nanog@Overkill.EnterZone.Net (John Fraizer) writes:
Beyond that, I'm _REALLY_ sick of people pissing and moaning about NAPS being congestion points. If you're so tired of the exchange point being slow, INVEST IN MAKING IT FASTER!!!!! ...
That's what PAIX did. And Equinix for that matter. Exchange points aren't slow, though I admit that some of the ATM-based exchanges have hit their scaling limit. Exchange points based on Ethernet with rich PNI opportunities literally do not have scaling limits. The existence of the OPTIX/SAVVIS/InterNAP model isn't an an indictment of exchange points so much as it is of the routing system itself. However, bloating the global routing table with lots of discontiguous subnets isn't a workable solution. These networks need backbones and rich peering so they can negotiate deaggregation with just those peers who want the detail. Otherwise they're just value-added resellers of other networks who DO have backbones and rich private peering. (Which is what MIBH was, so I'm aware of the customer-perceived benefits in the value being added -- the difference is that MIBH was honest about it.)