Title: A Method for Distributed Computation of SemiOptimal Multicast Tree in MANET
1A Method for Distributed Computation of
Semi-Optimal Multicast Treein MANET
- Eiichi Takashima, Yoshihiro Murata, Naoki
Shibata, - Keiichi Yasumoto, and Minoru Ito.
- Nara Institute of Science and Technology, Shiga
University
2Outline of this presentation
- Background
- Proposed method
- Evaluation experiment
- Conclusion
3Background
- Video streaming - one of the most important
application in mobile ad-hoc network (MANET) - Objective Delivering video to many nodes in
MANET - Using a multicast tree
- Satisfying QoS constraints
- Bandwidth
- Delay
- Optimized for any given objective
- Power consumption (Mobile nodes are operated on
battery) - Maximizing number of receiver nodes
4Background
- Optimizing multicast tree on MANET
- A hard task - an NP-hard problem
- Dynamic network topology
- Limited capabilities of mobile terminals
- Computation
- Communication
5 Existing studies
- P. Sinha, et al. 1
- Distributed algorithm
- Good scalability
- No handling of multiple QoS constraints
- No optimization for a particular objective
- Li Layuan, et al.2
- Centralized algorithm
- Optimizes any objective with multiple QoS
constraints - Poor scalability
- Cost of gathering topology information
- Centralized computation
1 P. Sinha and R. Sivakumar and V. Bharghavan,
"MCEDAR Multicast core extraction distributed
ad-hoc routing", WCNC(1999),
2 Li Layuan and Li Chunlin, "QoS Multicast
Routing in Networks with Uncertain Parameters",
APWeb, (2003).
6Outline of this presentation
- Background
- Proposed method
- Evaluation experiment
- Conclusion
7Goal of this research
- Constructing multicast tree
- Satisfying all given QoS constraints
- Optimizing a given objective
- total power consumption
- tree stability
- Good scalability
- Distributed computation
8Our Approach
- We use GA (Genetic Algorithm) to construct
semi-optimal multicast tree - To realize distributed computation
- we compute multicast tree on several nodes in
parallel using GA - Each node solves a sub-tree for the whole
multicast tree - We divide MANET into multiple clusters
- Advantage of using GA
- Quick computation using results of previous
computation - Especially when topology change is small
9Hierarchical computation
- Two tier computation local trees and global
tree - A local tree connects nodes in a cluster
- The global tree connects clusters
node
cluster
Local tree
Global Tree
10Target Environment Assumption
- Service
- deliver small video (or audio) data from a sender
node to multiple receiver nodes in MANET - requirement transmission rate B, tolerable
end-to-end delay D - MAC protocol of wireless communication
- IEEE 802.11
- Mobile nodes
- move at speed of 4 Km/hour (pedestrian)
- can measure available bandwidth and delay to
neighboring nodes - can estimate approximate distances to neighboring
nodes by strength of radio wave signals
11Problem Definition
- Input
- topology info G(V,E), where V is set of nodes,
E is set of links - sender node sÎV
- receiver nodes Rr1,rm ÍV
- Output
- Multicast tree T(V,E), where V Í V, E Í E
- Constraints
- each link eÎE has available bandwidth no less
than B - total delay of each path in T is no more than D
- Objective
- maximize stability of T (links are connected for
longer time) - maximize service availability (more nodes can
receive video) - minimize total power consumption
- etc
12Typical Objective Functions
- Our method solves problem for intra-cluster and
inter-cluster separately ? use different
functions - Global Tree T maximize FG
- FG aNumberOfReceivers(T)
- - bNumberOfDelayViolation(T)
- g Stability(T)
- Local Tree T maximize FL
- FL NumberOfReceivers(T) e
Stability(T) - a, b, g, e are coefficients.
service availability
service availability
Tree stability
Tree stability
Term for power consumption can also be added
13Procedure Phase1 Cluster division
Top cluster head responsible to global tree
construction
Cluster division
e
Inter cluster
e
Gathering topology info in each cluster
S
Gathering topology info between clusters
e
e
e
Computation of global tree
Intra cluster
Computation of local tree
Cluster re-division
Cluster head responsible to local tree
construction
14Phase2 Gathering Local Topology Info
Cluster division
e
Inter cluster
e
Gathering topology info in each cluster
S
e
e
e
Gathering topology info between clusters
Intra cluster
Computation of global tree
Computation of local tree
(1) Cluster head floods request msg in its cluster
Cluster re-division
15Phase2 Gathering local topology Info
Cluster Division
e
Inter cluster
e
Gathering topology info in each cluster
S
e
e
e
Gathering topology info between clusters
Intra cluster
Computation of global tree
Computation of local tree
(1) Cluster head floods request msg in its
cluster (2) Each node received the message sends
back a message with its ID and link state info
including B/W and delay to neighboring nodes.
Cluster re-division
16Phase3 Gathering global topology info
(1) Each cluster head measures QoS info on paths
to cluster heads of adjacent clusters. (2) Each
cluster head sends the info to the top cluster
head.
