On 26/07/2007, at 12:40 PM, Philip D'Ath wrote:
Just thinking about all these Telecom peering points.
Assuming that Telecom allocate IP addresses as they are requested by incoming broadband connections, as opposed to by geographic peering point, then there will be no way to summarise addresses at these peering points.
Firstly, you mean "assign", not "allocate". Secondly, let's highlight, underline, and put in large type that you're assuming.
So this will imply that those ISP which do peer at these points will have to listen to a good 100k odd /32 announcements (at the larger peering points). I could easily see the routing table having to contain 200k to 300k additional /32 routes if you want to allow least cost routing to work its best. If there are redundant paths then this could increase a lot more.
Then consider the case when a /32 appears to be from Dunedin, and then the next second it appears to be from Wellington (because the Dunedin broadband user lost their session, and a new Wellington use just acquired their dynamic IP address). Then consider all the places these routes could be redistributed to within different ISPs, and potentially outside of their own AS. Then consider [non telecom] peering points that only accept a /24 or bigger. Not that anything is wrong with asymmetric routing, but doing it almost to the customer edge is going to make solving connectivity issues much more difficult (I can access this web site, but not that web site ...). I can see a lot of potential for route flapping, and for frequent routing updates to occur. I can also see the potential for short term route loops occurring.
That would be terrible, huh? It's a good thing they have a network engineering group with a higher than zero competence, so that their /own/ routers don't melt down, let alone routers in other networks that they peer with. -- Nathan Ward