Runtime Optimisation in WSNs for Load Balancing Using Pheromone - - PowerPoint PPT Presentation

Runtime Optimisation in WSNs for Load Balancing Using Pheromone Signalling

Ipek Caliskanelli James Harbin Leandro Soares Indrusiak Paul Mitchell David Chesmore Fiona Polack

Department of Computer Science

2 2

Outline

• Motivation and Background

 why do we work on load balancing on WSNs?  applied techniques- task mapping on WSN

• Problem Statement
• System Model
• Load Balancing Based on Pheromone Signalling

 Linking biological concepts with WSN  Algorithm

• Verification
• Conclusions

3 3

Motivation and Background What is a wireless sensor node?

• Small
• Autonomous
• Self-powered

Each node consist of

limited resources:

embedded processor, memory, battery, radio transceivers and environmental sensors.

4 4

Motivation and Background Why do we work on load balancing?

Processing capabilities Energy restrictions

Prevents to achieve the high performance efficiency in terms of service availability and Quality of Service (QoS)

5 5

Motivation and Background We propose

 task mapping optimisation

• to solve the energy VS service availability trade-off

We apply

 dynamic task mapping at runtime

• to represent the dynamic nature of the WSN
• to provide efficient solutions

We use

 Biological knowledge of the bee colonies

• To take the advantage of nature
• To mimic
• highly self-organised systems
• adaptability against changes

6 6

Problem Statement The primary design objectives

 maximisation of service availability  minimisation of energy consumption

7 7

System Model Three layer system model consist of

Application Model

Consists of tasks

Maping

Load Balancing process Selection of the QN, pheromone propagation and emission

Platform Model consists of sensor nodes

8 8

System Model § Platform Model

 Consist of set of nodes, N

• ,

memory capacity

energy capacity pheromone level

 Consist of set of links, L

• ,

sender

receiver

N ni ∈

} , , {

i i i i

h e mc n =

L l

k ∈

} , {

j i k

n n l =

9 9

System Model § Application Model – DAG representations

 Consist of set of tasks, T

• ,

memory footprint

energy consumption execution time

 Consist of set of communications, C

• ,

sender

receiver

T ti ∈

} , , {

i i i i

et e mf t = C c

k ∈

} , {

j i k

t t c =

10 10

System Model

• A service is provided by one or more network nodes, and

can be requested or triggered by end users, other nodes or even the environment

• Each service consist of multiple tasks
• A service is available if all of its tasks can successfully be

executed by the network nodes

11 11

Outline

ü Motivation and Background ü Problem Statement ü System Model

• Load Balancing Based on Pheromone Signalling

 Linking biological concepts with WSN  Algorithm

• Verification
• Conclusions

12 12

Outline

ü Motivation and Background ü Problem Statement ü System Model

• Load Balancing Based on Pheromone Signalling

 Linking biological concepts with WSN  Algorithm

• Verification
• Conclusions

13 13

Biological Background

In Bee Colonies;

• Honey bees need a queen bee to orchestrate the colony and

facilitate social interactions For This Purpose;

• Bees developed a special pheromonal system in order to

maintain the required harmony and orchestration Works as;

• Queen bee stimulates a unique pheromone for the worker

bees to realize the absence or presence of the queen bee

14 14

Biological Background

• This pheromonal mechanism works whereby the worker bees

lick the queen bee and pass the queen substance to each

• thers
• If there is no queen substance passed through the worker

bees; worker bees will then consider the queen as dead so they will grow a larva and will be feed with royalactin protein

**Royalactin protein induces the differentiation of honey bee larvae into queens; increases body size, ovary development and shortened developmental time in honey bees

• Whereas if worker bees receive the queen substance, they will

know that there is a queen bee to orchestrate the colony and take no action towards building a new queen

15 15

Load Balancing Based on Pheromone Signalling

Bees Pheromone Stimulation Sensor Network

Queen Bee Sensor node responsible for task mapping and execution (QN) Worker Bees Sensor node (WN) Pheromone Level Parameter used for QN selection Lifetime of Bee Operation Lifetime of the Sensor Node

TABLE 1: CORRELATION BETWEEN BEE’S PHEROMONE STIMULATION AND SENSOR NETWORKS

16 16

Load Balancing Based on Pheromone Signalling The objective of Pheromone Signalling (PS) Algorithm is

 to enable node differentiation at a scale that produces sufficient QNs to handle all the required system functionality  to avoid unnecessary redundancy

17 17

Load Balancing Based on Pheromone Signalling

By applying Load Balancing Algorithm;

• QNs stimulate pheromone
• Nodes accumulate pheromone
• Each node differentiate itself into QN depending on their

pheromone level We formalise the PS algorithm by describing its three parts which are executed on every node of the network

