[olug] internet partition

[Neal Rauhauser]

 I see there has been some talk about the internet partition due to the 
Level3/Cogent cat fight and I thought I'd take a moment to clarify some 
things.

 Internet providers are typically 'multihomed' - they get service from 
two or more carriers, they establish BGP peering sessions with the 
carrier's routers, and they receive at least one address block from one 
of the carriers that is a /24 or larger - a block with a mask of 
255.255.255.0 containing 256 addresses. This block is 'announced' via 
the BGP protocol and the world learns it can reach the block via 
whichever of the carriers is 'closer' to them.

 This stuff can all be tweaked a dozen different ways - adjusting the 
advertised path length, filtering what you send and accept, etc - the 
behavior of an exterior gateway protocol allows for much more policy 
expression than the knobs you find on an interior protocol like OSPF.

 Level 3 and Cogent recently stopped peering directly with each other. 
Circuits were taken down on one side over some business issue and that 
is that. What is unfunny about this is that one of the two providers, 
apparently Level 3, has dropped all routes to Cogent as well as their 
peering session. You can get from Cogent to Level 3 via several other 
peering arrangements, but Level 3 is playing hardball for some reason 
and filtering all routes to Cogent. This isn't a technical problem - 
orders came down from on high to whack Cogent and L3's peering guys have 
done just that.

  If I were a customer of either company I'd punish them over this sort 
of conduct. I'd stop paying the bill, I'd call my rep twice a day to 
reassure him that I wouldn't be renewing, and I'd contact one of my 
other carriers right away to get new IP addresses if I were using a 
block from the offending carrier.

  Besides doing a lot of BGP work around town I also find myself 
'multihoming' small offices as well. It is my studied opinion that this 
is a gigantic nuisance. If you don't share an interior gateway protocol 
from one of your two providers (just you try and order that!) you're 
pretty much talking application level awareness of gateway failure. In 
English - write a cron job that checks your gateways periodically and 
use the one that is up.

   This is another talk I could give at an OLUG meeting if people were 
inclined to listen ...