1. Autonomic and Collaborative Protocols in Wireless Delay Tolerant Networks Stavros Toumpis Department of Informatics Athens University of Economics and Business Athens, Greece CROWN KICKOFF, 11/5/12 1

2. PART A: Delay Tolerant Networks 2

3. Definition Delay in the delivery of packets is very large (specifically, comparable to the time needed for the topology to change substantially) Two cases 1.Very large delays are necessarily large (e.g., interplanetary networks [Burleigh et al. 2003]) 2.Very large delays are a design choice (e.g., Zebranet, Juang et al. 2002). Delay is conscientiously traded off. In the context of wireless networks, large delays typically translate to communication by physical transportation of data (either partially or exclusively) 3

4. Applications Interplanetary networks Sensor networks (Zebranet) The Internet Vehicular Networks, for certain kinds of traffic (GeoDTN+Nav) 4

5. Recent History Infostations [Goodman et al. ‘97] Epidemic Routing [Vahdat/Becker ‘00] Mobility Increases the Capacity of Wireless networks [Grossglauser/Tse ‘01, Toumpis/Goldsmith ‘04] Data Mules [Shah et al. ‘03] Zebranet [Juang et al. 02, Small/Hass ‘03] Delay Tolerant Architecture [Jain et al. ‘03] Spray and Wait [Spyropoulos et al. ‘05] MaxProp [Burgess et al. ‘06] 5

6. Earlier History Much work in Operations Research in the context of dynamic flows and networks (which are functions of time) Ford/Fulkerson constructed maximal flows in ‘54 and maximal dynamic flows in ’58! Ogier studied minimum delay routing and related problems in the ‘80s. Ferreira et al. [Ferreira 04,10] and Merugu et al. [Merugu et al. ‘04] applied dynamic flows in the context of (wireless) DTNs 6

7. Classification of DTN Analysis Do we know the topology evolution of the network? – If YES, then we can study it using tools from network optimization theory, notably dynamic flows and networks – If NO, then we can use tools from probability and related fields (e.g., stochastic control) 7

8. Why are DTNs interesting in the context of this project? If Wireless Networks added a spatial component to the analysis of networks… … then Delay Tolerant Networks add a time component… …and room for innovation is still there… … particularly in the areas covered by this project. 8

9. Part B: Autonomous and Collaborative Protocols in DTNs 9

10. Organization of the project 10 WP1: Understanding and influencing uncoordinated interactions of autonomic wireless networks WP2: Optimization through network coordination WP3: Autonomic and collaborative protocols in Wireless DTNs Task 3.1: Autonomic operation of wireless DTNs Task 3.2: Coordinated operation of wireless DTNs Task 3.3 Realistic wireless DTN protocol design

11. Task 3.1: Autonomic operation of wireless DTNs Main tool: probability theory DTNs are frequently partitioned, therefore autonomic operation based on local decisions is appealing/necessary – Nodes decide on next hop of packet – Nodes decide on what packet to transmit/delete – etc. Resources (bandwidth, buffer spaces) are typically constrained, so selfish behavior is expected – Nodes would like to only transmit/store their own packets, etc. 11

12. Sub-Task 3.1.1: Geographic Routing In geographic routing, next hop for a packet is decided according to location of destination and topology near the current packet holder. Problem: find optimal behavior for current holder, based on local knowledge. An obvious tradeoff exists between transportation cost and packet delivery delay. Tool for performing the analysis: Stochastic Geometry 12

13. Subtask 3.1.2: Delay-Throughput tradeoff of DTNs Fundamental tradeoff: The more copies of a packet are transmitted, the faster it will arrive at its destination, but the smaller the throughput becomes. Goal: evaluate the performance of network coding particularly in terms of this tradeoff – What is the optimal tradeoff – How well do practical protocols achieve it? 13

14. Task 3.2: Coordinated operation of wireless DTNs Main tool: network optimization theory and dynamic flows. Main goal: study the tradeoff of delay with other metrics in a systematic manner. Common approach: – First, find optimum tradeoffs – Then, find good heuristics and compare them with optimum tradeoffs 14

