Sean writes:

The issue Andrew and Tony were raising was a bit more simple and a bit more complicated on several fronts. Let's start with a picture:

MAE-WEST PBNAP MAE-EAST

39.0/10 39.64/10 39.128/10

A------| S----| A--| S------| M----| S--| M------| N9---| M--| N9-----| NE---| N9-| NE-----| Y----| NE-| X------| Z--|

In normal conditions:

How does X get traffic to Y? How does Y get traffic to X? How does X get traffic to Z? How does Z get traffic to X? How does Y get traffic to Z? How does Z get traffic to Y?

How does A get traffic to Y? How does Y get traffic to A?

How do they do it now? Without using those aggregated addresses, we have this same situation in play today with some midsize ISPs directly connected to one or a few interconnects. Usually, they buy transit to other interconnects etc. from one or more providers and peer with everyone else who's at the interconnects they are at (as a simplistic explanation).

Where can 39.0/9 be aggregated? Where can 39/8 be aggregated?

You broke my critical assumption, which is that we're going to draw the "39/8" boundary around northern california. A better diagram would be more along the lines of having FIX-W being 39.0/10, MAE-W 39.64/10, PB-NAP 39.128/10, and CIX 39.192/10. With that, everyone outside northern california has 39/8 pointing towards northern california, and we don't bother to do anything with /9's in the scenario. [Obviously, the actual weights of who's connected where would make the situation different in practice, but this is just an example].

In failure conditions:

If S falls off MAE-WEST, how does it get traffic to the customers of A, M, N9, NE and X that are addressed by 39.0/10?

Send it to PB-NAP and hope someone does the right thing? Or, better, have an agreement with someone to do the right thing in both directions if that sort of failure occurs.

How do A, M, N9, NE, and X get to S's customers that are addressed by 39.0/10?

Have S announce their 39.0/10 routes at PB-NAP temporarily, which breaks the aggregation somewhat but is still better than existing aggregation.

If A falls off MAE-WEST, how does it get traffic to the customers of A, M, N9, NE, and X that are in 39.0/10? How does it get to all of 39.0/9?

How would it do so now? Route through someone else to get to MAE-W.

How does the rest of the world get to A's customers that are addressed in 39.0/10?

Headache.

What happens in both these cases to traffic between Z and everything in 39.0/9, in each direction?

Assuming Z is paying N9 or M or S for additional backbone transit, then it goes down that pipe. Same as current situation.

These definitely are not hypothetical questions.

Nor are most of them insolvable, though they do need to be addressed.

If geographical or stratum-based allocation that is not _also_ provider-based is to fly, this is precisely the type of thing that must be dealt with.

Yes, you're right. The right solution may not be this, it may end up being hitting InterNIC over the head and making them allocate larger blocks to backbone providers, and having backbone providers not fuss too much about small to midlevel ISPs going multihomed, and having backbone providers allocate space for growth for small to midlevel ISPs. But, at this point, that's not happening either. -george