[Starlink] Starlink Digest, Vol 25, Issue 28

Thu Apr 20 20:03:30 EDT 2023

The newest use case is waiting for a response from AI chatbots. We're
already conditioned to not have this be immediate.

Attributes of this use case are small volumes both up and down but compute
intensity at the datacenter.

-----Original Message-----
From: Starlink <starlink-bounces at lists.bufferbloat.net> On Behalf Of David
Lang via Starlink
Sent: Thursday, April 20, 2023 6:12 PM
To: Eugene Chang <eugene.chang at alum.mit.edu>
Cc: starlink at lists.bufferbloat.net; David Fernández <davidfdzp at gmail.com>
Subject: Re: [Starlink] Starlink Digest, Vol 25, Issue 28

On Thu, 20 Apr 2023, Eugene Chang via Starlink wrote:

> Friends,
> As I follow the discussion of putting computers on satellite, I can
understand the attraction if I apply the discussion with a satellite and
server above my location, suspended by a skyhook. The geometry is very easy
(some variations of a triangle). However, with an LEO satellite (or MEO),
most of the time the server is not overhead, it is hidden by the horizon.
>
> Have I missed comments (or naively not understood comments) about how the
solutions work when the server is not overhead? I wanted to hear about data
locality and how the desired behavior varies according to the position of
the server. Does some of the proposed edge computing imply (or assume) the
data is needed on many satellites so that there is always a server overhead
with the needed data? (Then we have lots of data synchronization
challenges.) Clearly, this suggests there is a scaling problem for edge
computing solutions because for a single server, most of the time the
computer is not at the edge near me.
>
> What am I missing?

1. data like DNS where it really is the same everywhere and changes
relatively slowly so it works well in this environment. (streaming data
could possibly fall in this category, depending on how popular something is)

2. compute loads where you don't get the answer back immediately and can
wait until the next orbit to get the answer (or where the processing time is
large enough that the latency of sending the results around the world when
it's done are small compared to the time it takes to generate the response)
This is not what people think of, as it's not people waiting for the answer
from a browser, but there is a lot more number-crunching than you think.

But yes, there are large categories of servers that this won't work for.

David Lang

> Gene
> -----------------------------------
> Eugene Chang
> eugene.chang at alum.mit.edu
> +1-781-799-0233 (in Honolulu)
>
>
>
>
>
>> On Apr 20, 2023, at 2:24 AM, David Fernández via Starlink
<starlink at lists.bufferbloat.net> wrote:
>>
>> Well, O3b MPower (MEO satellites) is offering independent one hop 
>> dedicated access to the (Microsoft Azure) cloud as "killer 
>> application". If the cloud is on the satellite, half-hop.
>>
>> Starlink GWs are near Google Cloud datacenters.
>>
>> Blue Origin is on the mission to move Amazon Cloud to orbit, 
>> eventually, maybe, leaving the Earth as a garden to enjoy, without 
>> any industry on the surface (in a century, maybe). Kuiper will offer 
>> one hop access to Amazon Cloud, then half-hop.
>>
>> What seems a crazy idea today will be eventually implemented later, 
>> like Starlink (Teledesic failed, fingers crossed Starlink does not go 
>> bankrupt, although I would expect it be saved by Department of 
>> Defense, as Iridium was saved).
>>
>> As we were discussing recently, maybe starting with anycast DNS 
>> servers on satellites is a first step to consider, before embarking 
>> any other type of cloud servers.
>>
>> Regards,
>>
>> David
>>
>>> Date: Thu, 20 Apr 2023 04:33:00 +0000
>>> From: Ulrich Speidel <u.speidel at auckland.ac.nz>
>>> To: "tom at evslin.com" <tom at evslin.com>, 'Michael Richardson'
>>> 	<mcr at sandelman.ca>, 'starlink' <starlink at lists.bufferbloat.net>,
>>> 	"e-impact at ietf.org" <e-impact at ietf.org>
>>> Subject: Re: [Starlink] DataCenters in Space (was Re:  fiber IXPs in
>>> 	space)
>>> Message-ID:
>>> 	
>>> <SY4PR01MB697983BB5DEB1B2AA0B2690BCE639 at SY4PR01MB6979.ausprd01.prod.
>>> outlook.com>
>>>
>>> Content-Type: text/plain; charset="windows-1252"
>>>
>>> Where do I even start? The lack of substantial bandwidth between 
>>> space and ground? The extra latency between ground and space 
>>> compared to terrestrial cloud, especially as terrestrial cloud edge 
>>> can move much closer to customers when space can't? The fact that 
>>> every LEO satellite is both a few
>>> 100 km from every customer and out of the customer's range depending 
>>> on when you look? That low temperatures in space don't mean 
>>> superconductive chips that produce zero heat, and that that heat is 
>>> difficult to get rid of in space? That generating power in space is 
>>> orders of magnitude more expensive than on the ground?
>>>
>>> Just because Starlink can provide a service somewhere between DSL 
>>> and low to medium grade fibre to a few million around the globe it's 
>>> not "done". Even with 10x the number of satellites and a couple of 
>>> times the current capacity per satellite, Starlink isn't going to 
>>> supply more than a couple of 100 million at best, and that's not 
>>> even accounting for growth in demand from IOT...
>>>
>>> --
>>>
>>> ****************************************************************
>>> Dr. Ulrich Speidel
>>>
>>> School of Computer Science
>>>
>>> Room 303S.594 (City Campus)
>>> Ph: (+64-9)-373-7599 ext. 85282
>>>
>>> The University of Auckland
>>> ulrich at cs.auckland.ac.nz<mailto:ulrich at cs.auckland.ac.nz>
>>> http://www.cs.auckland.ac.nz/~ulrich/
>>> ****************************************************************
>> _______________________________________________
>> Starlink mailing list
>> Starlink at lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/starlink
>
>