Top Banner
The Multi-agent The Multi-agent System for Dynamic System for Dynamic Network Routing Network Routing Ryokichi Onishi Ryokichi Onishi The Univ. of Tokyo, The Univ. of Tokyo, Japan Japan
27

The Multi-agent System for Dynamic Network Routing

Jan 06, 2016

Download

Documents

The Multi-agent System for Dynamic Network Routing. Ryokichi Onishi The Univ. of Tokyo, Japan. Contents. Related theme and paradigms MANET environment, AntNet, Miner ’ s Model Our proposal multiplying entries, evaluating entries Simulation and result effect of each model and formula - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: The Multi-agent System for Dynamic Network Routing

The Multi-agent System The Multi-agent System for Dynamic Network for Dynamic Network RoutingRouting

Ryokichi OnishiRyokichi Onishi

The Univ. of Tokyo, JapanThe Univ. of Tokyo, Japan

Page 2: The Multi-agent System for Dynamic Network Routing

ContentsContents

Related theme and paradigmsRelated theme and paradigmsMANET environment, AntNet, Miner’s MANET environment, AntNet, Miner’s

ModelModel

Our proposalOur proposalmultiplying entries, evaluating entriesmultiplying entries, evaluating entries

Simulation and resultSimulation and resulteffect of each model and formulaeffect of each model and formula

ConclusionConclusion

Page 3: The Multi-agent System for Dynamic Network Routing

Computer and wirelessComputer and wireless

computer networkscomputer networks- decentralized management- decentralized management

wireless networkswireless networks- centralized management- centralized management

Page 4: The Multi-agent System for Dynamic Network Routing

MANET environmentMANET environment

An epoch-making way of

wireless communication

multi-hopmulti-hop

wirelesswirelessad-hocad-hoc

peer-to-peerpeer-to-peer autonomousautonomous

Page 5: The Multi-agent System for Dynamic Network Routing

How good is the MANET ?How good is the MANET ?

DBDB

interneinternett

DBDB

interneinternett

the MANETthe MANETenvironmentenvironment

the usualthe usualenvironmentenvironment

A basestation isn’t a must for their communicationA basestation isn’t a must for their communication more devices communicate with basestationsmore devices communicate with basestations

Page 6: The Multi-agent System for Dynamic Network Routing

Miner’s Miner’s modelmodel

MA moves to a neighbor MA moves to a neighbor node the latest RA from node the latest RA from destination came through.destination came through.

the latest RA from Nodethe latest RA from Node Q Q

MAMA

Routing AgentRouting Agent

Message AgentMessage Agent

PP

QQ

AADD

CC BB

destinationdestination

sourcesource

Agent-based architectureAgent-based architecturefor wireless networkfor wireless network

Page 7: The Multi-agent System for Dynamic Network Routing

AntNeAntNett

Ants follow & deposit pheromone trails.Ants follow & deposit pheromone trails.

pheromonepheromone frequencyfrequency

Pheromone trails are piled up on the Pheromone trails are piled up on the ground.ground.

quantityquantityThe rich food is, the more ants deposit.The rich food is, the more ants deposit.

freshnessfreshnesspheromone evaporates along time.pheromone evaporates along time.

Agent-based algorithmAgent-based algorithmfor wired networkfor wired network

Page 8: The Multi-agent System for Dynamic Network Routing

Problems of AntNet Problems of AntNet

the blocking problemthe blocking problem If a good route is broken,If a good route is broken,

searching another route needs long time.searching another route needs long time. the shortcut problemthe shortcut problem

Even if a better route appeared,Even if a better route appeared,this new route is seldom discovered.this new route is seldom discovered.

Our routing agents walk randomly,Our routing agents walk randomly,and don’t follow pheromone trails.and don’t follow pheromone trails.

Page 9: The Multi-agent System for Dynamic Network Routing

About our modelAbout our model

algorithm (mind)algorithm (mind) Making good routes in a sense of Making good routes in a sense of

probabilityprobabilityby ants’ path-finding modelby ants’ path-finding model

framework (body)framework (body) A simple decentralized managementA simple decentralized management

by multi-agent systemby multi-agent system

Page 10: The Multi-agent System for Dynamic Network Routing

ContentsContents

Related theme and paradigmsRelated theme and paradigmsMANET environment, AntNet, Miner’s MANET environment, AntNet, Miner’s

ModelModel

Our proposalOur proposalmultiplying entries, evaluating entriesmultiplying entries, evaluating entries

Simulation and resultSimulation and resulteffect of each model and formulaeffect of each model and formula

ConclusionConclusion

Page 11: The Multi-agent System for Dynamic Network Routing

Multiply entries Multiply entries (model example)(model example)

MA moves to a neighbor MA moves to a neighbor node the most EAs from node the most EAs from its destination came its destination came through.through.

