1. LCN 2007, Dublin 1 Non-bifurcated Routing in Wireless Multi- hop Mesh Networks by Abdullah-Al Mahmood and Ehab S. Elmallah Department of Computing Science University of Alberta Ahmed Kamal Department of Electrical and Computer Engineering Iowa State University Research supported by NSERC

2. LCN 2007, Dublin 2 Outline Introduction Problem Formulation Solution Approach Some Related Work Simulation Results Concluding Remarks

3. LCN 2007, Dublin 3 Introduction General Objectives  Fixed wireless broadband access  Support applications with different service requirements (e.g., high data rate, low delay jitter etc.)

4. LCN 2007, Dublin 4 Introduction Mesh BS Mesh BS (y) Mesh BS (x) Mesh BS Internet Gateway Subscribers d1(x)d1(x) d2(x)d2(x) d |D(x)| (x) … Demand D(x)

5. LCN 2007, Dublin 5 Introduction Mesh BS Mesh BS (y) Mesh BS (x) Mesh BS Internet Gateway Subscribers s1(x)s1(x) s2(x)s2(x) s |D(x)| (x) … Accepted Demand S(x) flow f(x, y)

6. LCN 2007, Dublin 6 Problem with Splitting Flows Example: Routing streaming data End user experiences poor stream quality or unusual delay Packets 1,2, … 7 3,4 1,2,5,6,7

7. LCN 2007, Dublin 7  Non-bifurcated Flows A sequence of uniquely identifiable packets Indivisibly follows the same path without rerouting  Routers do not need synchronization  Interference follows protocol model (conforming to RTS-CTS-DATA-ACK sequence in IEEE 802.11 family of standards) System Model

8. LCN 2007, Dublin 8 Problem Formulation with Single Channel Notations  f(X,Y): aggregate flow between sets of routers X and Y  f(D(x)), f(S(x)): sum of flow values in the vectors D(x) and S(x) respectively  E int T (x): set of edges having one end within interference range of x  f(E int T (x)): sum of flow values along edges in E int T (x)  C(x): Available channel capacity at router x

9. LCN 2007, Dublin 9 Problem Formulation with Single Channel Objective: Subject to: Channel capacity constraint: l(x) = f(x, V) + f(V, x) + E int T (x) Maximizef(V, GW) sum of all flows from V to gateway x f(V, x)f(x, V) E int T (x ) C(x)C(x) ≤

10. LCN 2007, Dublin 10 Flow conservation constraint: Flow indivisibility constraint: d i (x) Î D(x) is assigned a single route from x to a gateway in G Problem Formulation with Single Channel f(x, V) Outgoing Flows f(V, x) + f(S(x)) Incoming Flows and Accepted Flows = g y x di(x)di(x) di(x)di(x) di(x)di(x)

11. LCN 2007, Dublin 11 Remarks We seek assignment of flows to edges Implementation: we use source routing in order to realize a set of computed flows Challenges  Achieved flow values = planned flows?  Any improvement over a sophisticated (ad-hoc) routing protocol (e.g. DSR)?

12. LCN 2007, Dublin 12 Solution Approach Flow augmenting paths (FAP) are used in classical network flow problems Example: Our problem under certain constraints is NP-complete Traditional FAPs do not work for our problem sbat sbat 3/84/4 1/9 7/80/4 5/9

13. LCN 2007, Dublin 13 Solution Approach We propose the use of Interference Constrained FAPs (IC-FAP) IC-FAPs take into account interference in system model Challenges  Finding IC-FAPs efficiently (i.e., in polynomial time?)  Suitability of using IC-FAPs for finding near optimal solutions (i.e. Are IC-FAPs sufficient?)

14. LCN 2007, Dublin 14 IC-FAP: An Example

15. LCN 2007, Dublin 15 An IC-FAP Search Heuristic Idea  Forward to nodes closer to gateways  Keep track of interference and consider alternative paths

16. LCN 2007, Dublin 16 An Example j kih dfe a bc g 1 1 1 1 (3,5):  g  (4,4):  g,f  (4,0):  g,f  (0,0):  g,f,c,b  (3,0):  g,f,c 

17. LCN 2007, Dublin 17 Some Related Work [Draves et al.: MobiCom 2004]: “Routing in multi-radio, multi-hop wireless mesh networks”  Proposed a metric based on transmission delays  Chooses channel that is likely to decrease delay

18. LCN 2007, Dublin 18 Some Related Work [Raniwala and Chiueh.: INFOCOM 2005]: “Architecture and algorithms for an IEEE 802.11-based multi-channel wireless mesh network”  A heuristic solution  Three stages : tree construction, node to interface binding and interface to channel binding  Key idea is balancing traffic load

19. LCN 2007, Dublin 19 Some Related Work [Kodialam and Nandagopal: MobiCom 2005]: “Characterizing the capacity region in multi-radio multi- channel wireless mesh networks”  Models the problem as a linear program  Provides a framework for estimation of capacity region  Flows are allowed to split among multiple paths  Assumes synchronous operation of routers

20. LCN 2007, Dublin 20 Some Related Work [Alicherry et al.: Journal on Selected Areas of communication 2006]: “Joint channel assignment and routing for throughput optimization in multiradio wireless mesh networks”  Assumes a synchronous model  Formulates solution as a linear programming relaxation  The algorithm works in stages of solution refinement  The final solution allows bifurcation of flows

21. LCN 2007, Dublin 21 Simulation Results Topology WMN Parameters Channel Capacity 100 units Radio Range of Mesh Router 112 m Radio Range of Subscriber Units 29 m Maximum Subscriber per Router 10 Flow Demand per Subscriber 1 unit Traffic Parameters 1 Unit of flow 40 Kbps (Application Data) Application-level Packet Size Uniform: [200, 300] bytes Packet Inter-arrival Time Uniform: [30, 50] ms

22. LCN 2007, Dublin 22  Findings: The average throughput is higher at different traffic loads. Simulation Results Comparison with DSR : Average throughput

23. LCN 2007, Dublin 23  Findings: The minimum throughput is also higher at different traffic loads. Simulation Results Comparison with DSR : Minimum throughput

24. LCN 2007, Dublin 24  Findings: The delay jitter is comparatively less Simulation Results Comparison with DSR : Delay Jitter

25. LCN 2007, Dublin 25 Simulation Results: Insights One source of improvement from DSR results from route stability DSR is not designed for (max-min) fairness

26. LCN 2007, Dublin 26 Concluding Remarks The