They actually propose to use this on top of TCP and UDP (or any other transport). They want to address content using names (like bittorrent magnet links), not address machines like DNS does. See this http://named-data.net/project/execsummary/ and this http://named-data.net/project/archoverview/ (also they have this on Github https://github.com/named-data/ndn-cxx). They want to use caching at router level as a substitute for client-side congestion control. I'll reproduce below some sections.

Routing and Forwarding

NDN routes and forwards packets on names, which eliminates four problems that addresses pose in the IP architecture: address space exhaustion, NAT traversal, mobility, and address management. There is no address exhaustion problem since the namespace is unbounded. There is no NAT traversal problem since a host does not need to expose its address in order to offer content. Mobility, which requires changing addresses in IP, no longer breaks communication since data names remain the same. Finally, address assignment and management is no longer required in local networks, which is especially empowering for embedded sensor networks.

(...)

Caching

Automatic in-network caching is enabled by naming data. Since each NDN Data packet is meaningful independent of where it comes from or where it may be forwarded to, a router can cache it in its content store to satisfy future requests. Upon receiving a new Interest, the router first checks the Content Store. If there is a data whose name falls under the Interest’s name, the data will be sent back as a response. The Content Store, in its basic form, is just the buffer memory in today’s router. Both IP routers and NDN routers buffer data packets. The difference is that IP routers cannot reuse the data after forwarding them, while NDN routers are able to reuse the data since they are identified by the data names. For static files, NDN achieves almost optimal data delivery. Even dynamic content can benefit from caching in the case of multicast (e.g., teleconferencing) or packet retransmission after a packet loss. Cache management and replacement is subject to ISP policies and is one of our research topics.

Caching named data may raise privacy concerns. Today’s IP networks offer weak privacy protection. One can find out what is in an IP packet by inspecting the header or payload, and who requested the data by checking the destination address. NDN explicitly names the data, arguably making it easier for a network monitor to see what data is being requested. One may also be able to learn what data is requested through clever probing schemes to derive what is in the cache. However NDN removes entirely the information regarding who is requesting the data. Unless directly connected to the requesting host by a point-to-point link, a router will only know that someone has requested certain data, but will not know who originated the request. Thus the NDN architecture naturally offers privacy protection at a fundamentally different level than the current IP networks. Transport

The NDN architecture does not have a separate transport layer. It moves the functions of today’s transport protocols up into applications, their supporting libraries, and the strategy component in the forwarding plane. Multiplexing and demultiplexing among application processes is done directly using names at the NDN layer, and data integrity and reliability are directly handled by application processes where the appropriate reliability checking, data signing and trust decisions can be made.

An NDN network is designed to operate on top of unreliable packet delivery services, including the highly dynamic connectivity of mobile and ubiquitou...