mobile agents for adaptive routing presented by hong-jiun chen & manu prasanna gianni di caro...

46
Mobile Agents for Adaptive Routing Presented by Hong-Jiun Chen & Manu Prasanna Gianni Di Caro & Marco Dorigo

Post on 20-Dec-2015

216 views

Category:

Documents


0 download

TRANSCRIPT

Mobile Agents for Adaptive Routing

Presented by Hong-Jiun Chen & Manu Prasanna

Gianni Di Caro & Marco Dorigo

Outline

IntroductionOverview of Routing AlgorithmsCommunication Network ModelAntNetOther Routing AlgorithmsExperiment SettingsExperiment ResultsConclusion

Hong-Jiun

Manu

Introduction

AntNet Real ants have been shown to be

able to find the shortest paths by using only the pheromone trail deposited by other ants

I’m Real Ant

Introduction

AntNet A new routing algorithm for

telecommunication networks An adaptive, distributed, mobile-

agents-based algorithm Apply it in a datagram network

Introduction

Terminology Routing Throughput Delay (Latency)

Introduction

Routing It refers to the activity of building

forwarding tables, one for each node in the network, which tell incoming data which link to use to continue their travel towards the destination node.

Introduction

Throughput It is the number of bits which the

network is able to carry in a given period of time

Introduction

Delay (latency)1. Propagation delay2. Queuing delay3. Processing delay4. Transmission delay: The time

elapsed from the moment the first bit of the message is transmitted till the last bit of the message is transmitted

Outline

Introduction

Overview of Routing AlgorithmsCommunication Network ModelAntNetOther Routing AlgorithmsExperiment SettingsExperiment ResultsConclusion

Routing Algorithm

GoalTo direct traffic from sources to

destinations

1. Network performance 2. Costs

Routing Algorithm

The performance metrics: throughput (bits/second) delay (second)

Static or Adaptive?

Outline

IntroductionOverview of Routing Algorithms

Communication Network ModelAntNetOther Routing AlgorithmsExperiment SettingsExperiment ResultsConclusion

Communication Network Model

Apply on datagram networks without concerning congestion and admission control

FIFO When links resources are available,

they are reserved and the transfer is set up

The time it takes a packet from one node to another depends on its size and the link transmission characteristics

No ACK

Outline

IntroductionOverview of Routing AlgorithmsCommunication Network Model

AntNetOther Routing AlgorithmsExperiment SettingsExperiment ResultsConclusion

2

AntNetDescribe it by 6 simple steps:

A

E

F

C

G

D

S

Dest.

Prob.

NextHop

D 0.50 A

D 0.50 F

E 0.50 A

E 0.50 F

G 0.50 F

G 0.50 A 5

3

2

3

3

4

1

2

5

1

1I’m Forward Ant

1. Forward ant Fsd is launched

2

AntNet

A

E

F

C

G

D

1

2. Ssd (k) is inserted, time elapsed is stored in stack

S

5

3

2

3

3

4

1

2

1

5

5A

0S

2

3

AntNet

2. keep it going to next hop

A

E

F

C

G

D5A

8C

0S

S

1

5

3

2

3

3

4

1

2

1

5

5

6

4

2

3

AntNet

3. A circle is detected

A

E

F

C

G

D5A

8C

11E

15F

18C

0S

S

1

5

3

2

3

3

4

1

2

5

1

4

6

5

2

3

AntNet

3. A circle detected, delete all the nodes in that circle from the stack

A

E

F

C

G

D5A

8C

11E

15F

18C

0S

S

1

5

3

2

3

3

4

1

2

5

1

2

3

AntNet

A

E

F

C

G

D

S

1

5

3

2

3

3

4

1

2

5

1

5A

8C

11E

15F

18C

0S

OLD

3. Start over from the last node without circles

5A

7G

0S

NEW

3

2 4

AntNet

4. Destination node reached

A

E

F

C

G

D5A

7G

9D

0S

S

1

5

3

2

3

3

4

1

2

5

1

1

I’m Backwar

d Ant

AntNet

4. Destination node reached, the ant Fsd generates another backward ant Bds

A

E

F

C

G

D5A

7G

9D

0S

S

5

3

2

3

3

4

1

2

5

1

1

AntNet

5. Backward ant pops its stack to know the next hop node

A

E

F

C

G

D5A

7G

9D

0S

S

5

3

2

3

3

4

1

2

5

1

1

2

AntNet

5. Backward ant pops its stack to know the next hop node

A

E

F

C

G

D5A

7G

0S

S

5

3

2

3

3

4

1

2

5

1

2

1

3

AntNet

5. Backward ant pops its stack to know the next hop node

A

E

F

C

G

D

5A

0S

S

5

3

2

3

3

4

1

2

5

1

3

2

1

4

AntNet

5. Backward ant pops its stack to know the next hop node

A

S

E

F

C

G

D

0S

5

3

2

3

3

4

1

2

5

1

3

2

1

4

AntNet6. Whenever the Backward ant arrives a

node, it updates 2 things: 1. A List Trip(i , i

2)

