[e2e] TCP in outer space

[Cannara]

David,

Since some of these points are covered in other emails, I'll just say, as you
said yourself, that capacity is ever cheapening and so there's less reason now
than before to prop up the interoperability god on trick crutches, like
RED-to-TCP burps.  Interoperability is great and many companies, for years,
interoperated with others via gateways (in the true sense of the word),
whether TCP/IP-SNA, IPX-SNA, IPX-TCP, yadda yadda.  A most memorable
experience was walking into a large corporation years ago to assess a network
problem, only to see traffic with SMB on DECnet (PathWorks) on IPX tunnelled
in IP.  Talk about interoperability.  Try it the other way -- UDP on IPX, for
instance.  :]  

One issue is that the TCP/IP family is a close-knit one, with interlayer
knowledge and dependencies that, if anything, make interoperability at end
stations an issue, forgetting such global interoperability issues such as IP
Multicast in real-world nets -- aw 28 bits is good enough for anyone.  

No one suggests other protocol families solved the congestion issue for
wide-open-access networking, and TCP/IP demonstrates that as well.  One could
easily add the queue-drop 'control' trick to any network whose traffic is
mostly via the transport and whose transport backs off, as say do DecNet,
XNS...

Alex


"David P. Reed" wrote:
> 
> At 09:31 PM 4/12/01 -0700, Cannara wrote:
> >Indications of parochialism were clear years ago, when Xerox had a wordwide,
> >functioning, XNS network, followed by 3Com's worldwide XNS-like network,
> >paralleled by the Marine Corps worldwide Banyan Vines network, various Netware
> >networks, etc., etc.  These were indeed largely datagram networks.
> 
> You forgot Bitnet and UUCP... and of course WWMCCS and the wonderful
> roaming capabilities that work so well in interconnecting cell phones with
> the POTS network.
> 
> And I suppose they all interoperated, so we could have a WWW?
> 
> I know the Vines and Netware architectures in detail, and they did not have
> any particularly good congestion control at all (who needs it when you are
> primarily focused on fast LANs).
> 
> You want to balkanize, why not just connect every system to every other one
> with a dedicated line, and use private protocols.  Who cares about
> innovation or interoperation, just as long as the links are fully occupied
> by bits?
> 
> The point that of the goals of the Internet, interoperation is number one,
> and heterogeneity is number two, seems to be beyond some people's
> comprehension.
> 
> >- David