Scaling the Internet is a multi-dimensional problem, spanning not just the number of nodes, but also the capabilities of those nodes, the diversity of security and protection demands, the diversity of available bandwidth, and the complexity of network topology. A new architecture must recognize the multiplicity of demands, from small embedded systems to very large distributed organizations, across a wide range of administrative boundaries and requirements. We believe TRIAD is a promising new architecture on which to base the future of the Internet, which addresses all of these issues, not just the low-level problem of too few network identifiers.
TRIAD solves the problems of NAT, allowing it to be accepted as a legitimate part of the Internet architecture, and sensibly deployed for its many uses. TRIAD also provides an significantly improved directory service, ensuring better naming support for applications and allowing multicast naming, mobility, policy-based routing and wide-area load balancing to be implemented at the higher level, rather than complicating the network layer. Moreover, TRIAD is largely IPv4 and DNS compatible for end hosts, leaf and backbone networks, simply requiring extensions to NAT-capable boundary routers. Thus, it retains the key aspects that have allowed the Internet to be so successful to date, including end-to-end semantics, and addresses one of its major deficiencies, namely lack of a dependable directory service.
Compared to IPv6, TRIAD is more backwards compatible, more deployable, more efficient and more secure while providing the same end-to-end semantics and recovery relative to network failures.
TRIAD, as the name suggests, is based on three key ideas. For one, TRIAD makes the user-assigned host or multicast channel DNS name the only global identifier, making packet addressing local and transient, and thereby short, efficient and automatically assignable. It integrates the directory system into the network infrastructure of routers to ensure availability and trust that matches the network itself.
Second, TRIAD uses a name-based packet pseudo-header, a natural approach, given the replacement of addresses by names in general. This approach supports end-to-end semantics even though the packet addresses are translated at each relay agent. TRIAD-TCP also uses name-based connections and name remapping as part of recovery to make the translation and relaying state in the network ``soft'', providing the same end-to-end resilience to failure as the original Internet architecture.
Finally, TRIAD introduces a relay layer and a shim protocol WRAP between IPv4 and transport protocols, providing extensible addressability between address realms and independence of addressing within each realm. In particular, local address structure can be completely hidden from the rest of the Internet and external Internet structure can be completely hidden from the local realm. Moreover, intra-realm communication can optimize out WRAP, incurring the same packet overhead in size and processing as IPv4. The simplicity of WRAP makes it feasible to implement in hardware in the next generation of switch/routers, allowing wire speed relaying, even at the highest performance levels.
A broader contribution of this work is the recognition of the value of NAT and directory services to the Internet. Both are critical and integral aspects in practice, yet both were omitted from the original architecture. They clearly need to be incorporated as we have done in TRIAD if there is to be an architecture that matches reality. In particular, the view of NAT as an interim hack is just plain wrong.
We believe that the primary competition to TRIAD at this stage is the continued ad hoc deployment of NAT and application-level proxies, not IPv6. Continued growth of the Internet without a guiding architecture will significantly detract from its reliability, security and eventually, utility.