15. Current status of Task 3.2 Delay versus cost tradeoff – i.e., if we wait more, we will transport data with smaller cost [Tasiopoulos et al. 12] Delay versus data volume tradeoff – i.e., if we wait more, we will transport more data [Gitzenis et al. ‘12] Delay versus storage capacity tradeoff – i.e., if we wait more, we need less storage capacity [Iosifidis/Koutsopoulos ‘11] 15

16. Task 3.3 Realistic wireless DTN protocol design Main tool: simulations Aim: develop simulator for simulating the operation of DTNs in the 10,000 node regime Currently available tools: – Generic, i.e., NS2, OMNET, etc. These are not good fits for DTN research, because large delays mean large buffers. – Specialized (for example, ONE), but slow Our approach: fast, dedicated simulator written in C. 16

17. Current status of Task 3.3 Features of our partially built simulator: – Can handle 10,000+ nodes – Uses realistic channel model and includes a realistic slotted MAC protocol – Can handle a variety of mobility models – We have evaluated a variety of well known DTN protocols such as Spray and Wait, GeoDTN+Nav, Geocross, etc. – We have created our own protocol, DTFR (more later) [Sidera ‘11] Main aim: use simulator to test ideas and protocols coming from other workpackages and tasks. 17

18. Part C: Flow Optimization in Delay Tolerant Networks using Dual Decomposition Savvas Gitzenis (Informatics and Telematics Institute, CERTH, Greece) George Konidaris, Stavros Toumpis (Informatics Department, AUEB, Greece) (RAWNET 2012, 18/5/12, Paderborn 18

19. Our work Mobile Wireless DTNs – Topology changes due to node mobility Objective: Single commodity flow optimization Challenge: Flow Optimization is a hard problem in wireless networks even in non-DTN setting Contributions: 1.Fast non-causal centralized algorithm (taking into account the structure of the problem) 2.Heuristic causal centralized algorithms (Heuristic causal decentralized algorithms are subject of future work) 19

20. Network Model (1/2) 20 N nodes 1,2,…,N with set of links A T time epochs 1,2,…, T – Topology (i.e., link properties) remains fixed during each epoch – (following Ferreira ‘02 and others) Traffic flow in epoch t is x(t)={x ij (t), (i,j) in A}, t=1,2,…, T – NB: x(t) describes volume of data, not data rate. x(t) must be inside capacity region R(t) At transition from epoch t-1 to epoch t, node i must have volume of data less than buffer size B i ( t ), i=1,2,…, N, t=2,…, T Internal buffer size vectors B(t)={B i (t), i=1,2,…N}

21. Network Model (2/2) 21 Let y i (t) be the data volume at node i at start of epoch t. Let y(t)={y i (t), i=1,2,…, N} Let z i (t) be the data volume at node i at end of epoch t. Let z(t)={z i (t), i=1,2,…, N} Let input cost function C i (y i (1)), i=1,2,…, N Let utility function U i (z i (T)), i=1,2,…, N Let external buffer size vectors B(1), B(T+1) such that

22. Capacity Region Evolving Graph (CREG) 22 Replica t ↔epoch t Vertices i t, t=1,…,T correspond to node i for different replicas Storage vertices s t i will be used in dual decomposition

23. Problem DTN Utility Maximization (DTNUM) 23

24. A key idea Complexity of problem is dominated by the capacity regions, which are often very hard to describe accurately (e.g., Johansson&Soldati, ’06) – Even simple flow maximization problems can be shown to be NP-complete (e.g., Ephremides&Truong ’90). Therefore, lumping multiple capacity regions in same problem is a bad idea. We will use duality to make sure that we never have to worry about more than one capacity region at a time. 24

25. Solving DTNUM directly 25 Epochs, T Nodes, N Computation Time, T (sec)

26. Solving DTNUM by Dual Decomposition 26 Epochs, T Nodes, N Computation Time, T Computation Time, T (sec)

27. Causal Algorithms Finding optimum involves knowing complete network evolution beforehand. Greedy DTNUM Algorithm: – Do greedy maximization at each epoch – Initial costs are assumed 0, and buffer spaces increase Geographic DTNUM Algorithm: – Greedy optimization takes into account node locations and direction of their movement – May be thought of as generalization of geographic routing 27

28. Performance (a): Optimal algorithm, (b) Geographic algorithm (c): Greedy algorithm, (d): Optimal algorithm, with 1/10 the speed of nodes 28 Total Epochs Volume