Cluster Division
Gathering topology info in each cluster
Inter cluster
e
e
Gathering topology info between clusters
S
e
e
e
Computation of global tree
Intra cluster
Computation of local tree
Cluster re-division
17Phase4 Computation of global tree
(1) Top cluster head (and some nodes) computes
global tree by using island model GA.
Cluster Division
Gathering topology info in each cluster
Inter cluster
e
e
Gathering topology info between clusters
S
e
Computation of global tree
e
e
Intra cluster
Computation of local tree
Cluster re-division
18Phase4 Computation of global tree
(1) Top cluster head (and some nodes) computes
global tree by using island model GA. (2)
Information of global tree is sent to each
cluster head in the tree.
Cluster Division
Gathering topology info in each cluster
Inter cluster
e
e
Gathering topology info between clusters
S
e
Computation of global tree
e
e
Intra cluster
Computation of local tree
Cluster re-division
19Phase5 Computation of local tree
Cluster Division
Inter cluster
e
e
Gathering topology info in each cluster
S
e
Gathering topology info between clusters
e
e
Intra cluster
Computation of global tree
Computation of local tree
Cluster re-division
Each cluster head computes local tree which can
be grafted to global tree
20Phase5 Computation of local tree
Inter cluster
Cluster Division
e
e
Gathering topology info in each cluster
S
e
e
e
Gathering topology info between clusters
Intra cluster
Computation of global tree
Computation of local tree
The island model GA is used for computation of
local tree
Cluster re-division
21Phase5 Computation of local tree
Inter cluster
Cluster Division
e
e
Gathering topology info in each cluster
S
e
e
e
Gathering topology info between clusters
Intra cluster
Computation of global tree
Computation of local tree
The info of local tree is sent to each node in
the tree
Cluster re-division
22Phase5 Computation of local tree
Inter cluster
Cluster Division
e
e
Gathering topology info in each cluster
S
e
e
e
Gathering topology info between clusters
Intra cluster
Computation of global tree
Computation of local tree
The semi-optimal multicast tree has been
constructed among nodes.
Cluster re-division
23Phase6 Cluster re-division
Inter cluster
Cluster Division
e
e
Gathering topology info in each cluster
S
e
e
e
Gathering topology info between clusters
Intra cluster
Computation of global tree
Computation of local tree
After a while, MANET is clustered again and
procedure from phase2 is repeated to reflect
change of topology.
Cluster re-division
24Outline of this presentation
- Background
- Proposed method
- Evaluation
- Conclusion
25Evaluation
- Criteria
- Advantage of GA for computing multicast tree
- Feasibility in practical environment
- Superiority to existing method
26Advantage of the proposed algorithm
- Objective is to investigate
- scalability against number of nodes
- efficiency of re-computation when topology
changes - Experimental Configuration
- Mobility model of nodes
- Random way point, 4 Km/hour
- PC (laptop) for executing algorithm
- CPU Intel(R) Pentium(R) M processor
1500MHz,Windows XP,cygwin 1.5.18,gcc version
3.4.4.
27Result of (re)computation time of tree
Seconds
- Computation time
- 6 sec for 800 nodes
- 1 sec for 100 nodes
- Re-computation time
- shortened to 60
sufficient
Number of nodes
Computation time ? approximation of computation
time Re-computation time ? approximation of
recomputation time
28Feasibility in practical environment
- Evaluated the following points with 1000 nodes on
30 clusters (33 nodes per cluster) - Computation cost
- Required bandwidth for MANET
- Experimental result
- Computation time 0.04 second
- Needed bandwidth6.3K bps
- Proposed method is feasible in practical
environment.
29Superiority to existing method
- Investigated performance of our method
- Show superiority to existing method
- Index transition of packet arrival rate as time
progresses - Experimental configuration
- Area size
3000m3000m - Number of nodes
1000 - Simulator
GTNetS - Radio Range
160m - MAC layer protocol IEEE802.11 (Max. 2Mbps)
- Max of Speed 4 Km/hour
- Mobility model random
waypoint
30Comparison with existing method
- AQM (on-demand multicast routing method)3
- Proposed method
- Optimized for communication stability
- Optimized for the number of receivers
- Optimized for power consumption
3K. Bur and C. Ersoy. Ad Hoc Quality of
Service Multicast Routing. Computer
Communications, 29(1)136148, December 2005.
31Transition of packet arrival rate
AQM Stability . of receivers Power-saving
second
The proposed method is superior to AQM in terms
of packet arrival rate
32Conclusion
- We proposed a new multicast routing method for
MANET. - To construct the semi-optimal multicast tree
satisfying several QoS constraints for any given
objective - We show that the proposed method is feasible in
practical environment.
33 34Result of power consumption
Unit Watt-second
35Power consumption
- Compared item
- Transmission power consumption in 20 seconds
- 20 seconds reconstruction interval of multicast
tree