[Starlink] CDNs in space!

[Ulrich Speidel]

On 1/09/2022 8:24 am, David P. Reed via Starlink wrote:
>
> Having looked into this a lot, CDNs don't account for very much 
> Internet traffic. There's a lot of marketing propaganda out there that 
> suggests this might be true, but when you try to track down the 
> source, it's almost always from an old slide deck presented by a 
> company that sells CDN servers.
>
Well, most web sites may not be using CDNs but almost all frequently 
used sites do. This ranges from the Netflixes and Vimeos of this world 
to Facebook, Twitter, pretty much all significant media organisations 
(even our granny paper, the NZ Herald, does), most larger universities 
(we use Panopto for our lecture recordings and other videos, and most of 
our peer institutions do the same, and they all use CDNs). Most web 
sites that I use on a daily basis are CDN-based. General consensus now 
is that currently, CDN traffic accounts for >70% of Internet traffic by 
volume and growing, and that's what matters here.

If you didn't have CDNs but origin servers only, traffic on 
international links would be much, much higher than it is now. Read: It 
would kill them.

> That doesn't mean that CDN servers don't help a fair bit, but no, 
> today's Internet wouldn't fall apart if we didn't have CDN servers 
> withint 2 msec. of the edge.
>
> Also, CDN's need to be BIG to hold all the videos that people might 
> choose to watch at any particular time.
>
They are big. But they don't need to hold all videos people might choose 
to watch. A CDN server only needs to hold each video that at least one 
user in that neck of the woods has already requested, and that only for 
a given amount of time. But here's an added point: What users want to 
see is highly geographically correlated. Terrestrial CDN servers only 
serve a particular region. A CDN server on a LEO would have to serve 
every region of the world.
>
> So I'm just pointing out that the business case for CDN's in space to 
> merely solve Starlink's potential issues is probably not great. Maybe 
> 10 years from now. Maybe never. The idea that everyone watches TV and 
> the same few seconds of content of a few shows that are 
> extraordinarily popular - well, that dog don't hunt. It doesn't 
> justify multicast either.
>
> So let's improve the discussion here. The Internet, for the forseeable 
> future, at the edge, is unicast.
>
Yes! The problem is that, for any direct-to-site satellite service, the 
edge is on the wrong side of the satellite link.
>
> CDN's reduce costs for big companies like Netflix, not because they 
> are "close to the watcher" but because it is easier to not have 
> everything centralized in a single point of failure and needing a huge 
> pipe because all customers aggregate demand into their central HQ.
>
> There are some advantages at the top level peering in distributing CDN 
> sites into the networks of various access providers (like Comcast) 
> because you can get cheaper pricing.
>
> But lets think about CDNs in space. They don't deal with the actual 
> bottleneck in each satellite to ground.
>
Exactly.
>
> The uplink from central ground stations has LOTS of capacity, it won't 
> be the bottleneck if Starlink or any other satellite service for that 
> matter balances its design.
>
> And if they get inter-sat links (laser or maybe RF, if the 
> interference problem with other satellites is resolved at the WRC or 
> other international body that regulates RF in space) working, where 
> are the CDNs going to maintain state? The reachable ones for any 
> satellite in the constellation will be moving relative to the user's 
> dishy, unless it the CDN is in geosync orbit).
>
> Starlink isn't a media company. It doesn't want to own all the 
> content, or even host all the content.
>
But it's got to transport all the content, each and every time somebody 
wants it. No international subsea fibre cable needs to do this.
>
> So I'm pretty darn skeptical!
>
> At the moment, Starlink is a "last mile" service in a virtual sense. 
> It connects from the public internet at some pretty high performance 
> point of presence, hopefully has a pretty good latency, and doesn't 
> let its users saturate any satellite (lest huge queueing delay build 
> up). Given that, anywhere there is a ground station the public fiber 
> backbone can reach all the CDNs.
>
The problem is though that in reality, Starlink is a severely 
bandwidth-constrained "last thousand mile" service. If the CDN server 
sits on the public fibre backbone, then every time it serves a piece of 
content for the 2nd or subsequent time, it takes up scarce capacity on 
the satellite. That isn't the case in (proper) terrestrial networks, 
where aggregate bandwidth increases the closer you get to the end 
customer. My suburb here has almost all houses on 1 Gb/s FTTH now, so 
collectively, we have well over 1 Tb/s on our doorsteps. Granted, the 
CDNs that serve us probably only have between 1 and 10 Gb/s output, but 
there is just no bottleneck between them and us.
>
> It's hard to capitalize a last mile fiber service. Each home passed 
> costs about $2000, if we want to connect up (that includes almost all 
> actual rural residences, but not all the uninhabited parts of the 
> planet). For whatever reason, the telcos want no one else to put in 
> fiber, but also don't see much profit in extending their coverage, 
> either, because they aren't allowed to charge for all the content like 
> they used to. Fiber's great advantage is that it has extremely low 
> operational expense.
>
$2000 is about a year's worth of Starlink service with Dishy, right? I 
think again this boils down to what is very much a US-specific problem.
>
> This is what Starlink fills in for. It has lower capex, but HUGE opex. 
> It can't perform or scale very well compared to the asymptotic fiber 
> solution. So it has a pretty good short-term competitive position. And 
> they've gotten very smart in finding early customers who are eager to 
> buy in.
>
Indeed.
>
> However, it's important to realize that what drives this all is how 
> the Internet is used and who needs it where, and how much they are 
> willing to pay.
>
> One thing is clear - Starlink isn't the Internet of the future. It's 
> filling a niche (a large one, but a niche).
>
Exactly.
>
> The mistake Motorola made with Iridium was in not realizing that 
> cellular telephony was in all respects a better answer, and would be 
> cheaper, too. People did talk about putting CDNs in space with 
> Iridium, too.
>
> But one needs to understand what CDNs are useful for, and really 
> understand what part of the problem that is.
>
>
> _______________________________________________
> Starlink mailing list
> Starlink at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/starlink
>
-- 
****************************************************************
Dr. Ulrich Speidel

School of Computer Science

Room 303S.594 (City Campus)

The University of Auckland
u.speidel at auckland.ac.nz  
http://www.cs.auckland.ac.nz/~ulrich/
****************************************************************


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/private/starlink/attachments/20220901/ef48762b/attachment-0001.html>