1 constructing locally centralized applications by mobile agents in wireless sensor networks...
Post on 20-Dec-2015
214 views
TRANSCRIPT
1
Constructing Locally Centralized Applications by Mobile
Agents in Wireless Sensor Networks
2008/05/14Shunichiro Suenaga* (Nihon Unisys Ltd./The Graduate University for Advanced Studies)Shinichi Honiden (The University of Tokyo/ National Institute of Informatics)
ATSN-2008
2
Index
1.Overview
2.Problems and Requirements
3.Approach
4.Evaluation
5.Related Work
6.Conclusions
3
Overview• When an application is constructed by
multiple agents, existing works have problems– Architecture– Group Migration
• We propose– Architecture (Three different role of agents)– Group Migration (Agent generation)
4
2. Problems and Requirements
5
2.1. Assumed Environment
WSN (roof)
Base Station
LAN
Stock Management
• Warehouse• Goods: Arrival/ Relocation/ Shipping• “Stock Management” manages locations of goods.
6
2.2. Locally Centralized Application
Temp. Monitoring Intruder Detection
+ Goods location sometimes change
Addition and Deletion
(Arrival and Shipping)
Specific processing for each goods.
7
2.3. Requirements① Spatial sensing around the goods.
( multiple programs are needed to cover spatial area )
② Programs can move to the new location (of the goods) and continue the processing .
Existing Works )Reprogramming (Deluge, Mate) and Mobile Agent (Agilla,Actornet) can realize addition and deletion of programs in a WSN.
8
Reprograming VS MA Addition and Deletion of Programs Specific Node Reprogram
Update ! Update !
Bottle-neck!
Base-Station Base-Station
Reprogramming Mobile Agent
9
2.4. Problems
① Architecture to constructing LCA (Locally Centralized Application ) by multiple mobile agents
② Simultaneous Group Migration to move to the new location of the goods
10
Unsophisticated agent architecture introduces following possibilities
Disrupts periodical sensing Makes Light weight agent heavy
2.4. Problem ① ( Architecture)
11
2.4. Problem (Group Migration) ②
Migration failure makes application execution impossible
Individual migrations of agents introduce following possibility
・ Lost
・ Fall awayCan’t continue execution
12
3. Approach
13
3.1 Approach (Architecture)①
LCA Requirements
A) Sensing
Execution of sensing on several nodes.
B) Information collection
Exchange of information between base station.
C) Application specific processing
judgment of the sensing results, execution of the processing in each specific situation.
14
3.1 Approach (Architecture)①Name Description
C) Master Judgment of the sensing results,
Application specific processing,
Group Migration, management of the lifecycles of Slave-S,Slave-M.(Approach② )
A) Slave-S Sensing around the goods. Multiple Slave-S are deployed in a group.
S: Sensing / Stationary
B) Slave-M Master gives an order to information collection Slave-M. (Information Exchange between node and base station)
M: Mobile
15
3.1 Approach① ( Architecture ) LCA (Wine Example)
When wine moves, group also needs to move..
16
3.2 Approach (Group Migration)②■Success Rate of migration (existing works)
P: Failure rate ( common in WSN )
N: number of programs
Success Rate= (1-P)^N
Decreasing N is a solution
Slave-Ss and a Slave-M are included in Master code.
Our Approach: Make N 1.
17
3.2 Approach (Group Migration)②
• Slave-S , Slave-T are included in Master• Master generates Slave-S and Slave-T
Mote
Middle
Mote
Middle
Migration
Transmitted
transformed
Re-generation
transformed
18
3.2 Approach (Group Migration)②
1. Master Injected
2. Slave Deployment
3. Start
4. Slave Killed
5. Migration
6. Deployed
19
3.3 Approach② ( Group Migration)
■ Deployment Pattern (Static)
- number of Slaves- Deployment Pattern
- All Surrounding node- N-Hop
■ Static Deployment Pattern Example
- 4 Slave-S are deployed on 4 of 1hop neighbors.
- 1 Slave-M is deployed on 1 of 1hop neighbors.
