FIND (Future Internet Design) Project Cluster

Project Objectives

WINLAB is an active participant in NSF's future Internet initiatives including FIND (Future Internet Design) and GENI (Global Environment for Network Innovation).  In 2006, WINLAB faculty members and their collaborators were awarded the following FIND projects under the NSF NeTS program:

1. NeTS-FIND: Postcards from the Edge:
   A Cache-and-Forward Architecture for the Future Internet.
   Profs. Roy Yates, D. Raychaudhuri, and Sanjoy Paul
   (in collaboration with Prof. Jim Kurose of U Mass)
2. NeTS-FIND: CogNet - An Experimental Protocol Stack for Cognitive Radio Networks and Its Integration with the Future Internet.
   Profs. D. Raychaudhuri, N. Mandayam, Predrag Spasojevic (in collaboration with Prof. Joe Evans at U Kansas and Profs. Srini Seshan and Peter Steenkiste of CMU)
3. NeTS-FIND: A Geometric Stack for Location-Aware Networking.
   Prof. Marco Gruteser (PI), Prof. Rich Martin (Co-PI)

Each of these projects is intended to explore clean-slate networking approaches to mobile/wireless services in the future Internet.

Technology Rationale:

The WINLAB cluster of FIND projects are aimed at identifying the impact of wireless/mobile devices on the future Internet.  In particular, the “cache and forward” architecture represents an attempt to design a global network with efficient support for mobile content delivery over wireless channels.  The second project on “CogNet – An Experimental Protocol Stack for Cognitive Radio Networks” examines the impact of emerging cognitive radio technology on the future network.  The third project on “Geometric Stack for Location-Aware Networking” investigates issues of performance and scale which arise in geographic routing scenarios of increasing importance in mobile and vehicular applications.  Each of these projects is expected to lead to several new protocol concepts for the future Internet, along with extensive simulation and tesbed evaluations.  In the longer term, promising protocol results may be implemented as services on NSF’s GENI network when it is deployed.

Technical Approach:

(1) Cache-and-Forward Network: The cache-and-forward concept exploits rapidly decreasing cost of storage to provide support efficient content delivery to mobile users. Fundamental to this architecture is a smart "link layer" service that operates in a hop-by-hop store-and-forward manner with large data files. For mobile nodes, the architecture enables opportunistic push-pull delivery of files, both to and from the wired network. Routing to and from mobile terminals will exploit location information provided by an enhanced name service. Distributed caching of popular content will occur throughout the network which together with a novel file naming service will make peer-to-peer file transfer an efficient and scalable first class service.

Current work on the CNF project is focused on three main areas: (1) analysis and simulation of the hop-by-hop architecture and comparison with TCP/IP for different usage scenarios; (2) design of protocol mechanisms for integration of packet forwarding and content caching; (3) evaluation of hop-by-hop transport in wireless multihop scenarios characterized by disconnections and self-interference.  ORBIT/PlanetLab proof-of-concept experiments for a baseline set of CNF protocols are also planned. [This is a joint project with Prof. Kurose at U Mass].

(2) CogNet: This project has two major thrusts: the first is to identify broad architecture and protocol design approaches for cognitive networks at both local network and the global internetwork levels.  This architectural study should lead to the design of control/management and data interfaces between cognitive radio nodes in a local network, and also between cognitive radio subnetworks and the global Internet.  The second thrust is to apply these architectural and protocol design results to prototype an open-source cognitive radio protocol solution (the CogNet stack) and use it for experimental evaluations on emerging cognitive radio platforms. A number of architectural issues will be examined as we try to identify an efficient and complete solution – these include control and management protocols, support for collaborative PHY, dynamic spectrum coordination, flexible MAC layer protocols, ad hoc group formation and cross-layer adaptation.

The experimental component of this project aims to prototype an open-source Linux-based CogNet software protocol stack for use with emerging cognitive radio platforms (such as the GNU/USRP2 radio to be used as the baseline, the KU agile radio or the Lucent/WINLAB network-centric prototype), and make this software available for community research.  The prototype software will be validated in two steps: first in a wireless local-area radio network scenario with moderate numbers of cognitive radio nodes, and later as part of several end-to-end experiments using a wide-area network testbed such as PlanetLab (and GENI in the future).  We believe that the proposed CogNet software prototyping, open-source distribution and experimental validation aspect of the project is of particular significance to the wireless research community which currently lacks a well-supported experimental cognitive radio platform   [This project is a collaborative effort involving Rutgers Univ., CMU, KU, and Blossom Research].

(3) Geometric Stack: Using location information to optimize wireless networks has emerged as a powerful approach to scale capacity in high density or high mobility systems. Geographic routing (see Fig 2) gains significantly by reducing the overhead of maintaining or acquiring network topology information in addition to reducing the size of routing tables. In addition to unicast routing, geographic multicast (“geocast”) enables group communication as required by many sensing applications. This method of routing proves its advantage when locations of the intended recipients change frequently due to mobility.

These location-aware algorithms have to date been studied through simulation models. Thorough investigation of routing protocols and widespread adoption will only be possible if implementations in actual systems validate these results under realistic localization error, interference, and MAC protocol assumptions.  The geocast primitive transmits a message to all nodes within a defined geographic area. In a vehicle-to-vehicle communication scenario for example, geocasting may be used to warn following vehicles of an obstacle or sudden stopping. A number of open research questions are being investigated by research groups working on this dense vehicular network scenario.  One important issue is the design of the MAC layer protocol that addresses varying radio density and the broadcast/flooding nature of the packet traffic being carried.  Current 802.11 MAC protocols suffer from serious performance limitations due to hidden node and exposed node problems, so that it is desirable to investigate fundamentally new approaches.  The second issue is that of designing an efficient geocast routing procedure taking into account control overheads and the role of infrastructure network access points that allow for routing through the wired network where appropriate.  An interesting design issue is the comparison between clean-slate georouting approaches vs. overlay or hybrid routing techniques which continue to use IP in the core network and employ location servers for geographic address resolution.  The above problems are being studied via prototype implementations on the ORBIT grid emulator as well as the outdoor ORBIT field trial system with vehicular radio nodes.

Results-to-date and future work plan:

These FIND projects started in mid-2006 and are still at an early stage of development.  Initial protocol architecture designs have been completed for each of the three proposals (CNF, CogNet and Geometric Stack) and are being evaluated with initial simulation models, etc.  More complete protocol descriptions and validated software implementations are planned for 2007.  More realistic validation experiments will later be conducted on ORBIT/PlanetLab and GENI (when available).