There's a paper in this year's Infocom that manages to prove MPLS is
unnecessary for traffic engineering. It shows that for *any* set of
LSPs, it is possible to compute a set of routing link metrics that
result in exactly the same traffic flow. This is slightly surprising
since for N nodes there are typically O(N^2) LSPs but only O(N) links so
there's a huge difference in information content between the two
approaches. But their development shows that the problem is so
conditioned by flow balance contraints (ie., all the flow that enters a
node must leave it) that O(N) parameters (the link metrics) are
sufficient to completely specify the solution.
If anyone's interested, the paper is available at:
http://infocom.ucsd.edu/papers/744-2798113708.pdf
- Van
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:03 EDT