Traffic engineering tools

Thu Oct 28 19:41:43 UTC 1999

Prabhu --

> You continue to dismiss concepts (traffic engineering) that have proven
> themselves in practice in large networks.  And they continued to prove
> themselves long after you predicted they would not (years ago).

I continue to tell that there are simpler and more robust solutions to
the same problems.  And that it is much easier is not to create problems
in the first place than to think up clever patches later.

> Stating the obvious, "there's more fiber in the ground than all routers
> can fill up" is not the point.  The point is could we have built the Internet
> to its present point without traffic engineering?  The answer to that is a
> resounding NO!

If you consider Internet in its present shape working, i beg to differ.
It's a horrible kludge which nobody really understands.  It's extremely
brittle.

For one, i'm working for a company which takes the question "How can you
put 911 service on the Internet" very seriously.  If you look at the
things from this perspective you'll see that we have a long way to go,
and that making things even slightly more complicated than the bare
necessity is not the right way to go - if we want to have it debugged
in less than 100 years it took for telephony to become dependable.

> The next question is whether traffic engineering is useful for the future.

I see little use in discussing the present :)

> Remember these practitioners do not build networks based upon theory, but on
> practical implementations that prove themselves to work.  Perhaps the
> load-sharing mechanism you described for Pluris will work.  Perhaps not.

Works in Cisco's CEF.

> But to promote what hasn't proven itself and dismiss what has sounds
> suspicously like bad marketing dressed up in the language of mathematics.

I do not do marketing.  I do not speak for Pluris.  If you checked the
website you'll see that the company is doing all the TE bells and whistles
you find in any other router box.  In fact, if i behaved like a marketroid,
i'd insist on doing them - because not providing something which is in
vogue never wins a lot of customers.

As for mathematics and logic reasoning - please, if you see flaws in the
reasoning, pount them out.  The fact that something sometimes works doesn't
prove anything.  Following that line of argument, we all should be doing X.25 -
because some decade ago that was what nearly everyone was doing and the
networks in their present shape at the time couldn't be built without it.

--vadim