|
North American Network Operators Group Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical RE: Interesting social phenomenon...
>As you should well know, it's fairly difficult to break the internet as a >whole. Obviously one can break sections of it, and the "break" may >appear differently from one site to another. Therefore, although some >people may have had sever outages, some of us only lost sections >of the routes to AS 7007 and our email between several of us worked. That assumes a fairly rich mesh of alternate paths between providers. Unfortunately it fails when you limit you choices to only a few connections between a few providers. Instead of being able to lop off a small misbehaving section of the network, you are forced in to lobotomizing large chunks of your connectivity. In this case, if you only connected with Sprint, you lost connectivity to everything 'behind' Sprint. 'Behind' being relative to which side of Sprint you were connected. On the other hand, if you additional possible paths, you may have been able to maintain connectivity to more of the Internet even if you needed to shut Sprint's connections off for a time. It may be a difference between a 60% outage, and a 30% outage. All those little extra paths add up to better connectivity. And every path you don't use, is a piece of connectivity you lose. Sprint is just the current example, the same thing is true for maintaining connectivity to anyone you only see from 'one' place/provider/etc. InternetMCI had a filtering oops last year, and I'm certain other providers will have them in the future. Just because you have a good connection today, doesn't mean it will always be there. I think I'm going to dust off my draft on Guidelines for Responsible Internet Service Providers. So far I've received three comments, if you would like to review it one more time, it is available at <http://dranet.dra.com/draft-donelan-rnmg-latest.txt>. I plan on updating the section on out-of-band verification, and submit it as an informational RFC. -- Sean Donelan, Data Research Associates, Inc, St. Louis, MO Affiliation given for identification not representation - - - - - - - - - - - - - - - - -
|