[28742] in North American Network Operators' Group
Re: Optical Crossconnects and IP
daemon@ATHENA.MIT.EDU (Vijay Gill)
Sun May 14 22:30:34 2000
Date: Sun, 14 May 2000 22:28:20 -0400 (EDT)
From: Vijay Gill <wrath@cs.umbc.edu>
To: Michael Shields <shields@msrl.com>
Cc: nanog@nanog.org
In-Reply-To: <874s80efa5.fsf@challah.msrl.com>
Message-ID: <Pine.SOL.3.95.1000514221949.13512A-100000@mailserver-ng.cs.umbc.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Errors-To: owner-nanog-outgoing@merit.edu
On 15 May 2000, Michael Shields wrote:
> The approach laid out by Fortz and Thorup is very pleasant in its
> ability to show that a simple weight-based model can come close to
> optimal traffic engineering. But without any way to quantify how
> accurate our estimate of the demand matrix is, we cannot know whether
> the projected weights are actually even close to optimal. And
> considering the elasticity of demand, I would argue that an accurate
> demand matrix cannot be constructed for most Internet backbones given
> currently available tools and understanding.
Most. Not all. At some promising local ISP's in the Fairfax, VA, area,
there were actually fairly good demand matrix data available.
The macroflows due to the architecture employed were actually fairly
tractable and if one were to use ordinal (good enough) vs cardinal (best
possible) optimization techniques[1], resulted in fairly usable
parameters. The said parameters were useful for plugging into the traffic
engineering tools du jour (SPVC contracts at the time).
These data were also used to great effect in doing surviability analysis
and circuit planning.
[1] Overview of Ordinal Optimization, Dr. Yu-Chi Ho. Proceedings of the
33rd Conference on Decision and Control.
/vijay
