Wouldn't one of the ISPs have to advertise a longer prefix? I would think that the address space would come from only one of the providers, in which case the other provider would have to advertise this space on top of its own /20. It is irrelevant whether the two ISPs advertise one another, the longer prefix would be the first choice for the backbone traffic. If the longer prefix route goes down, traffic would still go to the /20 the other provider is advertising. The ISP who is advertising the route on top of its own /20 can't aggregate said route as it only can route to that portion of the address space defined in the longer prefix. Geoff Zinderdine ----- Original Message ----- From: "David Lott" <dlott@msncomm.com> To: <nanog@merit.edu> Sent: Thursday, September 14, 2000 12:34 PM Subject: Confussion over multi-homing
I've read the current policy on ARIN's allocation of space and I must admit that I'm still confused.
First, allow me to state the assumptions that I'm under. I understand the policy to state that if a business needs to multi-home and requires less space than a /20, then they should request this space from their ISP. I also understand that there are filters at the /20 boundaries in order to minimize the size of the routing table.
Question: Doesn't this break multi-homing for end users that need less than a /20?
For example, assume that the end user is connected to two regional ISPs (ISP-A and ISP-B). Neither of which have agreements with each other. However, they do share a common backbone with a national provider we will call ISP-Z. If ISP-Z has filters at /20 for both of the ISPs that it is connected to, then ISP-A address space will be the only space listened to on the ISP-A to ISP-Z link. The same would be true for the ISP-B address space only being listed on the ISP-B to ISP-Z link.
This creates a situation where address space from ISP-B would not be advertised through ISP-A and in effect, breaks multi-homing. Consider a
remote site attempting to reach the web server at the end user. DNS resolves the address to ISP-B address space. Also assume that the link between the end user and ISP-B is down. As the packet enters the national carrier ISP-Z's network, at some point the router will have to decide to send the pack on. If ISP-B is still advertising the remaining
portion of their network (say at the /20 boundary) then ISP-Z will forward the packet to ISP-B. This is normal and proper for a single homed address space. However, if the end user had their own micro allocation, their address space would be advertised to both ISP-A and to
ISP-B and in turn to the national carrier. As such, the destination network route would be dropped from the advertisement coming out of ISP-B and the only remaining route would be via ISP-A and the packet would still get there - if the end user had a micro allocation as per previous policy.
Also, let us further look a situation where ISP-B is down. When the national carrier detects ISP-B is down it will remove that particular route from it's table. In the old way of doing things with micro allocation to multi-homed end users, ISP-A would advertise the address space from the end user. It is my understanding that under the current policy, ISP-A would have to advertise the address space allocated to the
end user from ISP-B. If the address is less than a /20 and if the national carrier is filtering on a /20, wouldn't that cause the update to be dropped and thus not added to the routing table for the national carrier?
I guess my confusion could be cleared up if someone could describe how, under the /20 policy, an end user requiring multi-homing and less than a
/20 allocation would be able to survive one of their two ISPs going down
(remember the AT&T and MCI outages?).
Thanks,
PS. I have sent this email to the ARIN policy list but have not received any useful responses.
-- David Lott VP of Operations MSN Communications (303) 347-8303