• Differentiation cycle
• Propagation cycle
• Decay cycle

18 18

Load Balancing Based on Pheromone Signalling

First Part: Differentiation Cycle

Occurs on Periodic Basis

§ executes on nodes every TQN time units. § a node checks its current pheromone level hi. and will differentiate itself into either a QN or WN. § if a node differentiate itself as a QN, it propagates pheromone to its network neighbourhood. LISTING 1: PS DIFFERENTIATION CYCLE 1 every TQN do 2 if ( ) 3 =true 4 broadcast hd = {0, hQN} 5 else 6 =false

thresholdQN

<

hi QN i QN i

19 19

Load Balancing Based on Pheromone Signalling

Second Part: Propagation Cycle

Occurs on demand

§ executes every time a node receives a pheromone dose. § a node checks whether the QN that produces it is sufficiently near for the pheromone to affect it. § if the hd has travelled more hops than the threshold, the node simply discards it. If not, it adds the received dosage of the pheromone to its own pheromone level. LISTING 2: PS PHEROMONE PROPAGATION 1 When hd is received 2 if ( ) 3 4 broadcast hd’ = { }

thresholdhopcount

hd < ] 1 [ ] 2 [ hd

h h

i i

+ =

K HOPDECAY

hd hd ]. 2 [ , 1 ] 1 [ +

20 20

Load Balancing Based on Pheromone Signalling

Third Part: Decay Cycle

Occurs on Periodic Basis

• decay of the pheromone level of each node,

which happens every TDECAY time units to represent the elapsed time. LISTING 3: PS DECAY CYCLE 1 every TDECAY do 2

K h h

DECAY TIME i i

. =

21 21

Outline

ü Motivation and Background ü Problem Statement ü System Model ü Load Balancing Based on Pheromone Signalling

• Verification

 Simulation infrastructure  Advantages vs disadvantages of system-level simulation and real node deployment

• Conclusions

22 22

Outline

ü Motivation and Background ü Problem Statement ü System Model ü Load Balancing Based on Pheromone Signalling

• Verification

 Simulation infrastructure  Advantages vs disadvantages of system-level simulation and real node deployment

• Conclusions

23 23

Verification Design Goals;

• Short implementation time
• Minimum financial cost
• High performance efficiency

 System-level simulation model

To achieve the most accurate results real sensor datasheets are used

24 24

Verification Analysed;

• Two different size platform models in mesh network

topology, to capture the effects of the technique on the scalability

 Platform Model: 4x4 and 7x7 Mesh Topology

 Application Model: Three different types of DAG which contains 8, 10 and 14 tasks are designed and referred as Service

• The key parameters (decay and propagation period) and

showed their importance on the performance

• Short term effects on real sensor deployments

25 25

Verification - Real Sensor Deployment

• the total number of event

detections received over time and the number of packets transmitted in the network in total.

• the smaller the numbers of event

detection is efficient due to the minimal duplication.

26 26

Verification - Real Sensor Deployment

impact of queen hormone threshold thresholdQN upon the measured event processing and packet transmission load. differentiation algorithm tolerates a stable state with additional queens in the

• network. This leads to an

approximately 10% increase in the total redundant event

• processing. Total packet

transmissions also increased.

27 27

Verification – System Level Simulation

Experimental results: On 4x4 mesh network topology (a) % Events detected, (b) % Alive nodes.

1 2 3 4 5 6 7 8 9 10 10 20 30 40 50 60 70 80 90 100 Time (Week) % Events Detected 4x4 Mesh Network Baseline BS PS 1 2 3 4 5 6 7 8 9 10 10 20 30 40 50 60 70 80 90 100 100 Time (Week) % Alive Nodes

4x4 Mesh Network

Idle Baseline BS PS

(a) (b)

28 28

Verification – System Level Simulation

1 2 3 4 5 6 7 8 9 10 10 20 30 40 50 60 70 80 90 100 % Events Detected Time (Week)

4x4 Mesh Network

Idle PS 7200-7200 PS 7200-5000 PS 7200-3000 PS 7200-2000 1 2 3 4 5 6 7 8 9 10 10 20 30 40 50 60 70 80 90 100 % Alive Nodes Time (Week)

4x4 Mesh Network

Idle PS 7200-7200 PS 7200-5000 PS 7200-3000 PS 7200-2000

Experimental results: effects of different phormone decay period on 4x4 Mesh Network topology using : (a) % Events detected, (b) % Alive nodes. (a) (b)

29 29

Verification – System Level Simulation

1 2 3 4 5 6 7 8 9 10 11 12 13 10 20 30 40 50 60 70 80 90 100 Time (Week) % Events Detected

7x7 Mesh Network

Baseline PS 7200-7200 PS 7200-5000 PS 7200-3000 PS 7200-2000 1 2 3 4 5 6 7 8 9 10 11 12 13 10 20 30 40 50 60 70 80 90 100 Time (Week) % Alive Nodes

7x7 Mesh Network

Idle Baseline PS 7200-7200 PS 7200-5000 PS 7200-3000 PS 7200-2000

% Events detected (a), % alive nodes (b) for 7x7 Mesh Network topology. (a) (b)

30 30

Conclusion PS algorithm is verified with

 Real sensor deployments implemented to analyse the short term effects  System- level simulation implemented to analyse the long term advantages

Both results show;

 Decrease energy consumption  Increase service availability by the effective usage of service time

• f the network components by balancing the load of network

components across the resources

31 31

Conclusion

Our technique provides;

• Highly self-organised network structure
• Without the need of a central control mechanism over the

network

• Benefit in large networks by removing the unnecessary

redundancy as the results show

• By reducing the redundancy, in large network, we achieved

more than 25% longer network lifetime with higher level of service availability