[e2e] the evolution of deployability

Cannara cannara at attglobal.net
Tue Dec 17 21:32:08 PST 2002

Aaaahh George, it's always fun to see reactions to simple critiques.  In order
to minimize this thread, I'll add your 2nd comment here, obvious as it may

"But if you get 'good-enough' deployed then the sheer weight of inertia is
against you 'rolling the stone' any further up hill. But on the bright side,
at least the Valley is a long way below you..."

Those of us in the chip industry for some decades know why I's architecture is
mediocre, but this isn't a chip/prcessor forum, so let's skip on (we don't
need to even bring up M :).
TCP/IP was never designed to address the Internet's content, because its
designers didn't foresee even a portion of its expansive success.  Nor did
they even attempt any form of security as simple as a registry of unique,
allowed names/addresses, which might prevent the many silly security breaches
we've spent millions on trying to plug.  Since security and end-end
accessiblity is certainly what layers 5, 6, 7, 8, 9, whatever, depend on as
root networking functions, the Internet protocol suite has been a massive
failure, whether measured in $/code-byte or $ spent on trying to recover a
semblance of engineered security and reliability, even from important apps,
like SMTP ("HELLO" or "HELO" :).  

The Inet's success is one of marketing -- the idea that every Unix box
from any manufacturer came with it free was genius -- totally noncompetitive,
much as M functions.  Maybe you're right -- Decnet, XNS,
Netware, Vines, etc. should have been done that way?!  In any case, we have
what we have and like any bureaucracy, the Inet's current control is out of
the hands of those interested in substantive innovation -- NAT, MPLS, IPSEC,
IPv6, etc. demonstrate this.  NAT, at least, serves immediate needs of end
customers trying to just plain use the Internet in some way that helps them --
your "good-enough" conjecture.

Yes, the belief that "the Valley is a long way below" may die some day, but I
ain't counting the hours.  And, one can always fall back on the lazy canard of
unattainable perfection.  What bureaucracy doesn't try this one?  I still,
though, can hope that young folks will listen to, and study, the whole history
of networking and ask harder questions in forums, theses and committees than
the Inet bureaucracy has become used to.


PS For years, LAT was very successful as a private-WAN, terminal protocol,
just ask Perrier!  Of course the LAN-designed timers doubled the packets, but
LAT's well-designed robustness cared not -- you never got double charged for a
bottle and management didn't have to pay terminal operators overtime. {:o]

George Michaelson wrote:
> You know, DECnet probably looked better in theory than in practice.
> I doubt if any of the architects of LAT, who decided to make it so dependant
> on local-network timings were proud of that decision in hindsight. X25
> architects are probably still proud of the niche they filled.
> Maybe IPv6 is not as good as it looked on paper. But judged against its PEERS
> maybe its still worth it? Its about in-context decisions, not about what-if.
> Do we have to question *everything* *all the time* ? How far does this go? do
> we re-visit TUBA and SIPP?
> Designing protocols which fail at layer 8 or 9 is just as much designing
> failed protocols. Engineering isn't Science, its closer to Art precisely
> because it admits to the failure of the human experience, and mediates from
> the theoretical perfection to a practical, achievable reality.
> -George

More information about the end2end-interest mailing list