2. The Routing Table

A

E

F

C

G

D

S

5

3

2

3

3

4

1

2

5

1

3

1

4

AntNet

A

E

F

C

G

D

S

5

3

2

3

3

4

1

2

5

1

1. Change A List Trip(i , i2)

It estimates arithmetic mean values i and associated variances i

2 for trip times from the node itself to all the nodes i in the network

4

3

2

1

AntNet2. Change The Routing Table

A

E

F

C

G

D

S

5

3

2

3

3

4

1

2

5

1

Dest.

Prob.

NextHop

D0.50

A

D0.50

F

E 0.50 A

E 0.50 F

G 0.50 F

G 0.50 ADest

.Prob

.NextHo

p

D0.75

A

D0.25

F

E 0.50 A

E 0.50 F

G 0.50 F

G 0.50 A

OLD

NEW

Outline

IntroductionOverview of Routing AlgorithmsCommunication Network ModelAntNet

Other Routing AlgorithmsExperiment SettingsExperiment ResultsConclusion

Manu

Other Routing Algorithms

Performance Comparisons

OSPFa robust routing protocol used in the internet

BFasynchronous distributed Bellman Ford algorithm with dynamic link metrics

SPFlink state algorithm with a dynamic metric for link cost evaluations

SPF_1F SPF with only 1 step of flooding

Daemon Ideal algorithm

Outline

IntroductionOverview of Routing AlgorithmsCommunication Network ModelAntNetOther Routing Algorithms

Experiment SettingsExperiment ResultsConclusion

Experimental Settings

Topology and Physical properties NFSNET with 14 nodes and 21 links Bandwidth of links = 1.5Mbit/s Link/node fault probability = 0 Local buffer capacity = 1GB Statistical multiplexing

Traffic Patterns

Experimental Settings

Static Model Constant bit rate

Dynamic Model Variable bit rate

Geographical Distribution of Traffic

Experimental Settings

Uniform-deterministic distribution Uniform-random distribution Uniform-deterministic-hot-spots Uniform-random-hot-spots

Outline

IntroductionOverview of Routing AlgorithmsCommunication Network ModelAntNetOther Routing AlgorithmsExperiment Settings

Experiment ResultsConclusion

Experimental Results

Performance of all algorithms near optimal for low and uniform traffic loadsAntNet especially good in CBR caseAntNet algorithm shows overall best performanceDaemon algorithm (used for comparisons)

Outline

IntroductionOverview of Routing AlgorithmsCommunication Network ModelAntNetOther Routing AlgorithmsExperiment SettingsExperiment Results

Conclusion

Conclusion

AntNet shows a robust behaviorReaction time of algorithm is acceptableImpact on network resources is neglectable

Strengths and Possible Weaknesses

Strengths Possible Weaknesses Good idea Nice buildup Time tested idea

(ants have been around for sometime… 80 million years)

Scalability issues are ignored

Setup costs and time?

Feasibility for wireless networks?

New Ideas

AntNet: new algorithm for adaptive routingStigmergy The term is defined in the Oxford English

Dictionary as The process by which the results of an insects activity act as a stimulus to further activity, and is used in the mobile robotics literature to describe activity in which an agent supplies changes to the world architecting its future behavior, usually in a useful way

Relevance to IES

If the goal of AI/Robotics is to make machines as intelligent as humans we should first start with imitating lesser intelligent animals (eg: ants)Social behavior, community behavior, cooperation among ants/bees can be applied easily in robotics

The Ants: A Community of Microrobots

Source: MIT Artificial Intelligence LabGoals push the limits of

microrobotics by integrating many sensors and actuators into a small package

form a structured robotic community from the interactions of many simple individuals

The Ants: A Community of Microrobots

Community behavior: Clustering around food

The Ants: A Community of Microrobots

Questions?