Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

21
1 Algorithms for Bandwidth Efficient Multicast Routing in Multi-channel Multi-radio Wireless Mesh Networks Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen Department of Computer Science and Engineering York University, Canada

description

Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen Department of Computer Science and Engineering York University, Canada. Algorithms for Bandwidth Efficient Multicast Routing in Multi-channel Multi-radio Wireless Mesh Networks. Outline. Motivation Problem Formulation - PowerPoint PPT Presentation

Transcript of Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

Page 1: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

1

Algorithms for Bandwidth Efficient Multicast Routing inMulti-channel Multi-radio Wireless Mesh Networks

Hoang Lan Nguyen and Uyen Trang Nguyen

Presenter: Hoang Lan Nguyen

Department of Computer Science and EngineeringYork University, Canada

Page 2: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

2

Outline

Motivation Problem Formulation The Proposed Algorithms Performance Evaluation Conclusion and Future Work

Page 3: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

3

Motivation

Theoretical results indicate that the throughput capacity of a single-channel wireless mesh network becomes unacceptable low as the number of nodes increases.

One of the most effective approaches to enhance network throughput is to use systems with multiple channels and multiple radios (MCMR) per node.

Research on multicast (one-to-many communication) has focused mostly on networks with a single channel.

– Traditional multicast routing algorithms designed for single-channel environment, for example Shortest Path Tree (SPT) or Minimum Steiner Tree (MST), are not suitable for multi-channel multi-radio networks as they did not consider the channel diversity.

Page 4: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

4

Motivation (cont.)

The problem of multicasting in multi-channel multi-radio networks has only been studied recently.

These studies addressed the multicast problem based on the following approach:

– First, a mutlicast tree is constructed based only on network topology

– Then, a channel assignment is applied on top of the constructed tree to optimize an objective function (such as maximizing throughput or minimizing delay).

– We call this approach “Routing first, Channel Assignment second”

Page 5: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

5

Motivation (cont.)

• Drawbacks of the “Routing first, Channel Assignment second” approach:

• It does not consider existing channel assignments currently used by other types of communications such as unicast

• The new channel assignment for multicast may conflict with the current unicast channel assignment.

Page 6: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

6

Our Approach

• We instead consider the reverse approach called “Channel Assignment first, Routing second”:• We construct multicast trees on top of currently deployed channel

assignments.

• Advantages of this approach:• Re-use existing channel assignments

• Have no channel conflict with other types of communications

• Easy to deploy in any existing multi-channel multi-radio systems

Page 7: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

7

Problem Formulation

• Given a multi-channel multi-radio wireless network with already allocated channel assignments, the objective is to:

• construct a multicast tree with minimum bandwidth consumption

• minimize the number of transmissions used by the multicast tree

Page 8: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

8

Problem Formulation (cont.)• The problem focuses on the number of transmissions each

forwarding node requires to multicast a packet to its downstream node(s) in the multicast tree

– Note: in single-channel network, it is always one

– However, in multi-channel network, it may be greater or equal to one due to channel diversity

• For example, for the multicast tree below (drawn in blue arrows), for every packet, forwarder N has to make two transmissions, one on channel 1 to node I and the other on channel 3 to node K

Page 9: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

9

Problem Formulation (cont.)

• The number of transmissions consumed by a multicast tree is the total of the numbers of transmissions summed over all forwarding nodes in the multicast tree

• We have proved that constructing a multicast tree with minimum number of transmissions in a multi-channel multi-radio network is a NP-hard problem

• We, therefore, find approximate solutions by proposing heuristic algorithms

Page 10: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

10

The Proposed Algorithms

We propose multicast routing algorithms that take into account the channel diversity in multi-channel systems in order to minimize the amount of network bandwidth consumed by the multicast tree.

Specifically, given a multi-channel multi-radio network with already allocated channel assignments, the algorithm constructs a multicast tree that minimizes the total number of transmissions required to deliver a data packet from the source to all multicast destinations.

It uses a proposed routing metric that maximizes the wireless broadcast advantage and minimizes interference among nearby (one-hop away) forwarding nodes.

The wireless broadcast advantage refers to the concept that a single transmission from a node can reach all of its one-hop neighbors. This is true in single-channel networks.

