Re: [TE-wg] TE use in today's networks

["Xipeng Xiao" <xiaoxipe@cse.msu.edu>]

Two authors of this much discussed paper:
http://infocom.ucsd.edu/papers/744.pdf
Yufei Wang and Leah Zhang, are my colleagues at Photuris.
We had a long discussion on the practical issues 
of the paper some time back.  Yufei told me that:
THE APPROACH PRESENTED IN THE PAPER IS INTENDED FOR
NETWORK PLANNING, NOT FOR REAL-TIME TRAFFIC ENGINEERING.

You may also find the discussion on the 
'irtf-rr@puck.nether.net' mailing list during 5/7/01-5/9/01
helpful. It clarified some common misunderstandings 
about the paper. See the following thread:
RE: interesting Infocom paper on traffic engineering via routing metrics

XiPeng

> > I suspect that one of the papers referenced is:
> >
> > http://infocom.ucsd.edu/papers/744.pdf
> 
> Two other papers (posted on the TE-WG list back in December 1999) include
> 
>   - A. Feldmann, A. Greenberg, C. Lund, N. Reingold, and J. Rexford,
>     "NetScope: Traffic Engineering for IP Networks," IEEE Network 
>     Magazine, March/April 2000.
>       http://www.research.att.com/~jrex/papers/ieeenet00.ps
> 
>   - B. Fortz and M. Thorup, "Internet Traffic Engineering by
>     Optimizing OSPF Weights," Proc. IEEE INFOCOM, March 2000.
>       http://www.ieee-infocom.org/2000/papers/165.ps
> 
> > There are serious considerations to be overcome before this is
> > practically implementable (ie, 10 significant digits on routing
> > metrics).
> 
> The techniques in the papers listed above work with integer weights.
> Experimental results seem to suggest that a small number of different
> weight values are often sufficient in practice.
> 
> > Additionally, there are subsequent ops problems with implementing it
> > (ie, dramatic impact when one link drops).
> 
> Depending on which link fails, the network load after the failure
> isn't all that bad.  Although some failures can cause problems, often
> one or two weight changes after the failure is enough to bring the
> network back to a happy place (analogous to the need to fail over to
> backup paths in MPLS).
> 
> > I suspect that this issue has been hammered to death on a mailing
> > list somewhere, anone have pointers?
> 
> There has been some discussion recently on the IRTF-RR list...
> Another reference listed there is
> 
>   http://www.eng.tau.ac.il/~shavitt/pub/DIMACS01-17.ps
> 
> -- Jen
>