] musk on mean latency

[Dave Taht]

I would rather like starlink to get latency for
voip/gaming/videoconferencing down to consistently (99.8%) below 40ms.

from https://spaceflightnow.com/2024/01/13/live-coverage-spacex-to-launch-falcon-9-rocket-on-starlink-mission-from-cape-canaveral/…
“The biggest, single goal for Starlink from a technical standpoint is
to get the mean latency below 20 milliseconds,” Musk said.

mean latency is a somewhat wrong goal. It misses the impact of slow
start. I figure he is mostly describing the difficult layer 2 work it
requires to accomplish that, everything from doing sat handoffs and
rerouting stuff on the ground, to needing more ground stations, which
is indeed a giant task.  I keep hoping that in particular they get
away from the "recalculate connectivity and bandwidth every 15
seconds", into something more responsive for normal traffic. DNS
servers on the sats would help too! and does not fit into the concept
of "mean latency" at all.

... and me, all I want to see is cake on the "glitchy", fq_codel on
the wifi, and for them to find some way to leverage at least some of
the techniques in LibreQos (like ebpf and cake) to somehow manage
their downlinks better.

It does look like they are now achieving sub 30ms latency to 1.1.1.1
in some places for dns.

https://home.sjh.at/starlinksmokeping/?target=DNS.CloudflareDNS1


-- 
40 years of net history, a couple songs:
https://www.youtube.com/watch?v=D9RGX6QFm5E
Dave Täht CSO, LibreQos

Re: [Starlink] musk on mean latency

[Luis A. Cornejo]

[-- Attachment #1: Type: text/plain, Size: 2484 bytes --]

I don’t have my Starlink anymore, but I’ve just noticed that at least with
Speedtest screen shots, the bufferbloat seems to be much more under
control. E.g.

https://www.reddit.com/r/Starlink/s/THVeQ3r43D

Hopefully it’s not just an “optimization“ for the test. Maybe Nathan has
had a positive influence in that respect.

I still own the v2 hardware and would love to supports them instead of
Verizon, but if they had a more affordable 50/5 package for $50 which is
what Verizon charges for their 50/5 LTE service I would jump ship and run
some flent tests.

Anybody in this list able to confirm the better bufferbloat results?

-Luis

On Mon, Jan 15, 2024 at 9:22 AM Dave Taht via Starlink <
starlink@lists.bufferbloat.net> wrote:

> I would rather like starlink to get latency for
> voip/gaming/videoconferencing down to consistently (99.8%) below 40ms.
>
> from
> https://spaceflightnow.com/2024/01/13/live-coverage-spacex-to-launch-falcon-9-rocket-on-starlink-mission-from-cape-canaveral/
> …
> “The biggest, single goal for Starlink from a technical standpoint is
> to get the mean latency below 20 milliseconds,” Musk said.
>
> mean latency is a somewhat wrong goal. It misses the impact of slow
> start. I figure he is mostly describing the difficult layer 2 work it
> requires to accomplish that, everything from doing sat handoffs and
> rerouting stuff on the ground, to needing more ground stations, which
> is indeed a giant task.  I keep hoping that in particular they get
> away from the "recalculate connectivity and bandwidth every 15
> seconds", into something more responsive for normal traffic. DNS
> servers on the sats would help too! and does not fit into the concept
> of "mean latency" at all.
>
> ... and me, all I want to see is cake on the "glitchy", fq_codel on
> the wifi, and for them to find some way to leverage at least some of
> the techniques in LibreQos (like ebpf and cake) to somehow manage
> their downlinks better.
>
> It does look like they are now achieving sub 30ms latency to 1.1.1.1
> in some places for dns.
>
> https://home.sjh.at/starlinksmokeping/?target=DNS.CloudflareDNS1
>
>
> --
> 40 years of net history, a couple songs:
> https://www.youtube.com/watch?v=D9RGX6QFm5E
> Dave Täht CSO, LibreQos
> _______________________________________________
> Starlink mailing list
> Starlink@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/starlink
>

[-- Attachment #2: Type: text/html, Size: 3726 bytes --]