MODELLING MOBILE AGENT APPLICATIONS BY EXTENDED

UML ACTIVITY DIAGRAM

Miao Kang, Kenji Taguchi

Department of Computing,Univerisity of Bradford,Richmond Road, Bradford, West Yorkshire, UK

Keywords: Mobile agent, UML, UML Activity diagram, UML Metamodel

Abstract: Mobile agent technology has gained increasing importance in recent years. However, little work has been

done

in defining notations/languages to capture and model mobile agent applications. This paper presents

extensions of UML activity diagrams for modelling mobile agent applications, which capture specific

features of mobile agents such as mobility, cloning and communications.

1 INTRODUCTION

A mobile agent is a computing entity that can move

around different hosts on the network while carrying

its state and procedures. Mobile agent technology

can be deployed for distributed systems such as e-

commerce, distributed information retrieval and

some other services such as personal assistance and

telecommunication network services.

Industry has made a lot of effort to implement

obile agent programming languages. However

most of the works only address implementation

issues and do not address design issues on the

development of mobile agent applications.

In this paper, we present an effective new

approach t

o model specific features of mobile agents

by extending UML. UML can also be extended or

adapted to a specific method, organization, or user

by the built-in extension mechanisms. Its wide

acceptance in industry is one of our main reasons to

adopt UML as our base language.

UML is also widely accepted in Agent Oriented

Soft

ware Engineering (AOSE). Unfortunately UML

cannot be used in a straightforward way for

modelling agent-based systems, since it lacks basic

notion of agenthood. Hence it needs to be extended

by incorporating new structural elements which

could enhance the base language’s expressive power.

Agent UML (AUML) (Bauer, et. al., 2001) is one of

those approaches which synthesize a growing

concern for agent-based software methodologies

with increasing acceptance of UML for object-

oriented software development.

The notion of locality is a key concept in mobile

agent

languages and systems (Taguchi and Dong,

2002), which poses constraints on communications

between agents and how they can move around

locations. For instance, in the agent-place model,

which is the basic computational model of Telescript

(White, 1996), communications between agents are

only allowed at the same location, which means that

an agent must move to a certain location where the

agent can communicate with other agent.

In this paper, we extend activity diagrams in

UML to m

odel mobile agent applications. Activity

diagrams are mainly used for modelling business

process and are also good at modelling algorithmic

behaviour of a computation. The major extension is

to give a new structural meaning to swimlanes to

model behaviours of mobile agents. This visual

representation makes it easy to capture mobility of

agents. Since moving from one location to another is

simply represented by a transition between two

swimlanes. A metamodel is a model of models. It

aims at modelling, visualizing the system, specifying

the structure and the behaviour of the system and

helping in building the system, documenting the

decisions made while building the system (Booch,

1999). We use the UML metamodel to model the

syntax of our extension in this paper.

In order to demonstrate applicability of our

not

ation, we designed an electronic auction system

in our proposed notation and implemented it in Java

Agent DEvelopment (JADE) programming language

(Bellifemine, et. al., 1999).

This paper is organised as follows: the next

sect

ion presents our new approach by extending

UML activity diagrams to model mobile agent

519

Kang M. and Taguchi K. (2004).

MODELLING MOBILE AGENT APPLICATIONS BY EXTENDED UML ACTIVITY DIAGRAM.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 519-522

DOI: 10.5220/0002635505190522

 SciTePress

applications. The syntax of the extended activity diagram is defined by using UML metamodel in

section 3. We then compare related works and

finally conclude our work and explain our future

plan of our work.

2 OUR APPROACH

We will use UML Activity diagrams to model

dynamic behaviour of mobile agent applications.

UML has extension mechanisms such as tagged

values, stereotypes and constraints. Tagged values

are mainly used to show extra information.

Stereotype allows us to extend the UML elements

that already exist by decorating them with extra

semantics. It is presented within a pair of matched

guillemots.

Specific features of mobile agents, which we

model, are mobility, cloning, messaging between

agents, and the concept of nested places.

We introduce a new stereotype << Host >> +

parameter for swimlane, which represents the

location. The parameter is the unique name (address)

of the location. It should be noted that this is just a

conventional extension of UML. Moving between

hosts can be now captured by a transition between

swimlanes. A transition is represented by an activity

“go to host” (see (1) in the figure 1).

Now we introduce a new notation, Subswimlane

that is not realised by the built-in extension

mechanisms in UML. A subswimlane is also shown

as a swimlane with an icon in the higher right corner

representing a nested swimlane (see (2) in the figure

1). An example of subswimlane is given in (5) in

figure1 in which Shopping Centre accommodates

Directory Service, Ticket Shop and Flower Shop.

Mobile agents may travel among these shops

gathering information or making trade.

We use signal sending and signal receipt in

UML Activity diagram to enable agents to exchange

messages. A new rule is added, which is that

whenever an agent would like to send a message to

another agent, the parameters(message description)

inside these icons must be exactly the same between

signal sending and expecting signal receipt. For

example, a user defined “Query” type signal send,

must have the same type in the expecting signal

receipt (see (3) in the figure 1).

Cloning is another important behaviour of

mobile agents, which includes several subactivities.

This activity allows an agent to clone itself first,

then to send the cloned agent to another host. This

behaviour could be repeated in order to send the

required number of clones. We use subactivity

notation to include subactivities and use Dynamic

Concurrency State to show iterations without having

to construct a loop (see (4) in the figure 1).

We designed and implemented an electronic

English auction system in JADE (Bellifemine, et. al.,

1999). JADE is a FIPA-compliant Agent Platform.