However, this is not always the case in multi-channel networks due to channel diversity

Page 11: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

11

The Proposed Routing Metric

• Channel utilization µu(c) at node u is the number of incident links on u that are assigned channel c

• Channel metric δu(c) at node u is defined as 1 / µu(c)

– small δ values imply high channel utilization

Page 12: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

12

The Proposed Routing Metric (cont.)

• Link cost w(u,v) of link (u,v) is defined as δu(c) / δv(c) where c is the channel used by link (u,v)

– The term δu(c) in the link cost favors a transmitter with a channel highly utilized so that the channel can be used for as many receivers as possible. This is to maximize the wireless broadcast advantage.

– Given link (u,v) on channel c, the next-hop link (v,z) to be added should avoid channel c so that transmissions from u and v do not interfere

– Therefore, given a transmitter u with highly utilized channel c, we should choose v with lowly utilized channel c. This explains the term 1/ δv(c) in the link cost. This is to minimize interference among forwarding neighbors.

• Path cost of a path is the sum of link costs of the links on the path

Page 13: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

13

The Centralized Algorithm

• The centralized algorithm then uses the proposed link costs to build a minimum cost multicast tree based on a Steiner heuristic and greedy approach:

– The tree initially contains only the source.

– Then a multicast destination with the least path cost to the tree is added to the tree

• This repeats until all multicast destinations are added to the tree

Page 14: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

14

The Distributed Algorithm

• The distributed algorithm consists of two phases:

- Phase 1: a broadcast tree with minimized number of transmissions is built using distance-vector routing

- Phase 2: multicast tree is then constructed by pruning the broadcast tree

Page 15: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

15

Performance Evaluation• We call the proposed multicast trees MCMNT (Multi-

Channel Minimum Number of Transmissions) and compare with other types of multicast trees namely: – Shortest Path Tree (SPT)– Minimum Steiner Tree (MST)– Minimum number of Forwarder Tree (MFT) [Ruiz et al., ISCC’05]

Recall that, unlike MCMNT, the SPT, MST and MFT multicast trees are constructed based on network topology only, and do not take the underlying channel assignments into account.

• The performance evaluation was done using Qualnet simulator

Page 16: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

16

Experiment Setup

• 100 nodes uniformly distributed over a 1700m x 1700m network with random channel assignments

• Each node has a transmission range of 350m

• The data rate at the 802.11 physical layer is 11 Mbps

• The 802.11 CSMA/CA without RTS/CTS is used for multicast communications

• At the transport layer we do not use any flow or congestion control mechanisms to test the network performance under heavy loads

• The multicast source is placed at the center sending data at a constant bit rate, while multicast destinations are randomly scattered around the network

• Each experiment is run for 600 seconds of simulated time

• Each data point in the resulting graphs is averaged from five runs using random seeds and plotted with a confidence interval of 95%

Page 17: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

17

Experiment Scenarios

• We consider the following scenarios:

– multicast group size: number of multicast destinations varies from 20 to 80 nodes. The number of radios per node and the number of channels are set to 3. The source transmits at a rate of 200 packets/s

– multicast source rate: multicast source rate varies from 100 to 300 packets/s. The number of channels and radios per node is 3. The multicast group consists of 40 destinations.

– number of channels: number of channels is set to 1, 3, 5, and 7. The multicast group contains 40 destinations and the source

rate is 200 packets/s

Page 18: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

18

Experiment Results: Group Size

• MCMNT trees have the least transmission consumption and the highest packet delivery ratio

Page 19: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

19

Experiment Results: Source Rate

• This is also the case for the source rate scenario under heavy traffic loads

Page 20: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

20

Experiment Results: Number of Channels

• MCMNT also outperforms in various numbers of channels (except for the single-channel case)

– Note: MCMNT is not optimized for single-channel systems as there is no channel diversity in such environment

Page 21: Hoang Lan Nguyen and Uyen Trang Nguyen Presenter: Hoang Lan Nguyen

21

Conclusion and Future Work• We studied the problem of constructing multicast trees

with minimum number of transmissions in wireless mesh networks where multiple channels and radios are used

• Our proposed algorithms outperformed traditional multicast trees with respect to various performance metrics.

• Our future work:

– includes traffic load into the link and path cost computations for better load balancing and performance under dynamic network conditions