1. Architecture and Algorithms for an IEEE 802
Architecture and Algorithms for an IEEE based Multi-channel Wireless Mesh Network
Ashish Raniwala
In collaboration with Prof Tzi-cker Chiueh
Experimental Computer Systems Lab
Stony Brook University

2. Today’s Enterprise Wireless Networks
Motivation
Today’s Enterprise Wireless Networks
Examples: Bell Labs Holmdel, SUNY campus, Hilton
IEEE based Wireless Access
Wired Backbone (deployment, management)
Wired Backbone

3. Potential Solution: Wireless Mesh Network
Motivation
Potential Solution: Wireless Mesh Network
Multi-hop ad hoc network of wireless routers
Based on “Mobile Ad hoc Network” Architecture
Single-channel => Link interference => Low capacity
Wireless
Backbone

4. Hyacinth Goals Motivation
High-capacity Wireless Mesh => Multiple channels
Off-the-shelf hardware => No MAC modifications
Multi-channel
Wireless
Backbone

5. Hyacinth Architecture Research Issues
Outline
Motivation
Hyacinth Architecture
Research Issues
Load-balancing Routing
Traffic-aware Channel Assignment
Throughput and Latency Gains
Hyacinth Prototype
Conclusions
Other Mesh Projects

6. Interconnection Network ?
Network Model
Internet
NFS
ERP
Enterprise Resources
Gateways
Interconnection Network ?
Access Net

7. Hyacinth Architecture
Internet
NFS
ERP
Enterprise Resources
Gateways
Wireless
Backbone
Access Net

8. Hyacinth Architecture
Internet
NFS
ERP
Enterprise Resources
Gateways
Wireless
Backbone
Access Net

9. Hyacinth Architecture
Wired Network

10. Hyacinth Architecture
Wired Network 4 3 3 1 2 5 3 1 2 4 5
Virtual link
operating on
Channel 2
Mesh router
operating on
Channel 1 and
Channel 3

11. Connectivity Optimal Capacity
Research Issues
Interface Channel Assignment
Channel assignment => Bandwidth of virtual links
Connectivity vs. radio spectrum utilization efficiency
Workload awareness
Connectivity Optimal Capacity
Packet Routing
Routing => Traffic load on virtual links and gateways
Network-wide load balance
Interaction between routing and channel assignment
Goal: Maximize network cross-section goodput

12. Load-Balancing Routing: Problem30 20 40 10 30 10 40 50
Ingress/Egress Traffic 20
For each mesh node, find multi-hop path(s) to the wire such that
(1) load on the gateway nodes is balanced
(2) load on the intermediate nodes is also balanced

13. Load-Balancing Routing: Solution
802.1D-like Gateway Discovery Protocol
Each node joins one (or more) gateways
Protocol: ADVERTISE/JOIN
Parent-child relationship between nodes
Structure: Forest of trees rooted at gateway nodes.
Cache extra advertisements for failure recovery
Metrics
Hop-count
+ stable because mostly static
- load-imbalance
Gateway residual capacity
+ load balanced, adapts to traffic
- route flaps because dynamic
Path residual capacity
+ handles non-gateway bottlenecks
(1)
(2)
(3)

14. Traffic-Aware Channel Assignment: Problem60
110 70 30 40 40 30 40 30 20
For each mesh node interface, assign channels such that
the resulting capacity of virtual links matches their loads

15. Traffic-Aware Channel Assignment: Solution
Workload-Awareness
Why ?
Need to distribute load uniformly across channels.
How ?
1. Periodically construct a neighborhood channel-usage map
2. Re-assign channels to balance traffic load across channels
3. Coordinate with direct neighbors 10 40 30
Channel load
imbalance 10 40 30 10 40 30
Channel load
balanced

16. Traffic-Aware Channel Assignment: Solution
Channel Dependency Issue
Each node has a limited number of interfaces. Hence each interface is used to communicate with multiple neighbors.
Control Channel
- Physical: Extra NIC on dedicated channel
- Virtual: Multi-hop connectivity to neighbors C E
SOLUTION D B A
Channel Load Metrics
- Contention group size
Aggregated channel usage
Weighted sum of the two

17. Centralized Channel Assignment/Routing
Channel Assignment is NP-hard
Reduction from “Multiple subset sum” problem
Greedy Channel Assignment
Visit edges in order of “expected load”
Greedily assign locally optimal channel
Maintain previous channel assignments as constraints
Centralized Routing
Single-path iterative routing on residual graph
Randomized multi-path load-balancing routing
Overall Algorithm
Start with single-channel routing for initial load estimation
Iterate over channel assignment and routing until convergence

18. Performance Evaluation: Throughput Gains
Simulation Setup
60 nodes with 4 gateway nodes
2 NICs/node, 12 channels
30 random flows to wired net
Cross-section goodput X
Results
Baseline: Single-channel net
Single-NIC Multi-channel:
Marginal improvements
Identical CA: 2x improvement
Centralized CA: 6-7x gains
Distributed CA: 6-7x gains

19. Performance Evaluation: Latency Reductions
Simulation Setup
64 nodes with 4 gateway nodes
2 NICs/node, 12 channels
HTTP traffic requests/response
Traffic intensity:0, X, 2X, 3X, 4X
Results
Reduced average delay
Saturation point: 4x users with
multi-channel networking

20. Some Implementation Issues..
Interference Range Neighbor Discovery
- Brute-force method
- lowest encoding
Inter-channel Interference
- Antenna separation
- Channel separation
Coordinating Channel (and Route) Changes
- Preventing (N1,C1)(N2,C1)  (N1,C2)(N2,C1) | (N1,C1)(N2,C2)
- Exchange (Channel C1 Channel C2, Time T)
Antenna orientations
- Neighborhood information per antenna
Channel Quality Estimation
- Link errors, Channel encoding

21. Prototype Evaluation Configuration –
9 Win XP desktops, 2 gateway nodes
Two a NICs / node
User-level route/channel
assignment daemon
FTP Throughput –
5-times improvement in multi-channel mode
Should be higher for larger testbed

22. IEEE 802.11 beyond AP—mobile communication
Conclusions..
IEEE beyond AP—mobile communication
Multi-channel wireless mesh backbone
Multiple commodity cards per node
Workload-aware channel assignment
Load-balancing routing
Many-fold improvement with small increase in price
Research problems to work on –
(1) “Optimal” centralized load-balancing routing ?
(2) Distributed channel assignment for general wireless mesh ?
(3) Capacity of multi-radio wireless mesh networks ?
(4) Applications to IEEE a mesh networks ?
Project site:

23. Other Related Mesh Networking Projects..
Mesh Transport Protocol:
that achieves much better and robust performance than TCP over multi-hop wireless networks
Station-Transparent Mobility Management:
that supports end-user mobility across a WMN without any software pre-installed on the stations
Secure Routing Protocol:
to protect a WMN from compromised routers.
Directional Antenna Protocols:
to reduce long-term interference and achieve better spatial reuse of channels.
Miniaturized Mobile Wireless Network Testbed:
to provide a manageable, reconfigurable, controllable multi-hop wireless experimentation platform

24. Capacity Issue in Single-channel Mesh Network
Capacity Issues
802.11: MAC contention, PLCP header, ACK, bit errors
Ad-hoc: Single-channel across the network
=> Inter-path and Intra-path interference
Increasing Capacity
Frequency: Multiple channels
Spatial: Directional antennas, Transmit power control

25. NP-hardness Proof