Content Based Networking in DTNs

1. Content Based Networking in DTNs Dagstuhl Seminar on Information-Centric Networking December 7, 2010 Armando Caro acaro@bbn.com BBN Technologies Vikas Kawadia Niky Riga

2. Spectrum of Connectivity & Infrastructure Fixed Wired Infrastructure (Internet) Mobile Ad Hoc Networks (mostly connected, but no infrastructure) Highly Disrupted (never e2e paths)

3. Our DTN Perspective ?

4. Possible Applications of This Networking Paradigm • Disaster response • Initial survey, inventory, co-ordination and management • Sensor / information gathering • Real time road traffic monitoring • Forestry, mining, fishing, etc type of jobs • Extend connectivity to rural areas • Email & social networking • Online shopping • General bulletin boards (jobs, etc) • ‘Local’ social networking w/o infrastructure • Works in subways, cruises, adventure trips • Make new friends: mobile match.com

5. Why Content Based Networking? • Information search and access depend on connectivity & infrastructure • No search possible without access to a search engine • Traditional indexing techniques assume connectivity • i.e., not connected then doesn’t exist • In mobile & disrupted environments, only information within a connected cluster is available for retrieval • Today’s networks are used inefficiently as simple content-agnostic bit pipes • Content retrieval requires end-to-end connectivity • Critical links are potentially overloaded with several redundant copies

6. Replicating Content • Leverage routing control information to dynamically detect and adapt clusters that have “stable” connectivity • Maintain a copy of content in each cluster • Use DHT to spread burden of storage within a cluster • Benefit: Minimize dependency of inter-cluster transmissions

7. Managing Content Explosion (1/2) • Should not simply flood content to all clusters • Wastes bandwidth, storage, and battery resources • On-demand Reactive Caching • Only cache what was requested before (either locally or by a neighboring cluster) • Motivation: other nodes are likely to request the same content • Pro: Reduce transmission and caching of content that may never be requested in a cluster • Con: Delay in fulfilling initial request, which might be significant in case of disconnectivity

8. Managing Content Explosion (2/2) • Intelligent Pre-placement: • Use community detection techniques to predict future demands • Detect nodes that have the same usage patterns • Determine which content objects are related • Similar to Amazon’s suggestions: “other users like you viewed these other items too” • Any idle periods of resource utilization can be leveraged to pre-place content

9. Metrics & Evaluation • User-experienced latency • How long does it take for a user request to be satisfied? • In a connected environment, is latency comparable to current approach? • Recall (i.e., cache hit ratio) • How much of request content is retrievable from local or neighboring cluster’s cache? • Content throughput • How much content can we push through the network? • Relevance & precision • How relevant are the pre-placed content? • Have they been used to satisfy queries or just wasted resources? • Resource overhead • What’s the bandwidth and storage overhead in the system?

10. Challenges • Expressing content: Unique IDs vs Expressive Descriptors • Name mapping infrastructure in an Mobile Ad Hoc DTN? • Congestion avoidance • Especially when clusters opportunistically sync • Scheduling of pre-placed content • Prioritization: locally demanded vs transient content • User interface • Delayed responses • Potentially multiple content objects satisfying a single query and arriving at different times • Security (enough said ) • …