Name-Based Routing

In the TRIAD architecure, addresses are no longer a meaningful way to identify endpoints in the wide area. Instead, routing information takes the form of name suffixes, since names are the endpoint identifiers. A conventional routing table can be viewed as a directory mapping IP addresses to forwarding information; in TRIAD, the forwarding table is indexed by DNS name instead.

The Internet Relay Protocol (INRP) performs name-to-address conversion in TRIAD by using the routing information mantained by relay nodes. Name lookups are forwarded towards the authoritative server for the query name. The resulting address is built up by each of the relay nodes involved. Although routing information is stored as names, packet forwarding uses these addresses, so forwarding lookup can be quite efficent and names need not be placed in every packet.

The Name-Based Routing Protocol (NBRP) is a dynamic mechanism for updating the routing information in relay nodes. Its behavior is similar to the Border Gateway Protocol (BGP); just as BGP distributed address prefix reachability information among atonoumous systmes, NBRP distributes name suffix reachability among (and within) address realms. Routing using names helps to ensure the availability of those names, even in the absence of multiple address realms.

Both INRP and NBRP can be implemented as extensions to the current Domain Name System (DNS). Name-based routing can provide significant benefits for existing networks, even without WRAP. Among these are improved fate-sharing between naming and packet delivery, improved lookup latency, lower load upon root name servers, and a natural caching hierarchy.

Content Routing

A key problem in today's Internet is locating the nearest replica of a particular "content" such as CNN or Yahoo. Existing proprietary solutions such as Cisco's Distributed Director and Akamai's FreeFlow attempt to solve this by picking a particular replica server based on the originating IP address of a DNS request. This can result in relatively long latency on name lookups if the DNS server is much farther away in the network than the replica server chosen. Both services also rely on short time-to-live (TTLs) on name-to-address mappings in order to keep consistent with the current routing topology, which increases load on DNS servers.

NBRP provides a solution to replica location, called Content Routing. A "content name" such as "yahoo.com" is advertised by each of the replica sites, so that the initial name lookup proceeds directly to the closest replica. The content name maps to a list of replica servers, rather than directly to an address; this list may be reordered to reflect current preference values without locking a client into a particular (and possibly unavailable) server. (Naturally, the inital preference ordering provided by the nameserver at a replica site will list the local server first.) Finally, the TTL of a particular ordering is distinguished from the TTL of the entire list; this allows clients to keep current with routing changes without requiring servers to throw away the still-valid list of replicas.

Aggregation

A key problem in name-based routing is maintaining the routing database efficiently even in the presence of names that do not match the routing hierarchy. From a routing perspective, the DNS is effectively flat, with several million names under the ".com" label.

NBRP addresses this problem by combining collections of name suffixes into routing aggregates. Routing updates typically update a small number of aggregates rather than many individual entries. Aggregate membership is relatively long-lived, so that relay nodes can amortize the cost of learning the aggregate membership (through the directory service) over many routing updates.

This indirect aggregation technique may be applicable to IPv4 and IPv6 as well as name-based routing. It removes the necessity of matching names (or addresses) with the topology of the network by explicitly creating aggregates rather than relying upon allocation schemes or renumbering.