on agent-friendly aggregation in networks

Post on 22-Jan-2016

46 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

On agent-friendly aggregation in networks. ATSN 2008 (at AAMAS 2008) Christian Sommer and Shinichi Honiden National Institute of Informatics, The University of Tokyo Tokyo, Japan. Agenda. Sensor networks Aggregation Agent aggregation specifics Problem model: aggregation graph - PowerPoint PPT Presentation

TRANSCRIPT

Agent-friendly aggregation 1

On agent-friendly aggregation in networks

ATSN 2008 (at AAMAS 2008)

Christian Sommer and Shinichi HonidenNational Institute of Informatics,

The University of Tokyo

Tokyo, Japan

Agent-friendly aggregation 2

Agenda

• Sensor networks

• Aggregation

• Agent aggregation specifics

• Problem model: aggregation graph

• Computing a tour

Agent-friendly aggregation 3

Sensor networks

• Sense/measure the environment– Temperature– Sound– Vibration– Pressure– Motion– …

Agent-friendly aggregation 4

Sensor networks

Base station

Agent-friendly aggregation 5

Wireless sensor networks

Base station

Agent-friendly aggregation 6

Example: Sun SPOT Sensors

• Processing– 180 MHz 32 bit ARM920T core - 512K RAM - 4M Flash

– 2.4 GHz IEEE 802.15.4 radio with integrated antenna

• Sensor Board• Battery

– 3.6V rechargeable 750 mAh lithium-ion battery

– 30 uA deep sleep mode

Agent-friendly aggregation 7

Data aggregation

• Severe resource limitations (battery, sending power)

• Often high redundancy of sensor measurements (time and space)

• Aggregate data before sending it to the base station (e.g., AVG, SUM, MIN,…)

• Aggregation tree

Agent-friendly aggregation 8

Aggregation tree

Base station

Agent-friendly aggregation 9

Aggregation using a mobile (software) agent

• Code is sent through the sensor network…

• … runs on (all/some) network nodes …– collects and aggregates data

• … and returns to the base station.

Agent-friendly aggregation 10

Pros and cons of the agent approach

Advantages:• ability to use code /

aggregation function, which is– Application-specific– Dynamic– Non-local

Problems:• Time• Code size• Security• Aggregation tour

Agent-friendly aggregation 11

Pros and cons of the agent approach

Advantages:• ability to use code /

aggregation function, which is– Application-specific– Dynamic– Non-local

Problems:• Time• Code size• Security• Aggregation tour

Agent-friendly aggregation 12

Pros and cons of the agent approach

Advantages:• ability to use code /

aggregation function, which is– Application-specific– Dynamic– Non-local

Problems:• Time• Code size• Security• Aggregation tour

Agent-friendly aggregation 13

What route to take?

• Visit all nodes

• Energy-efficiency– Avoid visiting nodes/edges several times

(possible exception: base station)• Possibly not a tree-like structure!

Agent-friendly aggregation 14

Aggregation tree

Base station

Agent-friendly aggregation 15

Problem modelling

• Sensor network as undirected graph

Base station

Agent-friendly aggregation 16

Problem modelling

• Sensor network as undirected graph

Base station

Agent-friendly aggregation 17

Problem modelling

• Sensor network as undirected graph

Agent-friendly aggregation 18

Assumption

• Graph is known (to base station)

• (i.e. sensors and their adjacency is known)

• … and does not change, static

Agent-friendly aggregation 19

Hamiltonian cycle

• Given a graph G=(V,E)

• Find a cycle visiting all nodes

• Hard problem

Agent-friendly aggregation 20

Travelling Salesman (TSP)

• Given a weighted graph G=(V,E)

• Find shortest tour visiting all nodes

• Compare all Hamiltonian cycles

• Hard problem

Agent-friendly aggregation 21

Hard problems?

