Statement: Future Internet will be based on a new paradigm of dissemination aimed at supporting intermittently connected hosts in an efficient manner. Talk Title: Postcards from the Edge: A Cache and Forward Architecture for the Future Internet by Sanjoy Paul, WINLAB, Rutgers University, sanjoy@winlab.rutgers.edu; sanjoy.paul@gmail.com Abstract: First generation of networking was powered by circuit-switching in telephony networks, while the second generation of networking has been powered by p acket-switching in the Internet. In fact, the Internet set out to solve a problem, namely data transmission, which is very different from what the tel ephony network was designed to solve, namely, end-to-end circuit connection set up, and hence there was a need for new networking paradigm, namely, pa cket switching and a new set of protocols, namely, TCP/IP. Moreover, the Internet was built as an overlay on the telephone network with the objective of efficiently using the underlying resources to support the new paradigm of packet switching. While TCP/IP has been an elegant solution to for the ne tworking problem at hand several decades ago when it was designed, the current problems in networking are very different in nature and require a shift in our thinking process. The overwhelming use of todays network (more than 99% by most measurements) is for a machine to acquire a named chunk of data. The named chunk can be a webpage, a picture, a song, a movie/video or an email. For example, BitTorrent, which contributes more than 30% of traffic on the Internet, is all about acquiring named contents, so is YouTube and MySpace. YouTube hosts about 6M videos growing 20% every month, requires 45TB of storage and spends millions of $ bw/month for transporting video clips. MySpace last year generated an astounding 31.5 billion page-views/month. In order to solve this f undamental problem of efficient content dissemination, we have resorted to building ad-hoc overlay networks, namely, Akamai CDN, Apple Rendezvous, Son us mesh, BitTorrent and many others. Clearly we need a new networking paradigm that supports dissemination of named chunks of data in an efficient man ner. With the maturity of wireless networking technology, the number of endpoints connected wirelessly to the Internet has overtaken the number of wired en dpoints, and the difference between the two is widening. Compared to 500M wired hosts/servers as of 2006, there are more than 2 billion mobile phones 50% of which are Internet capable, not to speak of the proliferation of WiFi-enabled laptops and PDAs, and the whopping 5B-10B projected sensor deploy ments by 2015. Wireless mesh networks, sensor networks, and vehicular networks represent some of the new growth segments in wireless networking in add ition to mobile data networks which is currently the fastest growing segment in wireless industry. Wireless links are time-varying in bandwidth, error rate, and connectivity, and such networks beg for opportunistic transport, especially when the link bandwidth is high, error rate is low, and when th e endpoint is connected to the network. Connected is a binary attribute in TCP/IP meaning you are either part of the Internet and can talk to everyt hing or you are isolated. In addition, connecting requires a globally unique IP address that is topologically stable on routing timescale (minutes to hours). This makes it difficult and inefficient to handle mobility and opportunistic transport in the Internet. Clearly we need a new networking parad igm that avoids a heavyweight operation like end-to-end connection, and enables opportunistic transport. Technology trends in keeping with Moores law enable us to have 5GB semiconductor memory for $50 in 2006 as opposed to 256MB in 2000; 2.5 GHz CPU in 2 006 in contrast to 500Mhz in 2000, and 2GB storage for $1 in 2006 compared to 0.2 GB in 2000. Thus storage costs have plummeted and plenty of processing power is available at a significantly lower price point. Keeping in mind the need for dissemination networking which requires efficient distribution of named chunks, the desire to support intermittently conn ected endpoints which requires opportunistic transport of data, and exploiting the decreasing cost and increasing capacity of storage and processing, we propose Postcards from the Edge: A Cache and Forward (CNF) Architecture for the Future Internet. CNF architecture provides unified and efficient transport services to end hosts that may be wired or wireless; static, mobile, and/or intermittently connected; and either resource rich or poor. Fund amental to this architecture is a transport layer service that operates in a hop-by-hop store-and-forward manner with large files. To realize this arc hitecture, this project designs, implements and evaluates a new network architecture that incorporates the following elements: (1) reliable hop-by-hop transport of large files; (2) configurable protocol for reliable transport based on the characteristics of the link between adjacent CNF nodes, (3) c ontent-based rather than address-based adaptive routing, (4) push-pull architecture for opportunistic delivery of files both to and from the wired net work; (5) location-independent naming of content; (6) enhanced naming of devices to provide location information for mobile terminals; and (7) distrib uted caching of static and dynamic content to enable dissemination of content in an efficient manner. CNF network is an overlay on the Internet, just as the Internet is an overlay on the telephone network. Just as TCP/IP leveraged the existence of the telephone network to solve a fundamentally different problem of data transmission, CNF is designed to leverage the TCP/IP network to solve disseminati on of named chunks of data, which is a problem fundamentally different than what TCP/IP was designed to solve. CNF changes the viewpoint in that it fo cuses on data rather than on endpoints as in TCP/IP, and hence data is of paramount importance, not the supplier of data, and similarly, security and trust is based on the data, not on the pipe that is used to deliver the data. Finally anything that moves bits in time and space is leveraged for data dissemination in CNF. From a broad perspective, CNF architecture adopts just the right amount of core techniques from Peer-to-Peer, Overlay, and Disr uption-Tolerant Networking to make it really a networking architecture for the future Internet. * Thanks to Van Jacobson for helping to crystallize some of the core concepts of CNF.