MAMA

Explorer AgentExplorer Agent

Messenger AgentMessenger Agent

PP

QQ

AADD

CC BB

destinationdestination

sourcesource

three EAs from Node three EAs from Node QQ

Page 12: The Multi-agent System for Dynamic Network Routing

Multiply entries Multiply entries (table example)(table example)

Pheromone trails are piled up on the ground.Pheromone trails are piled up on the ground. More route information from EAs are held in More route information from EAs are held in

the routing tables.the routing tables.

destdest nextnextNN AA

OO BB

PP nullnull

QQ AARR CC CC

AAnullnullDDAA

nextnext

CCAAnullnullCCAA

nextnext

CCBBnullnullCCDD

nextnext

CCRRAAQQnullnullPPBBOOAANN

nextnextdestdest

multiplied up to 4 entriesmultiplied up to 4 entries

new new oldold

a single entrya single entry

Page 13: The Multi-agent System for Dynamic Network Routing

Evaluate entries Evaluate entries (model (model

example)example)

MA moves to a neighbor MA moves to a neighbor node which has the highest node which has the highest value of information on its value of information on its destination.destination.

Explorer AgentExplorer Agent

Messenger AgentMessenger Agent

PP

QQ

AADD

CC BB

destinationdestination

sourcesource

three EAs from Node three EAs from Node QQ

MAMA

Page 14: The Multi-agent System for Dynamic Network Routing

Evaluate entries Evaluate entries (table example)(table example)

two attached sub-entriestwo attached sub-entries timetime the number of hopsthe number of hops

desdestt

nexnextt

nexnextt

nexnextt

nexnextt

NN AA DD AA AA

OO BB CC CC DD

PP nullnull nullnull nullnull nullnull

QQ AA BB AA AARR CC CC CC CC

destdestnextnext nextnext nextnext nextnext

timtimee

hophopss

timtimee

hophopss

timtimee

hophopss

timtimee

hophopss

NNAA DD AA AA

2828 33 2626 66 2525 33 1717 44

OOBB CC CC DD

2222 99 2121 33 1515 22 1212 33

PPnullnull nullnull nullnull nullnull

nullnull nullnull nullnull nullnull nullnull nullnull nullnull nullnull

QQ AA BB AA AA2828 1313 2626 22 2020 1111 1616 22

RRCC CC CC CC

3030 22 2525 22 2020 33 1010 22

Page 15: The Multi-agent System for Dynamic Network Routing

Evaluate entries Evaluate entries (the way of (the way of evaluation)evaluation)

DDRRh-1h-1RR33RR22RR11SS

destination nodedestination node

Explorer AgentExplorer Agent

h

i EAh

iht

p 1

1

)1(The total reliabilityThe total reliability

source nodesource node

pp::the broken-link ratio a the broken-link ratio a timetime

tt::the time since info. the time since info. gottengotten

hh::#hops to the #hops to the destinationdestination

hhEAEA::#hops EAs move a #hops EAs move a

timetime

tp)1( EAht

p

1

)1(

EAht

p

2

)1(

EAh

ht

p

1

)1(

Page 16: The Multi-agent System for Dynamic Network Routing

Ant metaphor and our modelAnt metaphor and our model

[ [ Ant metaphor ]Ant metaphor ]

Pheromone trails Pheromone trails are piled up on the are piled up on the ground.ground.

Pheromone trails Pheromone trails evaporate along evaporate along time.time.

The rich food is, the The rich food is, the more trails ants more trails ants deposit.deposit.

[ [ Our model ]Our model ]

Each next-node Each next-node entry is multiplied.entry is multiplied.

Next-node info. is Next-node info. is evaluated with evaluated with freshness sub-info.freshness sub-info.

Next-node info. is Next-node info. is evaluated with evaluated with distance sub-info.distance sub-info.