• Hard in the worst case• But: there is hope for some graphs;

problems are solvable on average for these instances

• Unit disk model: n nodes are distributed uniformly at random in the unit disk, nodes within distance r (trans-mission radius) can communicate

Agent-friendly aggregation 22

Assumption

• Apart from base station, all sensors can send and receive within the same distance, not possible to adapt signal strength (due to unit disk model)

Agent-friendly aggregation 23

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

1) Remove trees, 2-core remains2) While no cycle is found, backtrack through

different rotations (permutations)1) Take a path from the list of partial paths2) Try to extend it at either side with an unvisited

nodeIf impossible,

1) If cyclic, search for a node with a yet unvisited neighbor (exists due to connectivity)

2) Else, for endpoints, check for another adjacent node on the path and rotate

Agent-friendly aggregation 24

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

1) Remove trees, 2-core remains2) While no cycle is found, backtrack through

different rotations (permutations)1) Take a path from the list of partial paths2) Try to extend it at either side with an unvisited

nodeIf impossible,

1) If cyclic, search for a node with a yet unvisited neighbor (exists due to connectivity)

2) Else, for endpoints, check for another adjacent node on the path and rotate

Agent-friendly aggregation 25

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

Agent-friendly aggregation 26

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

1) Remove trees, 2-core remains2) While no cycle is found, backtrack through

different rotations (permutations)1) Take a path from the list of partial paths2) Try to extend it at either side with an unvisited

nodeIf impossible,

1) If cyclic, search for a node with a yet unvisited neighbor (exists due to connectivity)

2) Else, for endpoints, check for another adjacent node on the path and rotate

Agent-friendly aggregation 27

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

1) Remove trees, 2-core remains2) While no cycle is found, backtrack through

different rotations (permutations)1) Take a path from the list of partial paths2) Try to extend it at either side with an unvisited

nodeIf impossible,

1) If cyclic, search for a node with a yet unvisited neighbor (exists due to connectivity)

2) Else, for endpoints, check for another adjacent node on the path and rotate

Agent-friendly aggregation 28

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

Agent-friendly aggregation 29

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

1) Remove trees, 2-core remains2) While no cycle is found, backtrack through

different rotations (permutations)1) Take a path from the list of partial paths2) Try to extend it at either side with an unvisited

nodeIf impossible,

1) If cyclic, search for a node with a yet unvisited neighbor (exists due to connectivity)

2) Else, for endpoints, check for another adjacent node on the path and rotate

Agent-friendly aggregation 30

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

Agent-friendly aggregation 31

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

1) Remove trees, 2-core remains2) While no cycle is found, backtrack through

different rotations (permutations)1) Take a path from the list of partial paths2) Try to extend it at either side with an unvisited

nodeIf impossible,

1) If cyclic, search for a node with a yet unvisited neighbor (exists due to connectivity)

2) Else, for endpoints, check for another adjacent node on the path and rotate

Agent-friendly aggregation 32

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

Agent-friendly aggregation 33

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

1) Remove trees, 2-core remains2) While no cycle is found, backtrack through

different rotations (permutations)1) Take a path from the list of partial paths2) Try to extend it at either side with an unvisited

nodeIf impossible,

1) If cyclic, search for a node with a yet unvisited neighbor (exists due to connectivity)

2) Else, for endpoints, check for another adjacent node on the path and rotate

Agent-friendly aggregation 34

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

Agent-friendly aggregation 35

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

1) Remove trees, 2-core remains2) While no cycle is found, backtrack through

different rotations (permutations)1) Take a path from the list of partial paths2) Try to extend it at either side with an unvisited

nodeIf impossible,

1) If cyclic, search for a node with a yet unvisited neighbor (exists due to connectivity)

2) Else, for endpoints, check for another adjacent node on the path and rotate

Agent-friendly aggregation 36

Hamiltonian cycle for unit disk graphs (Bollobas et al., Petit)

Agent-friendly aggregation 37

Conclusion

• If agent-based aggregation is benefitial in a sensor network, it can be done quite efficiently.

• (the algorithm of Bollobas et al. quickly computes an energy-efficient tour (a Hamiltonian cycle) in a unit disk graph)

Agent-friendly aggregation 38

Thank you

top related