Vadim replying to me wrote: | > Actually the distribution problem is pretty simple. | > | > In your forwarding table add a per-prefix cost. | > | > The cost is expressed in terms of tokens. | > | > The tokens are deducted from a token bucket. | > | > The number of tokens in the bucket (depth and refresh-rate) is | > arrived at by some engineering or sales process. | > | > This polices MEDs nicely. | | Hm. Aggregation breaks that scheme nicely. Absense of | aggregation breaks routing nicely. Remember Yakov's route push and route pull model. In the system above, when you have multiple connections to someone sending you only very short (aggregate) prefixes, you have an incentive to do a route pull in order to optimize your routing. That is, you can acquire information about the other network to help you choose the cheapest entry point. Moreover, the route pull does not have to be very dynamic in order to increase (as opposed to maximize) the number of packets you are able to send into the network. Sean. P.S.: There is also a direct trade-off -- choosing the most expensive entry point merely minimizes the number of packets one will be able to inject into the other network for any given token-bucket- refresh-rate. Rather than choose cheaper entry points, one could simply choose to buy a faster refresh-rate.