Middleboxes no longer considered harmful

In this paper, the authors propose an architectural solution to accommodating "middleboxes", or network elements that violate certain Internet layering, such as NATs and firewalls. They argue that these middleboxes are likely to be useful even with migration to IPv6, for reasons such as the need to maintain private address realms or to outsource firewall operations. To do this, Delegation-Oriented Architecture (DOA) is proposed, which sits on top of the IP layer.

From my understanding, each host is given a unique 160bit identifier (a cryptographic hash?) which makes each host uniquely addressable. EIDs can be resolved through the use of DHT to either an IP address of a delegate or another EID. Essentially, the architecture allows a sender to address a packet using an EID (of the receiver), while the receiver can delegate responsibility for handling/forwarding packets for itself to some intemediary (addressed by either EIDs or IP address). I am tempted to think of this as providing a mixture of routing and DNS on top of "normal" Internet.

What I am not clear about is who maintains responsibility for operating the DHT for EID resolution. I am also torn between whether this is meant to be operated among small network of users (i.e. each group of hosts could operate their own DOA) or is meant to be a netwide deployment, where everyone would share the same EID namespace. In the latter case, how would DHT scale?

Overall, it is a very intriguing idea. Does it have additional value over IPv6? I'm not sure if I would agree with the authors' claims that it does. Outsourcing of firewall functions can probably still take place without DOA, though it is likely that NATs will still have value (multiple hosts sharing a single internet connection).