1. Efficient allocation of resources in multiple heterogeneous Wireless Sensor Networks
Wei Li, Flávia C. Delicato
Paulo F. Pires, Young Choon Lee
Albert Y. Zomaya, Claudio Miceli
Luci Pirmez
Journal of Parallel Distributed Computing, 2014
Presented by Group 6
Tianyi Pan

2. Traditional Wireless Sensor Networks (WSNs)
Application specific design
Single target application
Goal of achieving energy efficiency
Limited use of versatile sensor nodes
WSNs have the capability to allow nodes to perform different tasks
System inefficiency – resource wastage

3. Resource allocation in IoT Middleware
WSNs in IoT
Targeting diverse applications
Collect and process data from different type of sensor nodes
Different application QoS requirements
WSN system required to integrate heterogeneous WSNs to perform multiple co-existing WSN applications
How to fully utilize the resources?
How to maintain energy efficiency?
Resource allocation in IoT Middleware

4. How to assign tasks to sensors?
System Architecture
APP
Tasks
How to assign tasks to sensors?

5. Application Model Application: <𝑉,𝐺,Δ𝑡, 𝐶>
tasks area duration confidence
Task: <service type, data sensing rate, data sending rate, workload>
Confidence: Probability of achieving an application goal when a subset of all tasks are executed.
Fire detection application: tasks to detect 𝐶 𝑂 2 ,𝐶𝑂, temperature, smoke
100% confidence when all tasks are executed
30% confidence when only temperature is detected
User defined

6. System Model 𝑘 WSNs 𝑊={ 𝑊 1 , 𝑊 2 ,…, 𝑊 𝑘 } in area 𝐺
Each WSN 𝑖 has a single sink 𝑆 𝑁 𝑖 and 𝑃 sensors
Each sensor node belongs to exactly one WSN
WSNs may overlap with others in sensing range
Sensing precision
Impacted by environmental noise, proximity to the target area, number of neighboring nodes
Precision 𝑀 𝑖 =1− 𝑁 𝑖 100 , where 𝑁 𝑖 is the noise level in [0,100]
Assumptions
Sensors have the same ratio interface, communication and sensing ranges
Interference not considered
Keep track of sensor states: location, residual energy, etc.

7. Energy Model
Communication, sensing and processing modules for each sensor
Only consider the case when modules are active
Omit processing module in this paper
𝐸= 𝐸 𝑐 + 𝐸 𝑠
Communication cost for transmit 𝑙 bit of data from source to destination:
𝐸 𝑐 𝑙,𝑠𝑟𝑐,𝑑𝑒𝑠 = 𝑢,𝑣 ∈ 𝑃 𝑠𝑟𝑐,𝑑𝑒𝑠 𝐸 𝑙, 𝑑 𝑢𝑣 , 𝐸 𝑙, 𝑑 𝑢𝑣 = 𝐸 𝑒𝑙𝑒𝑐 ×𝑙+ 𝜀 𝑎𝑚𝑝 ×𝑙× 𝑑 𝑢𝑣 2
Sensing cost for service 𝑖 for time period 𝑡
𝐸 𝑠 =𝐸 𝑅 𝑠 𝑖 ×𝑡

8. Resource Allocation Problem
System lifetime
The time until the first sensor among all WSNs completely depleted its energy
Given: 𝑚 applications, each application 𝐴 𝑥 has 𝑉 𝑥 tasks
Find: a mapping Π: 𝑉 𝑥 →𝑃 for each application
Such that: the application requirements are met and the system lifetime is maximized

9. Service and Requirement Analysis
Three main types of service
Sensing: continuous data reading, detect occasional event within an area
Computation: fuse incoming data temporarily/spatially, make local decision, eliminate noise/redundant data, store/forward data
Communication: claim usage/duration/bandwidth of selected wireless communication channel, determine route to destination
Application/task requirements
Data accuracy or precision
Network lifetime
Balance between accuracy/energy consumption

10. Fulfilling Application Requirements
Data accuracy
The best candidate is the one with the least noise
Energy conservation
Reduce sensing energy: use one sensor to concurrently serve multiple tasks
Reduce communication energy: choose the sensor that is closest to a sink
Combined: choose the sensor node with maximum weight
𝛼 𝐷 𝐷 𝑏𝑒𝑠𝑡 +𝛽 𝐸 𝑟𝑒𝑠𝑖𝑑𝑢𝑎𝑙 −𝐸 𝐸 𝑟𝑒𝑠𝑖𝑑𝑢𝑎𝑙
𝛼+𝛽=1

11. The Heuristic Allocation Algorithm
App Arrival
Decompose app to tasks
Task Req.?
Combined
Precision
Weight
Select sensor with largest weight
Select sensor with least noise
Select sensor with least energy cost
Last Task?
Yes
End
No, check next task

12. Performance Analysis - Setup
Parameter
Value
Application arrival rate
Poisson distribution with mean 10
Multiple app arrival rate
Range [0,1], with 1 being most likely, default 0.1
Max number of apps arriving concurrently 3
Tasks in application
[2,6], uniform
Workload of application
5,15 , uniform
Data sensing rate
∞ (continuous sensing)
Data sending rate
1,3 , uniform
𝐸 𝑒𝑙𝑒𝑐
50 𝑛𝐽/𝑏
𝜀 𝑎𝑚𝑝
10 𝑝𝐽/𝑏/ 𝑚 2
Initial energy
10 −5 𝐽
𝛼,𝛽
0.5
Not specified: service types for the tasks, geographical requirement of the tasks, capability of the sensors, noise, topology, how WSNs overlap

13. System Lifetime

14. Data Accuracy
●: 10% overlapping
■: 50% overlapping
▲: 90% overlapping

15. Energy Consumption
More overlapping: more chance for data sharing

16. Testbed Experiments: Sun SPOT Platform
Use real energy consumption data

17. Questions?