Dynamic Routing of Autonomous Mobile Agentsbillatnapier.com/bcs_agents/BSC_SGAI_2004_09.pdfBCS SGAI,...

34
Dynamic Routing of Autonomous Mobile Agents Christian Erfurth Wilhelm Rossak

Transcript of Dynamic Routing of Autonomous Mobile Agentsbillatnapier.com/bcs_agents/BSC_SGAI_2004_09.pdfBCS SGAI,...

Dyn

amic

Rou

ting

of A

uton

omou

s M

obile

Age

nts

Chr

istia

nE

rfurth

Wilh

elm

Ros

sak

BC

S S

GA

I, C

hris

tian

Erfu

rth

2

Con

tent

Bas

ics

and

Mot

ivat

ion

Aut

onom

ous

Mob

ile A

gent

s -C

once

pts

and

Cha

lleng

es–

Mod

ern

Mob

ile A

gent

Sys

tem

s–

Pos

sibl

e A

pplic

atio

n S

cena

rio

The

Rou

ting

Ser

vice

Pro

Nav

–Th

e Te

chni

cal S

olut

ion

Sum

mar

y

BC

S S

GA

I, C

hris

tian

Erfu

rth

3

The

Age

nt Is

sue

Pla

ceM

ulti

Age

nt S

yste

mA

gent

sS

oftw

are

Age

nts

Age

nt S

yste

mM

obile

Age

nt S

yste

mA

vata

rM

obile

Age

nts

Inte

llige

nt A

gent

sA

gent

Ser

ver

plac

epl

ace

agen

cyM

obile

Age

nts

–S

oftw

are

Ent

ities

/Obj

ects

, whi

ch a

re a

ble

to m

igra

te

BC

S S

GA

I, C

hris

tian

Erfu

rth

4

Initi

al S

tart

ing

Poin

t

A lo

neso

me

agen

tpublic class myAgent extends MobileAgent{

...

public void startAgent() {

...

go(“server.name.edu“,“nameOfMethode“);

}

public void handleMessage(Message msg) {

...

}

...

???

BC

S S

GA

I, C

hris

tian

Erfu

rth

5

Our

Goa

ls

Ena

ble

agen

ts to

be

–m

obile

–au

tono

mou

s–

pro-

activ

e

Use

r foc

uses

on

WH

AT

to d

o

Age

nt “k

now

s“ H

OW

and

WH

ER

E to

do

BC

S S

GA

I, C

hris

tian

Erfu

rth

6

Mod

ern

Mob

ile A

gent

Sys

tem

Firs

t sta

ge–

A s

peci

alis

ed a

gent

has

to b

e pr

ogra

mm

ed to

fulfi

l its

task

App

licat

ion

algo

rithm

–ag

ent’s

per

sona

lity

–Th

e iti

nera

ry n

eeds

to b

eov

erha

nded

to th

e ag

ent

Rou

ting

algo

rithm

Sec

ond

stag

e–

The

itine

rary

is e

stab

lishe

d by

an

agen

t at r

un ti

me

Dec

oupl

ing

of a

pplic

atio

n an

d ro

utin

g al

gorit

hmFo

cus

of a

pplic

atio

n al

gorit

hm

Third

sta

ge–

Sel

ectio

n of

pre

defin

ed m

odul

es fo

r app

licat

ion

algo

rithm

BC

S S

GA

I, C

hris

tian

Erfu

rth

7

Our

Vis

ion

Gen

erat

e a

path

thro

ugh

the

netw

ork

base

d on

use

r‘s

task

spe

cific

atio

n

TASK

plac

epl

ace

agen

cy

plac

epl

ace

agen

cy

plac

epl

ace

agen

cy=>

we

call

it itinerary

BC

S S

GA

I, C

hris

tian

Erfu

rth

8

Our

Vis

ion

(con

t‘d)

Exe

cute

and

dyn

amic

ally

mod

ify th

e iti

nera

ry

plac

epl

ace

agen

cy

plac

epl

ace

agen

cyR

ESU

LT

plac

epl

ace

agen

cy plac

epl

ace

agen

cy

BC

S S

GA

I, C

hris

tian

Erfu

rth

9

TheProNav

Fram

ewor

k

plac

epl

ace

agen

cy

Map

Mod

uleSe

rvic

e?

