Using INRP and NBRP as described above, a client request is mapped to a nearby content server within one round-trip time to a content router near to the client, without the need to contact off-path name servers. Latency is typically dominated by round-trip time to the content server, not by content routing; cache misses require only one RTT to do a new name lookup. Moreover, by increasing the number of content routers, this property is retained even as the Internet scales to ever larger size and increasing number of clients.
By making name lookup low-latency, INRP eliminates the need to perform multiple levels of redirection in DNS. Instead, low-TTL address records can be returned at the first layer of naming, to preserve sensitivity to network conditions. (Assuming, of course, that the content routers can handle the name lookup load required, which will be addressed below.)
By making INRP and NBRP open standard Internet protocols, all ISPs, router manufacturers and content providers can participate in this content routing layer, further enhancing the cost-effective scalability to the clients.
The key issue raised by our solution is the scalability of NBRP, given it is distributing naming and load information, not just aggregatable addressing information. Ideally, we would like to completely replace the current Domain Name System by INRP and NBRP, to remove dependence on root name servers-- themselves a large source of connection setup latency and scalability concerns.