On 15 May 2002 at 16:32, Fredrik Haborn wrote:
> I might missunderstood things at the meeting but what I understand
> several operators are running "hot potato" at the ix points.
> And what would the benefit be with only regional routing ?
I'm not saying the existing routing announcements should be replaced. The idea is to add
regional routing information to gain experience if what works in theory also work in practice.
The purpose is not to convert any of the national peering points (NPs) into a regional point,
but to use it to "simulate" a regional peering point. Why? Because it's there. People are
connected, things are up and running. Much easier than getting a switch, talking people into
ordering connections to the switch, setting up those connections and then doing testing.
I'd like to see a number of ISPs exchange lesser aggregated routing over one of the NPs
according to the specs. in the regional routing document.
That is, the routing announced being from their NP attached confederation/RR topology, and the
information received from peers kept within that same confederation/RR topology.
I don't think we would see any traffic-benefits of doing it, but what we might learn is
1. How much problem is it to configure one's network to announce relevant regional routes.
2. How much problem is it to configure filters to prevent received regional routes from
entering the backbone routing tables.
3. How active will the routing tables be (i.e. how frequent are updates for smaller prefixes
4. What are appropriate dampening parameters to ease the load on the boxes.
5. What - if any - operational issues may arise when something like this is run for some time.
Perhaps a smaller test with two or three providers who feel confident enough in their BGP
skills to mess around with a production environment. ;-)
I know of atleast two ISPs that have (almost) everything they need in terms of community
controlled routing to be able to get things working in..say..15-20 minutes or so. ;-)
PacketFront Sweden AB