1. 1 11 Distributed Channel Assignment in Multi-Radio 802.11 Mesh Networks Bong-Jun Ko, Vishal Misra, Jitendra Padhye and Dan Rubenstein Columbia University Microsoft Research WCNC 2007

2. 2 22 Outline Introduction Related Work Architecture and Model Distributed Channel Assignment Performance Evaluation Conclusion

3. 3 Introduction (1) In multi-hop wireless networks, the management of radio resources has a tremendous impact on the performance of the entire system e.g., transmission power control, frequency channel selection, routing, etc. Managing one resource type greatly impacts the management of other resources it requires the global status to be considered We focus on the wireless channel assignment problem in multi-hop wireless networks with multi-radio stations

4. 4 Introduction (2) What information should the channel selection be based on? The channels should be selected based only on locally available information The assignment of the channel should be based on the physical structure of the network quickly-changing component is handled by routing slowly-changing component is handled by channel assignment The change in channel assignment should not frequently alter the connectivity between nodes providing a stable channel environment for the end-to-end routing mechanism Our goal is to provide a diverse and quickly-stabilizing channel configuration

5. 5 Mesh Network Architecture Our focus is on how to utilize the 802.11 channels within the wireless network of mesh routers ignore the mesh clients and mesh gateways

6. 6 Related Work [11] propose a distributed algorithm that utilize only local traffic load information to dynamically assign channels and to route packets it works only for routers whose connectivity graph is a tree In [10], a central server that periodically collects dynamically-changing channel interference information it takes into account dynamically changing network status, while our channel assignment is based on more static information the performance gain of our mechanism observed in the real- world testbed experiment appears similar to what is shown in their simulation results

7. 7 System Model N nodes K wireless channels possibly overlap f(a, b): channel interference cost function provides a measure of the spectral overlapping level between channels a and b f(a, b) ≥ 0 and f(a, b) = f(b, a) δ tunable parameter S j : node j’s interference set all other nodes within three hops of node j are in node j’s interference set

8. 8 Channel Selection Algorithm Intuitively, a node would like to choose a channel upon which its transmissions are least likely to suffer interference from other senders’ transmissions

9. 9 Theorem 1 If every node selects its channel following Algorithm 1, within a finite number of channel changes by nodes, the channel assignment reaches a stable state where nodes cease changing channels

10. 10 Proof of Theorem 1 (1) Notation c i and c i’ be node i’s channel at time t and t’ respectively assume i is the only node that changes the channel between time t and t’., the sum of the interference levels for all nodes before and after node i’s channel change, respectively We will show that F’ < F

11. 11 Proof of Theorem 1 (2) Each node changes its channel only when it can decrease the interference level Therefore, for the changing node i, For each node j in S i, For all other nodes,

12. 12 Proof of Theorem 1 (3) the last inequality holds because The above inequality means F decreases monotonically whenever a node changes its channel F cannot decrease indefinitely and must stop decreasing after finite number of steps of nodes’ channel changes hence Algorithm 1 stabilizes

13. 13 Remark Each node’s greedy choice to improve its local objective results in the improvement in global objective of total interference level also eventually leads to a channel assignment in which all nodes are satisfied with their channel choice

14. 14 Applying to 802.11 Multi-radio Network We strive to find a good balance between the channel diversity and the network connectivity with the following assignment rules One interface of each node is dedicated to a default channel common to all nodes ensures the connectivity The assignments to each of the remaining interfaces are performed using Algorithm 1, with one exception the selected channel must be one of those already assigned to some neighbors within communication range  prevents the interface from being assigned a useless channel

15. 15 Network Testbed 14 nodes each node is equipped two 802.11 interfaces one from 2.4GHz 802.11g band, and the other from 5GHz 802.11a band default channel: 802.11a 14 concurrent TCP flows The destination of each TCP flow is chosen at random We avoided choosing single-hop destinations We generated four different sets of such TCP flows, with each set denoted by fset1, fset2, fset3, and fset4, respectively

16. 16 Comparison We consider three baseline channel assignment strategies to compare with our assignment samech all nodes are assigned the same 11g channel 11-rand each node is assigned one of 11 802.11g channels selected uniformly at random  corresponds to the case of δ = 0 3-rand each node is assigned one of three orthogonal 802.11g channels (i.e., channel 1, 6, and 11) selected uniformly at random

17. 17 Cumulative Distribution : Throughput of Individual TCP Flows

18. 18 Median Network Throughput

19. 19 Conclusions We presented a fully-distributed mechanism that assigns 802.11 channels to multi-radio nodes in wireless mesh networks Our assignment mechanism stabilizes to a desirable channel configuration it strikes a good balance between network connectivity and channel diversity It is sufficiently light-weight to be executed on large scale mesh networks We run experiments on our wireless mesh network testbed show that our channel assignment can increase the capacity between 20% ~ 50% over conventional mechanisms