BOOK ON ROUTING IN OPPORTUNISTIC NETWORKS Chapter 2: Social-aware Opportunistic Routing: the New Trend 1Waldir Moreira, 1Paulo Mendes 1SITILabs, © SITILabs, University Lusófona, Portugal University Lusófona 1 Goal of this Chapter Introduce different opportunistic routing approaches Learn about existing opportunistic routing taxonomies Show how social information improves data forwarding © SITILabs, University Lusófona, Portugal 2 Introduction Users want to be connected at all times Produce and consume content (prosumers) Devices capabilities contribute Powerful (e.g., processing, storage) Allow networks to be formed on-the-fly Opportunistic routing provides the means Allows the exchange of information even when end-toend paths do not exist between communicating parties © SITILabs, University Lusófona, Portugal 3 Introduction Issue: cope with link intermittency Due to node mobility, power-saving schemes, physical obstacles, dark areas Opportunistic routing relies on the Store-carry-and-forward paradigm © SITILabs, University Lusófona, Portugal 4 Introduction There are different routing approaches Ranging from network flooding to more elaborate replication schemes A new trend emerges amongst solutions Based on social similarity metrics (e.g., relationship, affiliation, importance, interests) Focus of this chapter Social-aware opportunistic routing Great potential for improving opportunistic forwarding © SITILabs, University Lusófona, Portugal 5 Opportunistic Routing Approaches Different approaches Single-copy Routing Epidemic Routing Probabilistic-based Routing • Frequency Encounters • Aging Encounters • Aging Messages • Resource Allocation © SITILabs, University Lusófona, Portugal 6 Existing Opportunistic Routing Taxonomies Focus mostly on the efficiency Achieve higher delivery rates Spare network resources The focus should also include Analysis of the topological features (e.g., contact frequency and age, resource utilization, community formation, common interests, node popularity) © SITILabs, University Lusófona, Portugal 7 Existing Opportunistic Routing Taxonomies © SITILabs, University Lusófona, Portugal 8 New Opportunistic Routing Taxonomy Social similarity metrics gained attention Human social behavior varies less than the one based on mobility Based on social behavior abstracted from contacts between people, time spent with them, existing relationships © SITILabs, University Lusófona, Portugal 9 Experimental Analysis Goal Show how opportunistic routing can benefit from social awareness Done in two scenarios Heterogeneous (synthetic mobility models) Real human traces © SITILabs, University Lusófona, Portugal 10 Experimental Methodology Each experiment run ten times to provide results with a 95% confidence interval Performance metrics Average delivery probability • Ratio between the total number of delivered and created messages Average cost • Number of replicas per delivered message Average latency • Time elapsed between message creation and delivery © SITILabs, University Lusófona, Portugal 11 Experimental Setup © SITILabs, University Lusófona, Portugal 12 Results on Heterogeneous Scenario Average Delivery Probability dLife and dLifeComm consider users’ dynamic behavior • Delivery rate over 74% Bubble Rap is affected by limited buffer (2 MB) © SITILabs, University Lusófona, Portugal 13 Results on Heterogeneous Scenario Average Cost Bubble Rap, dLife and dLifeComm have low cost as they use social similarity to replicate • Cost of maximum 546, 319, and 319, respectively to perform a successful delivery © SITILabs, University Lusófona, Portugal 14 Results on Heterogeneous Scenario Average Latency dLife and dLifeComm take longer to forward (strong social links or important nodes) Bubble Rap chooses forwarders with weak ties • Centrality does not capture dynamism © SITILabs, University Lusófona, Portugal 15 Results on Human Trace Scenario Average Delivery Probability Contact sporadicity affects • Bubble Rap and dLife: Delivery 25.5% • dLifeComm relies on node importance – Takes too long to reflect reality © SITILabs, University Lusófona, Portugal 16 Results on Human Trace Scenario Average Cost Bubble Rap, dLife and dLifeComm produced approx. 24.52, 24.56, and 28.79 replicas • With few extra copies almost the same delivery performance as Spray & Wait © SITILabs, University Lusófona, Portugal 17 Results on Human Trace Scenario Average Latency Bubble Rap had similar behavior as in previous scenario dLife and dLifeComm are affected by non-dynamism of user contact © SITILabs, University Lusófona, Portugal 18 Conclusions Despite the challenges in the scenarios Social-aware proposals that are able to capture dynamism of user behavior • Good delivery performance with low associated cost and a subtle increase in latency • Indeed have great potential in improving forwarding More improvements Consider point-to-multipoint communication Increase even more performance of social-aware solutions © SITILabs, University Lusófona, Portugal 19 Acknowledgements Thanks are due to FCT for supporting the UCR (PTDC/EEA-TEL/103637/2008) project and Mr. Moreira’s PhD grant (SFRH/BD/62761/2009), and to the colleagues of the DTN-Amazon project for the fruitful discussions. © SITILabs, University Lusófona, Portugal 20 BOOK ON ROUTING IN OPPORTUNISTIC NETWORKS Chapter 2: Social-aware Opportunistic Routing: the New Trend 1Waldir Moreira, 1Paulo Mendes 1SITILabs, © SITILabs, University Lusófona, Portugal University Lusófona 21