Map

Itin

erar

y

Map

/Req

uir.

Mirg

.Str

at.

Mig

rati

on P

lan

ner

Itin

erar

y

Rou

te P

lan

ner

BC

S S

GA

I, C

hris

tian

Erfu

rth

10

Rou

ting

Serv

iceProNav

–Th

e Te

chni

cal S

olut

ion

Ope

ratin

g Sy

stem

Java

VM

MAS

Tra

cy

Dom

ain

Serv

ice

Rou

ting

Serv

ice

Appl

icat

ion

Map

Mod

ule

Rou

teP

lan

ner

Mig

rati

onO

ptim

iser

plac

epl

ace

agen

cy

MAS Middleware

Net

wor

k

BC

S S

GA

I, C

hris

tian

Erfu

rth

11

Bas

is: I

nfra

stru

ctur

e –

Trac

y D

omai

n Se

rvic

e

Dom

ain

–Lo

gica

l net

wor

k of

age

nt s

erve

rs (D

omai

n N

odes

) w

ith a

cen

tral m

anag

er (D

omai

n M

anag

er)

Loca

l Dom

ain

Nod

es re

gist

er w

ith th

e D

omai

n M

anag

erD

omai

ns m

ay b

e ne

twor

ked

via

the

cent

ral

Dom

ain

Mas

ter

BC

S S

GA

I, C

hris

tian

Erfu

rth

12

Rou

ting

Serv

iceProNav

–Th

e Te

chni

cal S

olut

ion

Ope

ratin

g Sy

stem

Java

VM

MAS

Tra

cy

Dom

ain

Serv

ice

Rou

ting

Serv

ice

Appl

icat

ion

Map

Mod

ule

Rou

teP

lan

ner

Mig

rati

onO

ptim

iser

plac

epl

ace

agen

cy

MAS Middleware

Net

wor

k

BC

S S

GA

I, C

hris

tian

Erfu

rth

13

1 M

ap M

odul

e

Net

wor

k se

nsin

g m

odul

e at

age

nt s

erve

rs–

Nod

e in

form

atio

nS

ervi

ce o

ffer a

nd s

ervi

ce lo

catio

n, A

vaila

bilit

y,…

–E

dge

info

rmat

ion

Ban

dwid

th, L

aten

cy, …

Dyn

amic

Dom

ain

Map

–Lo

cate

d on

and

mai

ntai

ned

by e

very

nod

e w

ithin

the

dom

ain

as a

loca

l sna

psho

t of t

he n

etw

ork

–In

ter-

dom

ain

exch

ange

of k

now

ledg

e w

ith

sum

mar

ized

(com

pres

sed)

dom

ain

map

s

BC

S S

GA

I, C

hris

tian

Erfu

rth

14

Map

Str

uctu

re

Loca

l par

t–

Det

aile

d in

form

atio

nR

emot

e pa

rt–

Sum

mar

ised

loca

l map

s

BC

S S

GA

I, C

hris

tian

Erfu

rth

15

Eval

uatio

n M

ap M

odul

e

Exp

erim

ents

–To

det

erm

ine

the

qual

ity o

f mea

sure

d da

ta fr

om

sens

ors

–To

get

a fe

elin

g of

map

-abl

e dy

nam

ic b

ehav

iour

–an

d th

e ne

twor

k lo

ad c

ause

d by

the

expe

rimen

ts–

To a

djus

t par

amet

ers

BC

S S

GA

I, C

hris

tian

Erfu

rth

16

The

sens

or‘s

dat

a qu

ality

is q

uite

go

od!

We

are

able

to d

istin

guis

h ne

twor

ks!

Late

ncy

valu

es

BC

S S

GA

I, C

hris

tian

Erfu

rth

17

Som

e pe

aks,

but

the

qual

ity is

goo

d en

ough

!

Pro

cess

orlo

adin

fluen

ces

the

resu

lts

a lit

tlebi

tLate

ncy

valu

es

BC

S S

GA

I, C

hris

tian

Erfu

rth

18

The

sens

or‘s

dat

a qu

ality

is

good

eno

ugh!

