1. IEEE Infocom 2007 On a Routing Problem within Probabilistic Graphs and its application to Intermittently Connected Networks Joy Ghosh Hung Q. Ngo, Seokhoon Yoon, Chunming Qiao Messenger Server Department of Computer Science and Engineering, Yahoo! Inc. State University of New York at Buffalo, Sunnyvale, CA Buffalo, NY

2. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO In many networks such as MANET, DTN, ICMAN,..., links are probabilistic. Natural to formulate the following problem. A Routing Problem on Probabilistic Graphs s t pepe G: directed graph For each e=(u,v), p e is the probablity that u can deliver a packet to v All p e are independent Find a subgraph H maximizing Conn-Prob H (s,t) Subject to application-dependent constraints Find a subgraph H maximizing Conn-Prob H (s,t) Subject to application-dependent constraints

3. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Natural Questions That Follow (Talk Outline) 1. How do we construct G (and compute p e )?  Accuracy of p e  routing efficiency 2. What are the constraints for H? 3. Given H, how to compute Conn-Prob H (s,t)? 4. What’s the complexity of finding optimal H? 5. If complexity is too high, how to design good routing algorithms/protocols? 6. How useful is this model, anyhow?

4. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO 1. How do we construct G (and compute p e )? Short answer: application dependent Long answer:  Don’t care (e.g. Epidemic, randomized flooding, Spray-and-Wait, …)  Locally estimate delivery predictability/frequencies (e.g., ProPHET, ZebraNet, …)  Assume an Oracle (e.g., Spyropoulos et al. in Secon’04, Jain et al. Sigcomm’05, …)  Mobility modeling/profiling (e.g., random waypoint, group mobility, freeway mobility, …) We use random orbit model from our SOLAR frameworkrandom orbit model Main reasons: model built on real-world data traces, and we already have the simulation code for it

5. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO 2. Constraints for the delivery subgraph H Short answer: application dependent Long answer:  No constraint (blind flooding)  Acyclic (Epidemic)  Threshold on expected delivery time (e.g. some works on DTN) ……  Our proposal: |E(H)| ≤ given threshold k, because this will reduce Contention, thus message drops and retransmissions Data and bandwidth overheads

6. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO 3. Given H, how to compute Conn-Prob H (s,t)? Bad news: #P-complete  Classic s,t-reliability problem  Shown #P-complete by Valiant (1979) Good news:  Can be approximated with our simple and efficient heuristics to within about 85%-90% accuracy on averageour simple and efficient heuristics Note:  May have an FPRAS using Markov-chain Monte Carlo (along the line of Jerrum-Sinclair-Vigoda’s work, Journal of the ACM, 2004)  But: Long Standing Open Problem

7. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO 4. What’s the complexity of finding optimal H? Somewhat subtle: hard to compute objective function does not imply hard to optimize Bad News: #P-hard, as we showed in the paper Good News:  Our heuristic can find reasonably good H Note:  A wide-open research direction: approximation algorithms for #P-optimization problems.

8. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO #P-hardness of finding optimal H Given directed graph D = (V,E) with edge probabilities p e, source s, destination t; computing Conn-Prob D (s,t) is #P-complete in 1979 Let c be the least common multiple of all denominators of the p e We show that: a) If the optimal H can be found for any given G, then an efficient decision procedure for deciding if Conn-Prob D (s,t) ≤ c’/c can be designed for any c’ ≤ c. b) If such a decision procedure exists, then we can compute Conn-Prob D (s,t) with a simple binary search

9. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO #P-hardness of finding optimal H Add path with k=|E(D)| edges to D to get G with Π k i=1 p i = c’/c + ε, for any ε < 1/c Suppose finding optimal H can be done efficiently If H is the  Upper part  Conn-Prob D (s,t) > c’/c  Lower part  Conn-Prob D (s,t) ≤ c’/c

10. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO 5. Heuristic for finding good H (i.e. routing algo) Edge-constrained routing EC-SOLAR-KSP

11. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Simulation Parameters

12. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Edge-constrained routing – EC-SOLAR-KSP  EC-SOLAR-KSP1  L = |E|  EC-SOLAR-KSP2  L = 0.8 * |E|  EC-SOLAR-KSP3  L = 0.6 * |E|  PROB-ROUTE  P init = β = γ = 0.5; A. Lindgren, A. Doria, and O. Schelen, “Poster: Probabilistic routing in intermittently connected networks,” Proceedings of The Fourth ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc 2003), June 2003.  EPIDEMIC  Not shown in 2 nd figure as its Network Byte Overhead was much higher A. Vahdat and D. Becker, “Epidemic routing for partially connected ad hoc networks,” Technical Report CS- 00006, Duke University, April 2000.

13. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO 6. How useful is this model, anyhow? Short answer: we don’t know yet Long answer:  The good This model is potentially useful There are many interesting open questions Good solution can serve as benchmark  The bad and the ugly Edge probabilities may not be independent Practical applications can’t afford to use a centralized algorithm Intermediate vertices may have more information than the source at the point they are about to deliver packets  This may be a good thing, may be a bad thing!

14. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Concluding Remarks Contributions  Formulatied of a new routing problem within probabilistic graphs  Addressed several aspects of the problem: graph construction, complexity, routing heuristics Open problems  Approximation algorithm for this #P-Hard problem  Better distributive algorithm  Better Heuristics for this and other mobility models

15. IEEE Infocom 2007 Thank You! Questions?

16. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Mobile Users influenced by social routines visit a few “hubs” / places (outdoor/indoor) regularly “orbit” around (fine to coarse grained) hubs at several levels Sociological Orbit Framework

17. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Illustration of A Random Orbit Model (Random Waypoint + Corridor Path) Conference Track 1 Conference Track 3 Cafeteria Lounge Conference Track 2 Conference Track 4 Posters Registration Exhibits

18. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Random Orbit Model

19. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO How to compute model’s parameters? ETH Zurich traces  1 year from 4/1/04 till 3/31/05  13,620 wireless users, 391 APs, 43 buildings Mapped APs into buildings based on AP’s coordinates, and each building becomes a “hub”  Converted AP-based traces into hub-based traces Real-world Data Traces Hub-Lists (Binary Vectors) Users’ Mobility Profiles Hub Transition Probabilities Hub Staying Time Distributions Hub Transition Time Distributionss Model Trained Using EM-Algorithm Each Profile is a weighted Hublist, e.g. Profile = (0.4, 0.5, 0.9) Each profile is a cluster mean obtained via the Expectation Maximization (EM) Algorithm

20. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Examples of Hub-Lists and Mobility Profiles On any given day, a user may regularly visit a small number of “hubs” (e.g., locations A and B) Each mobility profile is a weighted list of hubs, where weight = hub visit probability (e.g., 70% A and 50% B) In any given period (e.g., week), a user may follow a few such “mobility profiles” (e.g., P1 and P2) Each profile is in turn associated with a (daily) probability (e.g., 60% P1 and 40% P2) Example: P1={A=0.7, B=0.5} and P2={B=0.9, C=0.6}  On an ordinary day, a user may go to locations A, B and C with the following probabilities, resp.: 0.42 (=0.6x0.7), 0.66 (= 0.6x0.5 + 0.4+0.9) and 0.24 (=0.4x0.6)  20% more accurate than simple visit-frequency based prediction  Knowing exactly which profile a user will follow on a given day can result in even more accurate prediction On any given day, a user may regularly visit a small number of “hubs” (e.g., locations A and B) Each mobility profile is a weighted list of hubs, where weight = hub visit probability (e.g., 70% A and 50% B) In any given period (e.g., week), a user may follow a few such “mobility profiles” (e.g., P1 and P2) Each profile is in turn associated with a (daily) probability (e.g., 60% P1 and 40% P2) Example: P1={A=0.7, B=0.5} and P2={B=0.9, C=0.6} On an ordinary day, a user may go to locations A, B & C with the following probabilities: 0.42 (=0.6x0.7), 0.66 (= 0.6x0.5 + 0.4+0.9), 0.24 (=0.4x0.6) 20% more accurate than simple visit-frequency based prediction Knowing exactly which profile a user will follow on a given day can result in even more accurate prediction

21. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Profiling illustration Obtain daily hub stay durations Translate to binary hub visitation vectors Apply clustering algorithm to find mixture of profiles

22. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Profile parameters for all sample users

23. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Approximation algorithm for Conn-Prob G (s,d) G = (V, E) where edge probability between nodes u and v is p e (u,v) In G, starting from s, all nodes choose at most k downstream edges to get G k = (V, E k )  (b) Weight of each edge in G k is set to  w e (u,v) = -1 * log (p e (u,v)) to get G’ k say

24. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Approximation algorithm for Conn-Prob G (s,d) Compute shortest path from s to all nodes in G’ k to get G sp = (V, E sp ) & assign BFS level Reset w e (u,v) = p e (u,v) & add all edges (v,d) that were in G to get G’ = (V, E’)  (d) Let P d (u,v) be delivery probability of node u to v Apply Approximation Algorithm to G’ to get P d (s,d)  Start with any u ≠ d with maximum level #  P d (u,d) = 1 – Π k 1 (1 – p i )  Where p i = w e (u,v i ) * P d (v i, d) for all edges (u,v i )

25. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Approximation algorithm for Conn-Prob G (s,d)

26. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Optimal algorithm for delivery probability Calculate all paths from s to d Apply Algorithm 2 by rules of inclusion and exclusion

27. IEEE Infocom 2007 State University of New York (SUNY) at BUFFALO Approximation ratio simulation