[e2e] TCP in outer space

[David P. Reed]

I hope this isn't boring the vast majority of e2e list participants.  (send 
complaints to me, not to the list!)

But now I'm really intrigued.  Here's why:  a "space Internet" is (should 
be?) motivated by the same rationale as the original Internet - a network 
of heterogeneous networks can incorporate technological and applications 
innovation much more effectively than dedicated, purpose-built Internet.

TDRS is a dedicated, highly "optimized" technology (at a point in time, for 
a class of applications).  Sounds wonderfully well thought out. But... Like 
all "optimized" technologies, it is brittle with respect to 
change.  Doesn't tolerate new infrastructures or new applications very well.

IP is, as we all know, "good enough" technology, optimized for nothing in 
particular, except for its ability to rapidly incorporate or support new 
stuff at the low levels and new apps and requirements at the high levels.

Now that there's enough going on in space, I'd say the time is NOW to start 
architecting a "space Internet".  Remember, the Internet is not the 
hardware, and never was.  You can run "space Internet" over TDRS and 
anything else out there.

I bet there is a cadre of young designers who can carry this torch, just as 
the young students and others carried the Internet torch.  It is clearly 
"enabling" technology for space exploration, and even more than ISS, it can 
enable all kinds of stuff to happen.  But the key has to be a net that 
follows design principles like the Internet's.  End-to-end, liberal on 
inputs, conservative on outputs, no specialization to underlying transport 
"features", no privileged "applications".

Comments below.

At 02:01 PM 4/10/01 -0700, Courtney, Bill wrote:
>One of the problems with using non-GEO satellites is that
>the topology of the constellation is changing. True,
>it changes in a predictable way, but the changes will,
>nonetheless, require network reconfiguration. With non-GEO
>satellites, not only might the inter-satellite links
>be breaking and making, but also the earth-space access
>links for each user will be continually changing.

Yup.  The fact that links are independent of technology is what makes it a 
"network".  In the Internet, we said, since topology is constantly being 
broken because of reliability and autonomy, we need a routing protocol that 
can incorporate change, and a transport protocol that doesn't bind 
end-to-end state into physical circuits.

>In addition, satellite coverage footprints in a
>constellation usually overlap, and for many
>constellations the overlaps are also in constant
>change. This means that uplink and downlink bandwidth
>allocations among the satellites will be subject to
>changing re-use constraints. This makes it difficult
>to ensure that users in any particular area on the
>ground will receive a constant amount of bandwidth. And
>fluctuating access bandwidth leads to all sorts of
>problems, as you can imagine.

Fluctuating access bandwidth is no more a problem than fluctuating 
application demand.  I suspect that fluctuating app demand is "hidden" in 
low average utilization of end-to-end fixed rate circuits.  To me, there 
are very few applications (voice is one) that really have "fixed rate" 
needs.  The rest are kludged by assigning a fixed rate circuit that can 
handle worst case traffic, then leaving most of the capacity idle.

It's hard to believe that computer-computer (and human-human mediated by 
com-puter) traffic is fixed rate.  So the matching of demand to capacity by 
buffering and dynamic allocation (e.g. datagrams and bursty VCs) is going 
to be an issue for space - why not admit it and design a network that does 
that.

>These and other difficulties must be balanced against the
>possibility that latency might be lowered with non-GEO
>constellations. (I say "might," because some of the
>non-GEO difficulties can be addressed at the cost
>adding latency.) Non-GEO can be a bear!!

It isn't just latency that might be lowered.  Like the Internet, by 
allowing the incorporation of a wide variety of underlying technologies, 
new applications can be introduced at low cost.  Every time we increase the 
level of resource sharing by creating a network to mediate access to a 
resource (i.e. space), we boost innovation by lowering the cost of entry.

It would seem to me that a "space Internet" would be as big a deal as the 
"fast, cheap, out-of-control" mars robot design philosophy is.

Of course, I hold no hope to sell this idea to the "Beltway Space Bandits" 
- the corporate biggies whose goal is to sell expensive engineering time to 
risk-averse gov'ts, rather than looking for ways to do stuff cheaply and 
innovatively.  But maybe there's an innovative scientist-type at NASA or JPL?


- David
--------------------------------------------
WWW Page: http://www.reed.com/dpr.html