[28647] in North American Network Operators' Group
Re: Optical Crossconnects and IP
daemon@ATHENA.MIT.EDU (Vadim Antonov)
Thu May 11 06:03:07 2000
Date: Thu, 11 May 2000 03:00:10 -0700
From: Vadim Antonov <avg@kotovnik.com>
Message-Id: <200005111000.DAA24358@kitty.kotovnik.com>
To: akyol@pluris.com, tli@procket.com
Cc: nanog@nanog.org
Errors-To: owner-nanog-outgoing@merit.edu
I would like to inject some sanity into the traffic engineering
discussion by reminding about some small fact of life:
Nobody knows how to measure how what the demanded/available bandwidth
ratio is when the circuit is overloaded.
In a steady more, utilization is s 100%; packet loss is not an accurate
gauge, by any means (it depends on delays, agressiveness os sources, queue
manageent policies, timing of source starts, and phase of the moon).
What it means that by and large traffic engineering is done by seat of
the pants. An attempt to create a close control loop with that kind of
source information is somewhat problematic. And so is "wavelength on
demand".
When circuits aren't overloaded, traffic engineering yields zero benefit.
--vadim
PS I just love engineers spending untold hours and companies spending billions
on a technology of questionable worth and little theoretical foundations,
instead of spending a small fraction of it on research first.
PPS Pluris core technology allows to treat traffic as a liquid - unlike simple
IP routing it works by splitting aggregated streams, which can be sent
along different paths. It makes any VC-based traffic engineering and
associated hardware overhead simply unnecessary. I'm more than a little
disappointed by the fact that this is apparently lost in the mind of the
company's current architects. (Admittedly, it still requires some
work on algorithms, but at least routing of colored liquids is much easier
computationally than routing of VCs).
