On Mon, Jan 12, 2009 at 04:51:36PM -0500, Joe Abley wrote: [snip]
In my experience prepending someone else's AS to a prefix has only been useful operationally only as a short-term, emergency measure (e.g. when trying to avoid a black-hole between two remote ASes, neither of whom shows any signs of fixing the problem).
Randy's application, and Lorenzo's before him also seem like short- term applications designed to explore answering operational questions.
Nit, weird paths (this one) and long paths (Lorenzo's) are different. There were known BGP implementations which choked and died on long as-paths, which (w|c)ould trigger outages. Weird paths which appear to involve your network triggers -at least- work.
Just because something is generally not used, or even if it's only worth using in an emergency, doesn't make it "sketchy".
Given the prevalence of BGP community-based remote control over your direct neighbor's neighbors, it has seemed to (to me) to decrease. Using a label allocated to someone else does indeed seem sketchy to many of us; while the injector knows they are doing it and the injectee can figure it out, there's a heck of a lot of other parties (and archives) without context. Encouraging the use of such approaches, rather than encouraging providers to provision customers without the ability to forge AS paths, is a step in the wrong direction.
Most knee-jerk reactions to AS_PATH manipulation sound to me like fear of the unusual.
Less fear and more annoyance; the waters are muddied and the unusual requires investigation, and in some cases explanation internally & externally. Propagating bad table hygiene doesn't promote network use, increase stability/robustness, or anything that could be viewed as best practice. All IMO, of course. Cheers, Joe -- RSUC / GweepNet / Spunk / FnB / Usenix / SAGE