in the paper by gao, griffin, and rexford, it is shown that backup transit low-pref relationships as done today can lead to routing loops. it's worth a look just for that. while the authors propose a protocol approach, i wonder if there is an operational approach. randy --- From: Jennifer Rexford <jrex@research.att.com> To: randy@research.att.com Cc: griffin@research.att.com Subject: backup relationships Date: Wed, 6 Jun 2001 15:24:31 -0400 (EDT) Randy, Following up on our conversation this afternoon about backup routes and convergence problems, see Lixin Gao, Tim Griffin, and Jennifer Rexford, "Inherently safe backup routing with BGP," Proc. IEEE INFOCOM, April 2001. http://www.research.att.com/~jrex/papers/infocom01.ps and the slides at http://www.research.att.com/~jrex/talks/infocom01.ps http://www.research.att.com/~jrex/talks/infocom01.ppt The slides give a relatively theory-free overview. The intro of the paper presents the problem and the key results of the paper, and the body of the paper has some examples worked out and a fair amount of Greek to prove that our proposals handle the problem. The paper builds on earlier work by Tim on the stable paths problem (INFOCOM'00) and the non-convergence of some BGP configurations (SIGCOMM'99), and by Lixin and me on showing how policies based on peer-peer and customer-provider relationships can prevent these problems (SIGMETRICS'00). Backup routes throw a wrench into the work that Lixin and I did; this INFOCOM'01 paper proposes a way to deal with that problem. See the section on "global sanity of BGP routing" at http://www.research.att.com/~griffin/interdomain.html#divergepapers for the papers. -- Jen