Page 17: The Multi-agent System for Dynamic Network Routing

ContentsContents

Related theme and paradigmsRelated theme and paradigmsMANET environment, AntNet, Miner’s MANET environment, AntNet, Miner’s

ModelModel

Our proposalOur proposalmultiplying entries, evaluating entriesmultiplying entries, evaluating entries

Simulation and resultSimulation and resulteffect of each model and formulaeffect of each model and formula

ConclusionConclusion

Page 18: The Multi-agent System for Dynamic Network Routing

Simulation Simulation (network model)(network model)

400400m squarem square

120m diameter120m diameter

Mobile NodeMobile Node100 [100 [units]units]3.6 [km/hr] const. vector3.6 [km/hr] const. vector60[60[m] radio wave rangem] radio wave range

Explorer AgentExplorer Agent100 [100 [units], move a secunits], move a secmovement history 10movement history 10random movementrandom movement

Gateway NodeGateway Node4 [4 [units], stationaryunits], stationaryinformation sourcesinformation sources60[m] radio wave range60[m] radio wave range

100100mm 100100mm200200mm

1 1 meter = 0.625 milemeter = 0.625 mile

Page 19: The Multi-agent System for Dynamic Network Routing

Simulation Simulation (subject)(subject)

[ Performance Characteristics ][ Performance Characteristics ] ConnectivityConnectivity Route lengthRoute length

[ Compared Models ][ Compared Models ] 1 entry per a destination as Miner’s model1 entry per a destination as Miner’s model 60 entries per a destination as the first 60 entries per a destination as the first

modelmodel 20 entries with 40 sub-entries for 20 entries with 40 sub-entries for

evaluation evaluation as the second modelas the second model

the ideal modelthe ideal model

Page 20: The Multi-agent System for Dynamic Network Routing

Result Result (The average connectivity over (The average connectivity over time)time)

Miner’s modelMiner’s model

the 1the 1stst proposal proposal

the 2the 2ndnd proposal proposalthe ideal modelthe ideal model

getting worse over timegetting worse over time

stable after 50 secondsstable after 50 seconds

Page 21: The Multi-agent System for Dynamic Network Routing

Result Result (The average route length over (The average route length over time)time)

Miner’s modelMiner’s model

the 1the 1stst model model

the 2the 2ndnd model model

the ideal modelthe ideal model

getting worse over timegetting worse over time

stable after 50 secondsstable after 50 seconds

Page 22: The Multi-agent System for Dynamic Network Routing

Result Result (average and standard (average and standard deviation)deviation)

ModelModelConnectivityConnectivity Route lengthRoute length

AveragAveragee

Std Std DevDev

AveragAveragee

Std Std DevDev

Miner’s Miner’s modelmodel 66%66% 9%9% 2.82.8 0.40.4

the 1st the 1st modelmodel 83%83% 7%7% 3.23.2 0.50.5

the 2nd the 2nd modelmodel 93%93% 5%5% 2.72.7 0.30.3

the the ideal ideal

modelmodel98%98% 2%2% 2.12.1 0.20.2

Page 23: The Multi-agent System for Dynamic Network Routing

Result Result (The average connectivity over (The average connectivity over agents)agents)

the 2the 2ndnd model model

approaching to the idealapproaching to the ideal

Page 24: The Multi-agent System for Dynamic Network Routing

ContentsContents

Related theme and paradigmsRelated theme and paradigmsMANET environment, AntNet, Miner’s MANET environment, AntNet, Miner’s

ModelModel

Our proposalOur proposalmultiplying entries, evaluating entriesmultiplying entries, evaluating entries

Simulation and resultSimulation and resulteffect of each model and formulaeffect of each model and formula

ConclusionConclusion

Page 25: The Multi-agent System for Dynamic Network Routing

ConclusionConclusion

We proposed ants’ path finding algorithm We proposed ants’ path finding algorithm suitable for the MANET environment.suitable for the MANET environment.

It was proved that our model was proper, It was proved that our model was proper, because …because … our model showed better performance our model showed better performance

than Miner’s model.than Miner’s model. the more route information were gathered,the more route information were gathered,

the better routing performance was the better routing performance was improved.improved.

Page 26: The Multi-agent System for Dynamic Network Routing

Future WorksFuture Works

breed our model breed our model compare our modelcompare our model

Page 27: The Multi-agent System for Dynamic Network Routing

Thank you very much!Thank you very much!

Please get our paper Please get our paper and other related materials atand other related materials athttp://www.sail.t.u-tokyo.ac.jp/~ryo

[email protected]@sail.t.u-tokyo.ac.jp