DARPA's Disruption-Tolerant Networking (DTN) program is developing technologies that enable access to information when stable end-to-end paths do not exist and network infrastructure access cannot be assured. DTN technology makes use of persistence within network nodes, along with the opportunistic use of mobility, to overcome disruptions to connectivity. In this paper, we describe the SPINDLE Disruption-Tolerant Networking system and related technology being developed at BBN under the DTN program. Using an open-source, standards-based core with a plugin architecture and well-specified interfaces, we enable independent development and insertion of innovative DoD-relevant technology while allowing the core system to be refined and engineered within a COTS context. SPINDLE technology innovations include: (i) routing algorithms that work efficiently across a wide range of network disruption, (ii) a name-management architecture for DTNs that supports progressive resolution of intentional name attributes within the network (not at the source), including support for "queries as names" and name-scheme translation, (iii) distributed caching, indexing, and retrieval approaches for disruption-tolerant content-based (rather than locator-based) access to information, and (iv) a declarative knowledge-based approach that integrates routing, intentional naming, policy-based resource management, and content-based access to information. We present preliminary results that show that the DTN approach outperforms traditional end-to-end approaches across a wide range of network disruption.

We describe PRioritized EPidemic (PREP) for routing in opportunistic networks. PREP prioritizes bundles based on costs to destination, source, and expiry time. Costs are derived from per-link "average availability" information that is disseminated in an epidemic manner. PREP maintains a gradient of replication density that decreases with increasing distance from the destination. Simulation results show that PREP outperforms AODV and Epidemic Routing by a factor of about 4 and 1.4 respectively, with the gap widening with decreasing density and decreasing storage. We expect PREP to be of greater value than other proposed solutions in highly disconnected and mobile networks where no schedule information or repeatable patterns exist.