29. Part D: On the Cost/Delay Tradeoff of Wireless Delay Tolerant Geographic Routing 29 A.Tasiopoulos*, Ch. Tsiaras $, S. Toumpis* *Informatics Department, Athens University of Economics and Business, Greece $ Department of Informatics, Communication Systems Group, University of Zurich, Switzerland (WOWMOM 2012, 25-28/6/12, San Francisco, CA)

30. Basic Idea In DTNs, there is a tradeoff between packet delay and cost (including transmission and storage cost) Currently, tradeoff appears implicitly in formulations [Juang et al. ‘02, Jain et al. 04, Laoutaris et al. 09, Small et al. 03, etc.] We want to capture this tradeoff formally and explicitly – Under optimal operation – Using practical protocols 30

31. Cost/Delay Evolving Graphs (C/DEGs) Time is divided in epochs The C/DEG is comprised of one subgraph for each epoch Transmission delay is 0. 31

32. Optimal Cost/Delay Curves (OC/DCs) Let two nodes i, j in a network The OC/DC C ij (t) is the minimum cost with which i can send a packet to j with a delay of at most t epochs. To calculate it, we need to find the minimum cost path between node i 0 and the set of nodes i 0, j 1,…, j t – Simple minimum cost path problem – Special structure permits fast calculation 32

33. Example OC/DCs 33

34. Achievable Cost/Delay Curves (AC/DCs) 34 Let two nodes i, j in a network The AC/DC C ij (t) is the minimum cost with which i can send a packet to j with a delay of at most t epochs, assuming optimization over the parameters of the protocol To calculate it, we need simulations

35. Example AC/DCs 35

36. Delay Tolerant Geographic Routing When node S has a packet for node D, it sends it to one of its neighbors using only its local topology and the location of D. Traditionally, there is no delay. Newer approach: wait for topology to change. – MoVe [Lebrun et al. ‘05] – AeroRP [Peters et al. ‘11] – GeOpps [Leontiadis et al ‘07.] – BRR, CR [Tasiopoulos et al. ‘12] 36

37. Rules for selecting next hop 1.MoVe: A selects node that will pass closest to D 2.AeroRP: A selects node that approaches D fastest 3.Min-Cost-per-Progress Rule: minimize 4.Balanced Ratio Rule: minimize 5.Composite Rule: minimize 37

38. 38

39. Part E: Delay Tolerant Firework Routing 39 A.Sidera $, S. Toumpis* $ Department of Electrical and Computer Engineering, University of Cyprus, Cyprus * Department of Informatics, Athens University of Economics and Business, Greece (Med-Hoc-Net 2011, Sicily, Italy)

40. DTFR Operation 1.Homing Phase: travel to estimated location of destination using delay tolerant geographic routing 2.Explosion Phase: create multiple copies 3.Spread Phase: systematically search for destination 4.Lock Phase: do routing in usual sense when in same partition with destination 40

41. Performance 41

42. Analysis of Delay Tolerant Geographic Forwarding Assume mobile node, placed according to spatial Poisson process at any given time. Assume a packet destination at an infinite distance. There is an obvious tradeoff between – Speed v p with which packet moves towards the destination – Transmission cost per distance, c p We find c p (v p ) curve for specific forwarding protocol, under some approximations. 42

43. Bibliography L. R. Ford, Jr. and D. R. Fulkerson, “Constructing maximal dynamic flows from static flows,“ Operations Research, vol. 6, no. 3, pp. 419-433, May-June 1958. D. J. Goodman, J. Borras, N. B. Mandayam, and R. D. Yates, “INFOSTATIONS: A New System Model for Data and Messaging Services,” in Proc. Spring VTC ’97. R. G. Ogier, “Minimum Delay Routing in Continuous-Time Dynamic Networks with Piecewise-Constant Capacities,” in Networks, vol. 18, pp. 303-318, 1988. A. Ephremides and T. V. Truong, “Scheduling broadcasts in multihop radio networks,” in IEEE Trans. on Communications, 1990. A. Vahdat and D. Becker, “Epidemic routing for partially connected ad hoc networks,” Technical Report CS-2000-06, Duke University, 2000. M. Grossglauser and D. N. C. Tse, “Mobility increases the capacity of ad-hoc wireless networks,” in Proc. IEEE INFOCOM, vol. 3, Anchorage, AL, Apr. 2001, pp. 1360-1369. P. Juang, H. Oki, Y. Wang, M. Martonosi, L.S. Peh, and D. Rubenstein, “Energy- efficient computing for wildlife tracking: design tradeoffs and early experiences with Zebranet,” in Proc. ASPLOS_X, Oct. 2002. A. Ferreira, “On models and algorithms for dynamic communication networks: the case of evolving graphs,” in Proc. Algotel 2002. 43

