+ All Categories
Home > Documents > Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach...

Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach...

Date post: 04-Jan-2016
Category:
Upload: brian-carter
View: 212 times
Download: 0 times
Share this document with a friend
Popular Tags:
23
Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University of Hamburg
Transcript
Page 1: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Deployment of Distributed Multi-Agent Systems

Alexander PokahrKarl-Heinz Krempels

Lars BraubachWinfried Lamersdorf

Presentation: Dirk Bade

University of Hamburg

Page 2: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 2

Outline

Motivation

What can be learned from OO-Deployment ?

Requirements

Deployment Reference Model

Prototype Implementation

Conclusion

Page 3: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 3

Observation (1)

• The agent-metaphor is … useful for modeling & building complex applications making complexity manageable at a conceptual level seen as natural successor of the OOP

• Agent-applications have already been developed for various domains, but …

Why aren‘t agent applications widely used ?

Page 4: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 4

Observation (2)

• Development of MAS is … quite difficult and error-prone not based on generally accepted methodologies poorly supported by tools

• Deployment of MAS is … difficult because of software‘s dynamic & distributed nature not supported by systematical guidelines

Our approach:• specify agent-applications at a high level

using constraints to declare system-properties,that have to be fulfilled for the application to work properly

Page 5: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 5

Observation Implication

Constraints• e.g. demand certain agent-roles or

services to be available• interpreted & supervised by deployment environment

High level deployment achieved through …• reference model for launching distributed

multi-agent applications specifies which, how many and in what order

agent instances shall be launched includes generic meta-model consisting of 2 layers

1. definition of agent-types2. ordered composition of agent-instances

applied to a prototype deployment manager

Page 6: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 6

Outline

Motivation

What can be learned from OO-Deployment ?

Requirements

Deployment Reference Model

Prototype Implementation

Conclusion

Page 7: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 7

OO-Deployment

Object Management Group, Deployment:

"The processes between acquisition of

software and the execution of software"

The deployment process phases

• installation• functional configuration• creation of a deployment plan• preparation phase• launching and runtime configuration

Page 8: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 8

Learnt from OO-Deployment (1)

Deployment of agent-based applications is similiar, but more dynamic and flexible

• installation infrastructure- & application specific agent software• functional configuration adjust agent start-parameters & tune number of agents• deployment plan + preparation phase placement of infrastructure- & application-code on nodes• launching and runtime configuration differs to a great extent …

Page 9: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 9

Learnt from OO-Deployment (2)

Launching & runtime configurationProperties of OO-applications

• hierarchical structure• usually launched using a single starting point

Properties of agent-based applications• bundle of autonomous, self-dependent actors more abstract launching-concept necessary

Directly related:Monitoring & dynamic reconfiguration

Page 10: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 10

Outline

Motivation

What can be learned from OO-Deployment ?

Requirements

Deployment Reference Model

Prototype Implementation

Conclusion

Page 11: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 11

Requirements for Launching and Configuring MAS (1)

Terms:• agent type vs. agent instance

type: static implementation parts instance: concrete description of an agent

• society type vs. society instance type: static application properties

composition of agent types + constraints instance: special application configuration

composition of agent instances + constraints

Page 12: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 12

Requirements for Launching and Configuring MAS (2)

Basic Management Services:• starting/stopping of agent- and society instances

on local platform by local platform management on remote platforms by requests to remote service-agents

initiating node

local agents startedby local platform

delegate startupof remote agents

delegate startup of remote subordinated agents

node running an agent-platform

and service-agent

Page 13: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 13

Outline

Motivation

What can be learned from OO-Deployment ?

Requirements

Deployment Reference Model

Prototype Implementation

Conclusion

Page 14: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 14

Deployment Reference Model (1)

Page 15: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 15

<agent name="Host“ package="examples.party“ class="HostAgent“ type="jade">

Deployment Reference Model (2)

Launching, monitoring & reconfiguration-services are based on specification of agents & societies.

The AgentType-model

<parameter name="guestsToWaitFor“ type="String“ optional="false"> <value> 5 </value></parameter>

Page 16: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 16

Deployment Reference Model (3)

The SocietyType-model

<society name="PartySociety" package="examples.party“> <imports> <import>examples.partyservice</import> </imports> <agenttypes> <agenttype name="examples.party.Guest"/> </agenttypes>

<societyinstance name="Small Party"> <agentinstance name="PartyHost" type="Host"> <parametervalue name="guestsToWaitFor"> 10 </parametervalue> <tooloption type="debug">true</tooloption> </agentinstance> <agentinstance name="Guest_1" type="Guest"/>

<societyinstanceref name="MySubSociety" societytype="PartySociety“ societyinstance="Female Guestpool"> <launcher name="ascml@vsispro3:1099/JADE"> <address> http://134.100.11.53:80 </address> </launcher></societyinstanceref>

Page 17: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 17

Deployment Reference Model (4)The Dependency-model <dependency active=“true”>

<agentinstance name="PartyHost"> <provider name="ascml@MyComputer:1099/JADE"> <address> http://192.168.1.2:7778 </address> […]

Page 18: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 18

Outline

Motivation

What can be learned from OO-Deployment ?

Requirements

Deployment Reference Model

Prototype Implementation

Conclusion

Page 19: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 19

Prototype Implementation (1)

ASCML service-agent

• self-contained component

• standardized interface easy porting to other FIPA-compliant platforms

Page 20: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 20

Prototype Implementation (2)

• Currently implemented for JADE- and Jadex platforms

• Currently used & tested by the DFG priority research program SPP 1083

• publicly available at the beginning of 2005 ?

Page 21: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 21

Outline

Motivation

What can be learned from OO-Deployment ?

Requirements

Deployment Reference Model

Prototype Implementation

Conclusion

Page 22: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 22

Conclusion and Outlook

• Agent application deployment should be better supportedesp. configuration and launching issuesconcepts besides agents are needed

(e.g. the notion of an agent application or society)ASCML as a first approach

• Combination with organisational approaches (AGR-Ferber, AUML-Odell)

Page 23: Deployment of Distributed Multi-Agent Systems Alexander Pokahr Karl-Heinz Krempels Lars Braubach Winfried Lamersdorf Presentation: Dirk Bade University.

Page 23

Questions

?

?

?

?

?

?

??

?

??

?

????

?

?

?


Recommended