Optimization and Dynamic Deployment are Future Work
Where are Slave-S and Slave-M deployed ? (Specified in Master code)
20
3.4 Implementation
• We extended Agilla (ICDCS 2005)
Points
• Hierarchy (Master,Slave)
• Location Management
• Kill Manager
21
Evaluation
22
4.1 Evaluation
① Architecture- Scenario base
- Drawbacks of our approach
success rate of kill
② Group Migration- Success rate of migration
23
4.2 Simulation Setting
TOSSIM ・ Topology : grid
・ size : 8x8
・ baud rate : 40Kbps
・ Lossy : 5feet, 8feet,10feet
24
4.3 Evaluation (1/3)①Scenario(How many times can a group execute processin
g around the goods? )
• Master moves to the new location• Master deploys 4 Slave-S to 4 of 1hop neighbors• Master deploys 1 Slave-M to 1 of 1hop neigbors .• Master gives an order to Slave-M to go to base st
ation• Slave-M obtains the new location of goods• Group start to migrate to the new location.
25
4.3 Evaluation (2/3)①
5feet 8feet 10feet
2times-Our Approach
2times-Agilla
100%
73.3%
93.3%
60.0%
73.3%
33.3%
3times-Our Approach
3times-Agilla
93.3%
53.3%
73.3%
40.0%
40.0%
20.0%
4times-Our Approach
4times-Agilla
93.3%
33.3%
66.7%
20.0%
40.0%
6.8%
5times-Our Approach
5times-Agilla
86.6%
26.7%
60.0%
13.3%
33.3%
0%
26
4.3 Evaluation (3/3)①
5feet 8feet 10feet
Success rate of Kill 90.9% 68.6% 52.4%
Success rate of Kill
(Retransmit)
100% 100% 100%
Slaves can be killed completely in simulation environment.
27
4.4 Evaluation (1/2② )• 4-times Group Migration• Agilla (three agents)• Our Approach ( 1Master→Master,SlaveS,Slave
M )
5lossy 8lossy 10lossy
Agilla 3.46 3.0 2.66
80% 60% 40%
Our Approach
4.0 3.73 3.66
100% 80% 67%
Spatial distance of agents are 1hop
28
4.4 Evaluation (2/2② )
5lossy 8lossy 10lossy
Agilla 2.86 2.06 1.2
53.3% 26.6% 13.3%
Our Approach
4.0 3.4 3.2
100% 73.3% 60%
Spatial distance of agents are 2hop
• 4-times Group Migration• Agilla (three agents)• Our Approach ( 1Master→Master,SlaveS,Slave
M )
29
Summary of Evaluation
① Architecture– Architecture looks fair– Disadvantage is overcome in simulation
environment
② Group Migration– Our approach has better success rate than
existing work
30
Related Work
■ WSN• Reprogramming (Deluge, Mate , XNP, MNP ,
FireCracker)– code mobility
• WSN-MA (Agilla,ActorNet,SensorWare)– group migration
■ Wired Network• MA(Mobile Space,Ishikawa et al. )
• Assumed Environment• purpose
31
Conclusions
We propose, Architecture and Group Migration
We show the efficiency of our approach in simulation environment
Our approach will be one of the solutions to realize multiple simultaneous applications in WSNs.
32
Thank you !
33
Appendix
34
2.1. Assumed Environment• “Stock Management” manages locations of
goods. • WSN is
① Arrival ② Relocation ③ Shipping
Shelf ID Shelf ID
Link goods ID and shelf ID
Link goods ID and shelf ID(Relocation)
Delete shipped goods ID
Stock ManagementWi-Fi
Shipping ID
PDA
35
Limitation
• Migration Failure of Master– makes application execution impossible– while, existing work also has this problem
• Ignore Execution State of Slaves– Our approach always kill Slaves– Not required in assumed application
36
Future Work
• Dynamic acquisition of Deployment Pattern
Now: Middleware deploys Slaves in according to the static deployment pattern
Future:Master can adopt the situation, i.e. battery level, link state, number of neighbors, etc..
37
Sample Code