Distributed Channel Assignment and Routing Multiradio Mutlichannel Multihop Wireless Networks Haitao Wu, Fan Yang, Kun Tan, Jie Chen, Qian Zhang, and Zhenshrng Zhang Journal on Selected Areas in Communications 2006

Outline  Introduction  Problem Statement  CCM: Metric for JCAR  Distributed Algorithm for JCAR  Simulations  Conclusions

Introduction  Multiple orthogonal channels are defined in IEEE standards.(3 for b/13 for a)  With cheaper hardware and diverse wireless technologies, many mobile devices are equipped with multiple radios.  These devices may construct a multiradio multichannel multihop wireless network.(M 3 WN)

Introduction  How to effectively leverage multiradio and multichannel to conquer/reduce the wireless interference that widely exists in the classical multihop wireless networks.  To effectively mitigate interference, both routing and channel assignment (CA) should be carefully designed.

Introduction  CA determines the connectivity between radios since two radios can only communicate with each other when they are on a common channel.

Introduction - Challenges  A quantitative measure of the performance gain of any new JCAR pattern to denote any specific combined solution of CA and routing.  There might be no common radio or common channel supported in the whole network for both data transmission and signaling.  The signaling overhead to obtain updated traffic and link-state information for JCAR in such a heterogeneous environment.

Introduction - Challenges  The overhead of a/b/g NICs on channel switching should be taken into account.

Solution Summaries  Channel Cost Metric (CCM) reflects the expected transmission cost weighted by channel utilization.  “ equivalent fraction of air time ” (FAT) reflects the channel busy time and provides a common reference value for heterogeneous radios.

Solution Summaries  A distributed algorithm that effectively selects the JCAR pattern which has the smallest CCM value among a subset of potential JCAR patterns.

Problem Statement Initial Channel Equips two NICs Flow1 on 11g NIC and channel g1 New Flow

CCM: Metric for JCAR  CCM represents the expected per-unit transmission time on each channel weighted by channel utilization and explicitly captures the effect of the interference from hidden links.

Intuition for Metric Definition  Expected Transmission Time, ETT l i is the sum of transmission time of all data packets sent on channel l over a time unit.

Intuition for Metric Definition  FAT represents the normalized overall channel utilization.  For channel l at node i, FAT, F l i, is defined as the ratio of the total air time consumed.

Intuition for Metric Definition Based on this metric, it is apparent that pattern2 is better than pattern1.

Expected Transmission Time The average payload size The average transmission time of one data frame Collision probability

Normalized Channel Occupation Time-Fraction of Air Time

A Close-Form Expression for CCM

Distributed Algorithm for JCAR  A better pattern is found, the system may be reconfigured, and the total channel cost is reduced.  JCAR guarantees this by performing a feasibility check before applying a new pattern.  JCAR reconfigurations are performed periodically.

Distributed Algorithm for JCAR 1.Identify some of the possible JCAR patterns and for each of the identified JCAR patterns: 1)Check feasibility F l i (new pattern) <= Th F ; 2)Check network connectivity; 3)Determine its CCM value if feasible 2.Select a feasible pattern which has the smallest CCM value. 3.Conduct the switching operation

JCAR Candidate Pattern Selection  Only changing channels within the channel set to which cannel l belongs  Only changing interface between nodes i and j  Combination of both channel change and interface switching

Feasibility for CA and Routing  A new JCAR pattern is feasible. (no flow suffers degradation in throughput) (no flow suffers degradation in throughput)

Network Connectivity  Some JCAR patterns may result in network partitions and these patterns should be avoid.  Connectivity Invariance Rule If any node-pair that was originally connected, the node-pair should still be connected under a new JCAR pattern.If any node-pair that was originally connected, the node-pair should still be connected under a new JCAR pattern.  Only patterns involved with switching channels (categories 1 or 3 above) between two nodes may result in network partition.

Network Connectivity l → l’ l l Chain puzzle may involve a large number of nodes It is difficult to synchronize the switching actions among all nodes

Network Connectivity  In the above case, node i, j should give up the channel switching.  The node initiates a switching request will check if there contains chain puzzle within its two hop information.

Distributed JCAR  If the CCM value under the new pattern is smaller than the current one by a predefined threshold TH CA, then node i starts the operation for CA and routing switching.  For interface switching, the action is taken locally, only between node i and j.  For channel switching, a distributed procedure is required to guarantee that all the neighbors accept such changes,.

Distributed JCAR the CCM value is smaller than TH CA To indicate that it wishes to switch channel l to l’ To avoid synchronized pattern selection triggered for nodes in a neighborhood ！

Simulations  Using NS-2 simulator  A mesh network on a grid with 36 nodes,  The channel model is TwoRayGround  Trigger threshold TH H is  Setting to a DCF with 54Mb/s with 54Mb/s

Simulations

Simulations

Simulations  Another evaluation is in a nine-node wireless testbed.  All nine nodes are DELL OPTIPLEX  Nodes 10, 24, 27, 28 and 31 are equipped with two a/b/g combo cards. combo cards.

Simulations

Simulations

Conclusions  Discussion how to improve the performance by joint CA and routing for a heterogeneous M 3 WN.  Presenting CCM a critical metric that quantifies the difference for various JCAR patterns.  A distributed algorithm, JCAR, is to select a JCAR pattern that has smallest CCM at each node locally.