44. S. Jain, K. Fall, and R. Patra, “Routing in a delay tolerant network,” in Proc. ACM SIGCOMM, Portland, OR, Aug.-Sep. 2004, pp. 145-157 N. Laoutaris, G. Smaragdakis, P. Rodriguez, and R. Sundaram, “Delay tolerant bulk transfers on the internet,” in Proc. ACM Sigmetrics 2009, Seattle, WA, June 2009, pp. 229-238. T. Small and Z. J. Haas, “The shared wireless infostation model – a new ad hoc networking paradigm (or where there is a whale, there is a way),” in Proc. ACM MOBIHOC, Annapolis, MD, 2003. R. C. Shah, S. Roy, S. Jain and W. Brunette, “Data MULEs: Modeling and analysis of a three-tier system for sparse sensor networks,” Ad Hoc Networks, vol. 1, no. 2-3, pp. 215-233, Sep. 2003. S. Burleigh, A. Hooke, L. Torgerson, K. Fall, V. Cerf, B. Durst, K. Scott, H. Weiss, “Delay Tolerant Networking: An Approach to Interplanetary Internet,” in IEEE Communications Magazine, June 2003. A. Lindgren, A. Doria and O. Schelen, “Probabilistic routing in intermittently connected networks,” in ACM SIGMOBILE MCCR, vol. 7, Jul. 2003, pp. 19-20. A. Ferreira and A. Jarry, “Complexity of minimum spanning tree in evolving graphs and the minimum-energy broadcast routing problem,” in Proc. WiOpt, Cambridge, UK, Mar. 2004. S. Merugu, M. Ammar, and E. Zegura, “Routing in space and time in networks with predictable mobility,” Georgia Institute of Technology, Tech. Rep. GIT-CC-04-07, 2004, available at http://hdl.handle.net/1853/6492. 44

45. S. Toumpis and A. J. Goldsmith, “Large wireless networks under fading, mobility, and delay constraints,” in Proc. IEEE INFOCOM, Hong Kong, China, Mar.-Apr. 2004. D. G. J. LeBrun, C.-N. Chiah, and M. Zhang, “Knowledge-based opportunistic forwarding in vehicular wireless ad hoc networks,” in Proc. IEEE VTC Spring, vol. 4, Florence, Italy, May-June 2005, pp. 2289-2293. T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Spray and Wait: an efficient routing scheme for intermittently connected mobile networks,” in Proc. ACM WDTN, 2005. J. Burgess, B. Gallager, D. Jensen, B. N. Levine, “MaxProp: Routing for Vehicle- Based Disruption-Tolerant Networks,” in Proc. IEEE Infocom 2006. M. Johansson and P. Soldati, “Mathematical decomposition techniques for distributed cross-layer optimization of data networks,” in IEEE JSAC, Aug. 2006. I. Leontiadis and C. Mascolo, “GeOpps: Geographical opportunistic routing for vehicular networks,” in Proc. IEEE WOWMOM, Helsinki, Finland, June 2007. A. Ferreira, A. Goldman, and J. Monteiro, “Performance evaluation of routing protocols for MANETs with known connectivity patterns using evolving graphs,” Wireless Networks, vol. 16, no. 3, pp. 627–640, Apr. 2010. K. Peters and A. Jabbar, and E. K. Cetinkaya and J. P. G. Sterbenz, “A geographical routing protocol for highly-dynamic aeronautical networks,” in Proc. IEEE WCNC, Cancun, Mexico, Mar. 2011 A. Sidera and S. Toumpis, “DTFR: A geographic routing protocol for wireless delay tolerant networks,” in Proc. Med-Hoc-Net, Favignana Island, Italy, 2011. 45