1. Distributed Channel Assignment in Multi-Radio 802.11 Mesh Networks
Bong Jun Ko (IBM T.J. Watson Research)
Vishal Misra (Columbia University)
Jitendra Padhye (Microsoft Research)
Dan Rubenstein (Columbia University)

2. Wireless Mesh Networks
Objective : Maximize throughput.
WMN: Multi-hop wireless network infrastructure for local/residential area networks.
Goal: better channel utilization  higher network capacity.
For scalability and adaptability, light-weight distributed solution is desirable.

3. Our Philosophy Why focus on channel assignment?
Decouple channel assignment and end-to-end routing.
Routing protocols adapt to dynamic traffic load, link quality, and even channel configuration (e.g., MR-LQSR1) .
Channel assignment focuses on quickly-stabilizing channel configuration based on physical topology.
More scalable than centralized, joint-optimization approaches.
There are K channels, and assume (for now) every node can transmit and receive from all channels simultaneously.
Approach: For each node, minimize the number of other interfering nodes on the same channel. x x
Node x’s interference range
1. R. Draves et al., “Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks”, Mobicom 2004.

4. Distributed Greedy Channel Selection
Let each node select its own channel.
Whenever it is needed, each node changes to a channel that minimizes the number of other nodes on the same channel in the interference range.
Q : Will this process converge? x x y y

5. > Distributed Greedy Channel Selection YES! (+1) (-1)
Let each node select its own channel.
Whenever it is needed, each node changes to a channel that minimizes the number of nodes on the same channel in the interference range.
Q : Will this process converge?
YES!
Proof :
N(x): # of nodes on the same channel for node x.
xN(x) decreases monotonically. x x y
>
(-1)
(+1)
Local optimization improves global optimization metric – Self-stabilizing!

6. Experience with 802.11 Mesh Networks
Practical limitations
Current transceivers can send or receive through only one channel at a time.
Neighboring nodes need to be at the same channel.
Conflicting goals: connectivity vs better utilization.
Multi-radio stations
1 common, default channel for all nodes
Variable channels selected by channel assignment algorithm
Links of variable channels: express way
Links of common channel: local roads

7. Performance Evaluation
Experiments on a 14-node testbed.
A default channel from a
Variable channels from g
Interference range : 3 hops
Routing protocol : MR-LQSR (Multi-Radio Link Quality Source Routing)
Aware of multi-radio, multi-channel environment
Preference given to channel-diverse paths
Measure end-to-end throughput of multiple, concurrent TCP flows with random source-destination pairs
Compare to
samech : all nodes are assigned the same channel for additional radio
rand : each node is assigned a channel uniformly at random for additional radio

8. Testbed

9. Individual TCP Throughput
CDF of all TCP flow throughputs in all experiments.
Flows over longer paths benefit the most.

10. Aggregate TCP Throughput
Measured average TCP throughput of all flows in each experiment, and took median value of 5 experiments.
50% higher than samech / 20% higher than random.

11. Conclusion
Developed a fully-distributed, self-stabilizing channel assignment algorithm for multi-hop wireless networks.
Experiments on multi-radio mesh network testbed.
Our mechanism shows improvements in network throughput by 50% and 20% compared to homogeneous and random assignments, respectively.
Open Problems
Theoretical running time and bounds of the distributed greedy algorithm?
Formal time-scale decomposition in radio resource control (e.g., channel, power, rate, route control).

12. Thank You

13. Backup slides

14. Other Results Channel Utilization (in %) of 802.11g channels samech
rand DA
10.1
15.1
22.7
Protocol Dynamics
# msgs
90.0
Bytes
2080
Time (sec)
32.4
Changes
0.22
Requests
0.70

15. Dealing with Delay and Asynchrony
Solution : a 3-way handshake protocol for distributed mutual exclusive operation.
REQUEST → ACCEPT or REJECT → UPDATE or ABORT

16. 3-way Handshake Protocolx

17. 3-way Handshake Protocolx
REQUEST
REQUEST specifies:
Intended channel change
Perceived channels of other nodes

18. 3-way Handshake Protocolx
ACCEPT

19. 3-way Handshake Protocolx
UPDATE
When a node ACCEPTed a REQUEST, it “freezes” its channel until corresponding response (UPDATE or ABORT) is received.

20. 3-way Handshake Protocolx

21. 3-way Handshake Protocolx
REQUEST

22. 3-way Handshake Protocolx
REJECT

23. 3-way Handshake Protocolx
ABORT

24. 3-way Handshake Protocol
REQUEST y x
REQUEST

25. 3-way Handshake Protocol
REJECT y x
ACCEPT
Break ties by predefined order of nodes
- if x < y, y will be accepted to change.

26. 3-way Handshake Protocol
UPDATE y x
ABORT
Show how to Resolve the livelock.