1. Presented by Tae-Seok Kim
Joint Channel Assignment and Routing for Throughput Optimization in Multiradio Wireless Mesh Networks
Mansoor Alicherry
Randeep Bhatia
Li (Erran) Li
Bell Laboratories,
Lucent Technologies
Presented by Tae-Seok Kim

2. Wireless LAN/MAN Evolution
Internet/WAN
Wireless routers with gateway functionality interface with wired network
Gateway
Internet/WAN
Wired backbone
Wireless Mesh
backbone
Wireless routers form a self organized & self configured wireless mesh backbone
Multi-hop wireless routing
Only last hop is wireless
Wireless Mesh Network (WMN)
Traditional wireless networks
Access points connected to a wired backbone

3. WMN performance issues
Internet
Active
Inactive
Interference limits capacity of wireless mesh backbone
Performance degrades as network size increases
Throughput degrades as the number of wireless hop increases.
Performance improvement options: Multi-channel, Multi-radio, Routing

4. WMN single radio nodes and multiple channels
Internet
Active 1 2
When used with single radio nodes may result in network partitioning.
One partition per channel
Multiple channels can be used to lower interference
More simultaneous transmissions are possible

5. WMN multiple radio nodes and multiple channels
Internet 3
Active 1
4 channels
2 radios per node 2 4
If enough node-radios and enough channels then interference may be completely eliminated
In practice small number of avail. channels and few radios per node
Interference still an issue
Channel assignment is crucial

6. Channel assignment and routing go hand in hand
Internet
Routing Paths
4 channels
2 radios per node
Internet 3 1 2 4 3 4 1 2
Routing Paths
Channel assignment and routing paths must be jointly computed.
Channel assignment and routing are inter-dependent.
channel assignment: link bandwidths and interfere extent
traffic routing: the traffic flows for each link
Channel assignments need to be done in a way such that the communication requirements for the links can be met.

7. WMN design problem: Joint channel assignment and routing to maximize throughput
Given:
A WMN G= (V,E)
K orthogonal channels
Number of radios I(u) per node u.
Traffic load l(u) for each node u directed for the internet.
Find:
A channel assignment for node radios (from the set 1..K)
A set of packet routes over the WMN backbone and their associated flow (traffic)
A fair throughput measure .
Such that:
An interference free communication schedule can route l(u) traffic for each node u
A periodic (with period T) time slotted schedule is assumed.
is maximized

8. Wireless model Interference Range Defines interfering links in E
Communication Link
Defines communication links (e.g. )
Communication Range
Link
Interferes with link
Does not interfere with link
Interfere link set for
( , ) is not a communication link
Link e has the maximum rate of c(e)
May depend on distance

9. Steps for WMN design problem
Routing: Solving LP
Channel Assignment
Flow Scaling
Interference Free Link Scheduling

10. Scheduling (given the routing and channel assignment)
Internet
Internet
2 channels 1 2
Necessary condition:
(e.g. c(2) = 8)
c(q)- approximation algo. for scheduling
Sufficient condition:
Scheduling problem is NP-hard even for 1 channel

11. Scheduling necessary condition continued
To show: In a given slot of an interference free link schedule either link e is active or at most c(q) links from can be simultaneously active
Do not interfere

12. Scheduling necessary condition continued
How many nodes all apart can be located in a circle of radius
How many non-overlapping circles of radius can be packed in a
circle of radius
The answer is c(q) (e.g. c(2) = 8)

13. Routing Problem: LP formulation (May not result in feasible channel assignment)
Internet
4 channels
2 radios per node
l()=2
l()=0
l(): node traffic load
Each link of capacity c(e)
Let q=2, c(q) = 8
Find maximum st. all incoming node flow can be routed to the internet node:
Flow conservation constraints
Scheduling necessary condition
Node radio constraints
Capacity constraints

14. Channel Assignment Algorithm Phase 1
4 channels
2 radios per node
[a]
[b]
[c]
[d]
Internet
[a]
[b]
[d]
[c]
Internet
Each f(e,i) = 1/4 1 1 1 1
Basic Flow Transformation Step: For an link e move flow f(e,i) for channel i to flow f(e,j) for channel j.
Normalize (by splitting) nodes to have approx. same number of radios I (I=2 here)
Using only first I channels, transform flows such a way that the network has a large number of connected components with small intra-component interference
Interference of a connected component A:
At end of the phase channel assignment is feasible however some constraints may be violated

15. Channel Assignment Algorithm Phase 2
Internet
[a]
[b]
[d]
[c] 1
4 channels
2 radios per node
[a]
[b]
[c]
[d]
[a]
[b]
[c]
[d]
Group into at most K (=4) groups the connected components in all the channels (first I (=2) channels):
A group can only contain connected components from the same channel
Greedily merged such that the merging causes the least increase in max group interference
Assign links in the i-th group to channel i
Scale flow (and ) to satisfy scheduling and node constraints
At end of the phase channel assignment is still feasible, all constraints satisfied by flow scaling

16. The overall algorithm
Theorem: The overall algorithm (scheduling, routing and channel assignment) is a Kc(q)/I approximation algorithm for the joint routing and channel assignment problem to maximize throughput.
K is number of available channels
I is minimum number of radios per node
c(q) depends on the interference model parameter q ( ). For instance c(2)=8.

17. Simulation Setup 802.11a radios 60 nodes networks
Link rates distance dependent (Max. 54 Mbps)
Max. num of available channels = 12 channels
60 nodes networks
Grid networks: 8X8 grid, Grid size , Random node distribution on grid points
Random connected networks: Nodes placed randomly in 500mX500m area (9 random connected topologies )
20 randomly chosen nodes have load of 20Mbps each
Vary number of gateway nodes from 2 to 12
Vary number of radios per node from 1 to 4 (equal per node)
Vary number of available channels from 1 to 12

18. Varying the number of available channels
As the number of available channels increase the per node throughput generally increases

19. Varying the number of radios per node & number of gateway nodes
As the number of radios per node increase the per node throughput generally increases (the number of channels is 12)
Highest increase in going from 1 to 2 radios per node
As the number of gateways increase the per node throughput generally increases

20. Algorithm performance compared with the worst case approximation bound
Upper Bound: LP optimal value(may not be feasible) λ* # of gateways: 8, # of radios: 3, # of channels: 12
Worst Case Bound: Approximation ratio performance bound (is feasible) λ*l(v)/W, where W=Kc(q)/I
Proposed Algorithm
At least 5.3 and 8.3 times better than Worst Case Bound
At most 4.0 and 2.4 times worse than the Upper Bound

21. Conclusion
Multiple radios and multiple channels can be used to alleviate interference in WMNs
Channel assignment and routing is key
Designed an efficient algorithm for joint channel assignment and routing for WMN
The algorithm has a constant approximation bound
Is able to make use of multiple radios/channels for increased per node throughput
Future work
Dynamic Channel Assignment
Routing based on distributed routing protocols (e.g. OSPF)