It re

flect

s th

e ac

tual

net

wor

k si

tuat

ion.Tr

ansf

er ra

teva

lues

BC

S S

GA

I, C

hris

tian

Erfu

rth

19

The

qual

ity is

goo

d fo

r co

nnec

tions

with

a s

mal

l ba

ndw

idth

!

Tran

sfer

rate

valu

es

BC

S S

GA

I, C

hris

tian

Erfu

rth

20

Eval

uatio

n M

ap M

odul

e

Ove

rload

Exp

erim

ent (

Ban

dwid

th)

FOR every known agency s

DO

do measurement with s;

WAIT duration;

LOOP.

BC

S S

GA

I, C

hris

tian

Erfu

rth

21

Rou

ting

Serv

iceProNav

–Th

e Te

chni

cal S

olut

ion

Ope

ratin

g Sy

stem

Java

VM

MAS

Tra

cy

Dom

ain

Serv

ice

Rou

ting

Serv

ice

Appl

icat

ion

Map

Mod

ule

Rou

teP

lan

ner

Mig

rati

onO

ptim

iser

plac

epl

ace

agen

cy

MAS Middleware

Net

wor

k

BC

S S

GA

I, C

hris

tian

Erfu

rth

22

2 R

oute

Pla

nner

Dyn

amic

Dom

ain

Map

–D

irect

ed g

raph

with

wei

ghte

d ed

ges

Rou

te p

lann

ing

mod

ule

–B

asic

ally

a T

SP

(NP

-com

plet

e)S

impl

e lo

cal o

ptim

isat

ion

algo

rithm

s (2

-Opt

, 3-

Opt

)S

ymm

etric

al v

s. a

sym

met

rical

dis

tanc

e m

atric

es

BC

S S

GA

I, C

hris

tian

Erfu

rth

23

Rou

te P

lann

er (c

ont’d

)

Loca

l opt

imis

atio

n al

gorit

hms

need

a s

tart

tour

–N

eare

st N

eigh

bour

, Ran

dom

, Pat

chC

ombi

ned

TSP

alg

orith

m–

Sta

rt: P

atch

; Ite

ratio

n: I3

-Opt

–R

each

es re

sults

0.5

% a

bove

opt

imum

in a

bout

0.3

s10

0 pl

aces

(asy

m.,

tmat

), P

entiu

m II

333

MH

z, J

ava2

–Ta

kes

calc

ulat

ion

pow

er in

to a

ccou

ntE

very

20m

s it

chec

ks la

st o

ptim

isat

ion

resu

lts

BC

S S

GA

I, C

hris

tian

Erfu

rth

24

The

mor

e di

ffere

nt th

e th

roug

hput

the

mor

e us

eful

the

calc

ulat

ion

Ran

dom

pat

h vs

. cal

cula

ted

path

BC

S S

GA

I, C

hris

tian

Erfu

rth

25

Per

form

ance

of p

ath

calc

ulat

ion

Itera

ted

3-O

pt im

prov

es a

pat

h fa

st!

Mor

e ite

ratio

ns le

ad to

bet

ter r

esul

ts

BC

S S

GA

I, C

hris

tian

Erfu

rth

26

Rou

ting

Serv

iceProNav

–Th

e Te

chni

cal S

olut

ion

Ope

ratin

g Sy

stem

Java

VM

MAS

Tra

cy

Dom

ain

Serv

ice

Rou

ting

Serv

ice

Appl

icat

ion

Map

Mod

ule

Rou

teP

lan

ner

Mig

rati

onO

ptim

iser

plac

epl

ace

agen

cy

MAS Middleware

Net

wor

k

BC

S S

GA

I, C

hris

tian

Erfu

rth

27

3 M

igra

tion

Opt

imis

er

Opt

imis

atio

n–

Tran

smit

only

cod

e un

its w

hich

are

nee

ded

at

rem

ote

agen

cies

=> WHAT

–D

iffer

ent m

igra

tion

stra

tegi

es p

ossi

ble

(pus

h, p

ull

and

varia

tions

; Cod

e se

rver

s) =

> HOW

Trac

y pr

ovid

es a

set

of m

igra

tion

stra

tegi

es–

Cho

ose

a/th

e su

itabl

e on

e–

Ada

pt s

trate

gies

–“D

ynam

ic M

igra

tion

Stra

tegi

es”

