[2542] in Commercialization & Privatization of the Internet

home help back first fref pref prev next nref lref last post

Re: Trust and the T-3 Network - encore

daemon@ATHENA.MIT.EDU (Martin Schoffstall)
Sat Mar 14 23:26:43 1992

In-Reply-To: <9203141917.AA06883@tmc.edu>
Date: Sat, 14 Mar 92 22:24:58 -0400
To: "Stan Barber" <sob@tmc.edu>
Cc: "Gordon Cook" <cook@tmn.com>, com-priv@psi.com
From: "Martin Schoffstall" <schoff@mail.psi.net>
Reply-To: schoff@psi.com

>FROM:   Stan Barber <sob@tmc.edu>
>
>My belief is that Merit (or any network provider) has pretty much stuck to
>one vendor's routers to insure interoperability and ease of management. [The
>enterprise MIBs usually have the really interesting stuff in them like memory
>usage and CPU statistics.] In a truely open networking environment, routers
>supporting the standard protocols would all interoperate at all levels. In
>reality this is not always true.

Actually no.

PSINet is very heterogenous integrating 3 different router vendors 
hopefully seamlessly together.  I hope to decommit one router vendor in 
the next 12 months and replace all its gear with yet another router 
vendor, so that we still have 3 vendors.  This has been a purposeful 
committment by the PSI team from day one.

No vendor meets all the price/performance/reliability/management "point" reqs
that are needed, even independant of good engineering/business 
objectives of above.

It is even possible that a vendor who is "tops" in 92 will be "mediocre" 
in 94.

Note that we are far from unique (except maybe in "style") because
I know of several public internets that are also heterogenous.

One final point.  While it a more complex network operator issue, I think 
it is important that even the SOFTWARE be different between the 
different vendors.  The router technology (and even FR technology)
is almost completely software, this heterogeniality
becomes even more critical due to
underlying technical structure and software development/engineering 
capabilities -- which can limit the ability for the router to "scale" 
through time.

What was a neat hack in 85 might not be good for the Internet in 92, it 
should be obvious to many in the audience that we have a clear "vendor" 
scaling problem in the area of software.  But I'm sure a "blue ribbon 
panel" can be whipped up at the drop of a gavel to "prove" me wrong.

Marty

home help back first fref pref prev next nref lref last post