TOUR: Time-sensitive Opportunistic Utility- based Routing in Delay Tolerant Networks

Transcription

1 TEMPLE UNIVERSITY TOUR: Time-sensitive Opportunistic Utility- based Routing in Delay Tolerant Networks Mingjun Xiao a,b, Jie Wu b, Cong Liu c, and Liusheng Huang a a University of Science and Technology of China, China b Temple University, USA c Sun Yat-sen University, China

2 Outline Introduction on utility-based routing Motivation Problem Solution Simulation Conclusion

3 Introduction: utility-based routing Concept : Utility-based routing [Jiewu 08, 12] Utility is a composite metric Utility (u) ) = Benefit (b) Cost (c) Benefit is a reward for a routing Cost is the total transmission cost for the routing Benefit and cost are uniformed as the same unit Objective is to maximize the utility of a routing

4 Introduction: utility-based routing Motivation of Utility-based Routing Valuable message: route (more reliable, costs more) Regular message: route (less reliable, costs less) message sender route 1 route 2 route k receiver Benefit is the successful delivery reward

5 Motivation Utility-based routing Delay Tolerant Network (DTN) delivery delay is an important factor for the routing design Time-sensitive utility-based routing

6 Time-sensitive utility model Benefit: a linearly decreasing reward over time Utility: u u( t) u s =b b( t) = b( t) δ c s { b t δ, = 0, c u i u(t) t b δ t > b δ b(t) s i d u d t

7 Problem Time-sensitive utility-based routing in DTN DTN: V={1, 2, }, λ i, j, c i (i, j V) source s, destination d, initial benefit b, benefit decay coefficient δ (single copy) Objective: maximizee[ E[u d ] or minimized s (u s )=b-e[ -E[u d ] for generality, minimize D i (u) u s =b δ c s u i { Di(u) u = E[ ] -ud t s i d u ud

8 A simple example Problem DTN: V={1, 2, 3, d}, λ i, j, c i (i, j V) source s=3 =3, destination d, initial benefit b=20 =20, benefit decay coefficient δ=2 Objective: minimize D 3 (u 3 ) λ 3,1=0.8 c 3 = 4 c1= 8 1 λ 1, d= 0.8 λ 3,d = d λ 3,2= c 2 = 4 λ 2, d= 0.2

9 Problem The key problem when a node i meets another node, whether the node i should forward messages to this encountered node, or ignore this forwarding opportunity, so that the node i can achieve the minimum D i (u)

10 Solution Basic idea: Time-Sensitive Opportunistic Forwarding Dynamically select relays: forwarding set R i (u) Opportunistic forwarding scheme: only forward messages to nodes in forwarding sets; ignore the other nodes outside of the set i d R i (u)

11 Solution Basic idea: Time-Sensitive Opportunistic Forwarding Forwarding set R i (u) is time-sensitive: vary with time, i.e., remaining utility u λ 3,1 =0.8 c 3 =4 c1=8 1 λ 1,d =0.8 λ 3,d = d λ 3, 2=0.2 2 c 2 =4 λ2,d=0.2 R*(u) 3 {d, 1, 2} {d, 2} {d} φ tim e u

12 Solution Determine optimal forwarding set Computation formula R i * (μ) R u) = arg min D ( u) i ( i R( u) R( u) N µ D ( u µ ) = ρ ( u)( µ u + D ( u )) du + i = i R( u) i, j j j j f 0 j R ( u) p ( µ ) µ i j d successful forwarding failed forwarding μ-u j D j (u j )

13 Solution Determine optimal forwarding set For a single node i : R i * (μ) Assumption: D j (μ-c i )=D j (u j ) are known D 1 (μ-c 1 )<D 2 (μ-c 2 )< <D m (μ-c m ) Method: : Greedily compute R i * (μ) R i * (μ): 1, 2,, k, k+1,, m Correctness: : Theorem 1 λ 3,1 =0.8 c 3 =4 λ 1,d = 0.8 λ 3,d = d λ 3, 2= 0.2 c1=8 1 2 c 2 =4 λ 2,d =0.2 D 3 ( μ ) D d ( 6 ) = 0 D 2 ( 6 ) = 5. 8 { d } D 1 ( 6 ) = 6 R * 3 ( μ = 1 0 ) { d, 2 } { d, 1, 2 }

14 Solution Determine optimal forwarding set For all nodes i V: R i *(μ) Method: : iteratively compute R i * (μ) for all i V V -1 rounds of computation Convergence: : Theorem 2 u R*(u) 3 (4, 2 0 ] {d} (0, 4] φ 1 3 d 2 u R*(u) 1 (8, 2 0 ] {d} (0, 8] φ u R*(u) 2 (4, 2 0 ] {d} (0, 4] φ u R*(u) 1 (8, 2 0 ] {d} u R 3 *(u) 1 (0, 8] φ (1 2, 2 0 ] {d,1,2} (8, 1 2 ] {d,2} (4, 8] {d} 3 d (0, 4] φ Round 1 Round 2 2 u R*(u) 2 (4, 2 0 ] {d} (0, 4] φ

15 Implementation Discrete Process D i (u) ~ ( u ) Di ~ D i (u) D i (u) μ 0 μ μ k μ n u R i (u) ~ ( u ) Ri μ 0 Estimation e rror ~ R(u) R(u) μ k μ μ n u

16 Discrete Process Implementation µ D ( µ ) ρ ( u)( µ u + D ( u )) du + i = R( u ) i, j j j j f 0 j R( u ) p ( µ ) µ µ ~ ~ D ( µ ) ~ ~ ρ ( u)( µ u + D ( u )) du + i R ( u ) = ~ 0 j R ( u ) i, j j j j ~ p f ( µ ) µ Theorem 3 gives the upper bound of estimation error of the discrete process

17 Real trace used Simulation Cambridge Haggle Trace Trace Contacts Length (d.h:m.s) Routing nodes External nodes Intel 2, : Cambridge 6, : infocom 28, : UMassDieselNet Trace 40 buses 55 days, Spring 2006

18 Simulation Algorithms in comparison TOUR (10 discrete sampling points) TOUR-OPT (100 discrete sampling points) SimpleUtility, MinDelay, MinCost Metrics Remaining utility Derivation Cost

19 Simulation Settings Parameter name Default Range Initial benefit Maximum forwarding cost Benefit decay coefficient Number of messages 30,000

20 Simulation Results Remaining utility vs. initial benefit, benefit decay coefficient, maximum forwarding cost

21 Simulation Results Derivation vs. initial benefit, benefit decay coefficient, maximum forwarding cost

22 Simulation Results Remaining utility vs. initial benefit and benefit decay coefficient

23 Conclusion Our proposed algorithm outperforms the other compared algorithms in utility. The larger the initial benefit and the smaller the benefit decay coefficient are, the larger the remaining utility would be. Our proposed algorithm can schedule different message deliveries to different paths.

24 Thanks! Q&A