BC

S S

GA

I, C

hris

tian

Erfu

rth

28

Mig

ratio

n O

ptim

iser

(con

t‘d)

Age

nt‘s

com

pone

nts

are

sele

cted

by

prob

abili

ty

of u

sage

Cal

cula

tion

of m

igra

tion

times

for d

iffer

ent

stra

tegi

es b

ased

on

map

‘s d

ata

Com

paris

on o

f res

ults

and

sup

pose

s a

stra

tegy

BC

S S

GA

I, C

hris

tian

Erfu

rth

29

TOD

O:

Eval

uatio

n M

igra

tion

Opt

imis

er

Opt

imiz

atio

nis

use

ful i

f:–

Env

ironm

ent‘s

dyn

amic

beh

avio

ur is

not

too

high

–It

can

be u

sed

agai

n an

d ag

ain

–Iti

nera

ry is

kno

wn

(nex

t-hop

-opt

imiz

atio

n)

Furth

er te

sts:

–E

ffici

ency

–E

ssen

tial q

uant

ity/q

ualit

y of

kno

wle

dge

BC

S S

GA

I, C

hris

tian

Erfu

rth

30

Rou

ting

Serv

iceProNav

–Th

e Te

chni

cal S

olut

ion

Ope

ratin

g Sy

stem

Java

VM

MAS

Tra

cy

Dom

ain

Serv

ice

Rou

ting

Serv

ice

Appl

icat

ion

Map

Mod

ule

Rou

teP

lan

ner

Mig

rati

onO

ptim

iser

plac

epl

ace

agen

cy

MAS Middleware

Net

wor

k

BC

S S

GA

I, C

hris

tian

Erfu

rth

31

Trac

y –

The

Und

erly

ing

Mob

ile

Age

nt S

yste

m

Trac

y2–

Trac

y2 d

efin

es o

nly

basi

c co

ncep

ts a

nd fu

nctio

ns

com

mon

to a

ll (m

obile

) age

nt to

olki

ts–

Cor

e se

rvic

es a

re im

plem

ente

d as

low

-cou

plin

g so

ftwar

e co

mpo

nent

s ex

tend

ing

a lig

ht-w

eigh

t m

icro

-ker

nel

–Fl

exib

le fi

ne-g

rain

ed m

igra

tion

prot

ocol

–D

iffer

ent m

igra

tion

stra

tegi

es, e

.g. a

gent

can

ada

pt

mig

ratio

n fo

llow

ing

spec

ific

para

met

ers

such

as

band

wid

th, l

aten

cy, c

ode

size

BC

S S

GA

I, C

hris

tian

Erfu

rth

32

Sum

mar

y

The

Rou

ting

Ser

viceProNav

supp

orts

–A

uton

omou

s se

arch

/mov

emen

t–

Pro

-act

ivity

–D

ynam

ical

ada

ptat

ion/

mod

ifica

tion

of it

iner

ary

–O

ptim

isat

ion

of m

igra

tion

Spl

ittin

g in

to d

omai

ns is

use

ful a

nd n

eces

sary

–N

o A

I is

nece

ssar

y to

cal

cula

te a

pat

h–

mor

e fle

xibi

lity

–S

olve

s th

e qu

antit

y pr

oble

mS

econ

d st

age

mob

ile a

gent

sys

tem

–P

erfo

rman

ce o

f mob

ile a

gent

s is

incr

ease

d qu

alita

tivel

y

BC

S S

GA

I, C

hris

tian

Erfu

rth

33

Futu

re W

ork

Impr

ove

Map

Mod

ule

–Fi

nd m

easu

rem

ent m

etho

ds w

hich

cau

se le

ss n

etw

ork

load

Alte

rnat

ive

infra

stru

ctur

e or

gani

satio

n

Che

ckProNav

with

oth

er (m

obile

age

nt) s

yste

ms

Mor

e co

mpl

ex s

imul

atio

ns (m

ore

node

s)

BC

S S

GA

I, C

hris

tian

Erfu

rth

34

Con

tact

Furt

her I

nfor

mat

ion

www.pronav.info

Than

ks!