Re: DVB & all (Re: ipsat BoF)

[Harri Hakulinen NRC/Tre <Harri.Hakulinen@research.nokia.com>]

It may be, that I am "lonely snake" here, but I am in
favour of "topologically correct reverse tunnels" when
receivers are (multihomed) hosts or routers for small
domains.

>Joel Halpern wrote:
> TCP Relay and related issues fall into the TCP Sat working group, where
> there are many debates on the right way to do this.  (My personal
> prejudices run end-to-end, but there are alteratives such as relays,
> spoofs, etc.)

I guess that it is much easier to "control" e.g. TCP connections, 
if reverse tunnel exists, or build many kinds of proxy systems
that are optimised for e.g. satellite access users. 

>Joel Halpern wrote:
> As for the routing mechanism, I believe that the current UDLR mechanisms
> would give the best approximation consistent with internet routing
> philsophy. 

Hum, 
is there something like "Internet routing philsophy for Dummies"
available ?

I have couple of books here, but they have several hundred pages each
and I have found that it is difficult to learn this by ourself.  

And then some comments to original mail:
> On Tue, 4 Aug 1998, nghia PHAM wrote:
> > We have the following satellite topology with our Skyplex onboard
> > multiplexing:
> > A number of udl feeds transmit their packets to
> > the same satellite transponder which, with onboard processing,
> > concentrates the uplink streams into a single downlink stream.

In your (DVB) case I would like to use "virtual channels" for
those feeds, ie. use different PID for MPEGTS sections that are
send by different feeds (almoust like vpi/vci and AAL5).  

So, effectively those feeds are separated, and if receiver want's
to use one specifig feed, he creates reverse tunnel to the feed
that seems to be good for some or the other reason.

> On Tue, 4 Aug 1998, nghia PHAM wrote:
> > I would like to see a routing mechanism which allows a udl receiver to
> > specify a route so that the last hop to it must transit through any of 
> > a given set of udl feeds (dynamically detected by the receiver). 

Well, I assume that receivers IP address should eventually come
from "feed subnet", which means that traffic automatically comes
back via selected feed. If you want this to happen for e.g. each
TCP connection separately, I don't have any other answer than
building several virtual interfaces to receiver (one for each feed) 
and selecting one of them when connection is made (bind system call). 

But I am not sure if this is "official" interpretation
how UDRL could be used as part of solution ?   

> On Tue, 4 Aug 1998, nghia PHAM wrote:
> > The selected udl feed should preferably result in the best throughput,
> > heuristically, the closest to the server being queried.

If client can somehow get some info about location of feed
and server, it may act like in previous chapter, but I am
not sure how easily this can be implemented in general case.

In restricted cases it might be rather easy to implement
different favours of http-redirect like solutions in server
side.

> On Tue, 4 Aug 1998, nghia PHAM wrote:
> > I do not know whether this is complex or trivial but if there is a BoF
> > in Chicago as requested by Walid, I can make a presentation
> > of our Skyplex system.

I have also seen nice description on your website, 
is it still there (PDF, if I remember correctly) ?

> > Nghia PHAM      npham@eutelsat.fr , nghia_pham@compuserve.com
> > Digital Systems Division
> > EUTELSAT - European Telecommunication Satellites Organisation
> > (www.eutelsat.org)
> 
> Yours,
> Joel M. Halpern                         jhalpern@newbridge.com
> Newbridge Networks Inc.

--
Harri.Hakulinen              
@research.nokia.com            Play the Snake 
Visual Communications          http://www.nokia.com/snake/
Tampere/Finland