At 05:50 PM 12/19/97 +0000, Sam Critchley wrote:

>

>There must be some really interesting stuff to deal with in there - is

>there anyone from Teledesic who'd be willing to share a little with 
us?

>I'd imagine they'd have to add all sorts of metrics into the decision,

>such as power differences between sending to a nearby satellite and

>sending to one that's far away, being able to route around busy parts
of

>the sky, all on top of a constantly changing table based on neighbours

>that are moving in relation to each other. Others?

>


Sam-


Speaking from what Mark Allman calls the "dark side" (ATM), these issues
are indeed really interesting.  Big buffers-- the facile and
oft-mentioned solution-- are fine so long as you're dealing with PVCs
provisioned at PCR, making everything more or less nice and
deterministic.  But as we all know it falls apart when you try to provide
stringent QoS guarantees for a full-blown ATM service with SVCs, random
holding times, real-world congestion, node failures, and all that good
stochastic stuff.


We were part of a (post-FCC filing) contract study team in '96 analysing
(and trying to solve) some of these issues for a satellite data network
which was, at the time, targeting ATM.  CDV was my own focus.  The
results of the study were fascinating and I suspect some of the analyses
would pertain to the TCP/IP QoS area as well.  Unfortunately they're the
property of the client and considered highly proprietary.


FYI, Bezalel Gavish (Vanderbilt U.) and colleagues have written a couple
of excellent papers touching on some of these issues ("LEO/MEO Systems"
and "The Impact of Intersatellite Communication Links on LEOS
Performance"), which will appear in the journal <italic>Telecommunication
Systems Special Issue on Satellite Communications</italic>, coming out
Real Soon Now (next couple of months-- I'll post a notice when it's
available).  Several other papers in the issue will also be of interest,
I think.  Gavish's email ddress is gavishb@ctrvax.vanderbilt.edu.


I suspect that-- just like switch internals-- the internal satellite
routing protocols (for ATM and TCP/IP) will remain as proprietary as
possible.  After all, end-to-end performance is really going to be the
tiebreaker between the various LEO systems as far as the subscribers and
investors are concerned, so why give the competition a leg up?


R/

Eric