In JADE, the communication between agents will

use the ACLMessage class, in which several

message types are provided so that agents could

understand what kind of message, e.g., INFORM,

REQUEST CONFIRM, they need.

Our system consists of two agents, an auctioneer

agent and a bidder mobile agent. The auctioneer

agent resides at seller’s host and manages an

interface to the seller to show the state of the auction

and controls the whole process. The bidder mobile

agent will circulate round potential bidders while

gathering bids by displaying it on GUI to bidders.

ure 1: Modellin

conce

ts in activit

activit

dia

ram of our new a

roach

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

520

ure 2: Get Available Locations subactivit

and Bid subactivit

There are several repeated activities such as

querying for available bidders’ locations from

management agent (AMS in JADE) in main

platform, querying auction information from

auctioneer agent and bidding with bidders and so on.

So “Get Available Locations” and “Bid” are

modelled as two subactivitites in a swimlane

depicted in figure 2.

3 UML METAMODEL FOR OUR

NEW MODEL ELEMENTS

In the previous section, we have defined a new

stereotype <<Host>> + parameter in order to capture

the notion of location. This is defined in the

metamodel in (1) in the figure 3. Another element

we define is subswimlane, which is syntactically

defined in (2) in the figure 3. The swimlane is also

syntactically defined as in (3) in the figure 3. We use

stereotype and tagged value to present the syntax of

moving from host 1 to host 2 of a mobile agent.

We use two tagged values, “Go”, which is

defined in (4) in the figure 3 and “Clone” (see (6) in

the figure 3) to represent the syntax of move and

clone behaviours ((5) and (7) in the figure 3)

respectively.

Figure 3: Metamodel for our new approach

MODELLING MOBILE AGENT APPLICATIONS BY EXTENDED UML ACTIVITY DIAGRAM

521

4 RELATED WORK

The most relevant work is the FIPA modelling

standard in deployment and mobility of multi-agent

systems (FIPA Modeling TC, 2003), which uses

activity diagrams and deployment diagrams as part

of AUML. AUML deployment diagrams statically

models movement of agents by incorporating a new

association with a stereotype <<moves>> and nods

in a network in which agents are modelled as

software components. Algorithmic behaviour of an

agent is modelled in an ordinary Activity diagram,

but movement is captured by using symbols for a

node and a decision which an agent will take in

order to migrate different nodes in a network.

Lind adopted UML Activity diagrams to model

agent interaction protocol without introducing any

new modelling elements (Lind, 2002). He used

swimlane as a construct which denotes a role of an

agent and synchronous communications by giving

new stereotypes. However, as mobile agent systems

being inherently multi-agent, it is hard to use his

notation for more complex communication patterns,

which require more channels between different

agents.

The most recent approach (Baumeister, et. al.,

2002) related to ours is an extension of UML class

diagram and activity diagrams to model mobile

system. Their notation, which is based on a mobile

calculus mobile ambient (Cardelli and Gordon,

1998), assumes both mobile objects and locations

could migrate to another location. The crucial

difference between ours and theirs is that we use a

different mobility model in which we do not allow a

location to move to another location while some

agents running in parallel.

The biggest advantage of our work is its

straightforward visual representation of mobility of

agent, which helps to capture dynamic and

algorithmic behaviour of mobile agents and to

understand how they interact with each other in

terms of locality.

5 CONCLUSION AND FURTHER

WORK

In this paper, we presented extensions to activity

diagram for modelling mobile agent applications. A

new stereotype <<Host>> + parameter for swimlane,

which represents the location is introduced in order

to capture mobility of agents. Other specific features

of agents such as communications, cloning are

defined by existing model elements with a new rule

of subactivities. We also proposed a new notation

subswimlane in order to model nested places.

One of drawbacks is that an activity diagram

basically models a single multi-threaded agent. Even

our proposed notation clearly visualize locality and

communications between agents, it is hard to verify

all communications and movement are eventually

synchronized between agents. The best way to

overcome this drawback is to translate activity

diagrams into some other formalism such as process

algebra, which provide verification procedure of the

model. Fortunately our concept of location can be

easily mapped into mobile calculi, such as mobile

ambient, which provides verification procedure as

well as semantic foundations of our notation.

REFERENCES

Taguchi, K. and Dong, J. S., 2002. An Overview of Mobile

Object-Z. In ICFEM’02, 4th

International Conference

on Formal Engineering Methods, LNCS2495 144-155.

Booch, G., Rumbaugh, J. and Jacobson, I., 1999. The

Unified Modelling Language User Guide, Addison-

Wesley.

White, J., 1996. Mobile Agents. In Software Agents,

Bradshaw, J., editor, MIT Press 437-472.

Bellifemine, F., Poggi. A., Rimassa, G., 1999. JADE – A

FIPA-compliant agent framework, In the Proceedings

of PAAM’99, London, April 1999, 97-108

FIPA Modeling TC, 2003. FIPA Modeling Area:

Deployment and Mobility.

Lind, J., 2002. Specifying Agent Interaction Protocols with

Standard UML. In

AOSE’01, 2nd International

Workshop on Agent-Oriented Software Engineering

held at the 5th International Conference on

Autonomous Agents,

LNCS 2222 136-147.

Baumeister, H., Koch, N., Kosiuczenko, P. And Wirsing,

M., 2002. Extending Activity Diagram to Model

Mobile Systems. In

NODe’02, International

Conference NetObjectDays, LNCS

2591 278-293.

Cardelli, L. and Gordon, A., 1998. Mobile Ambients. In

FoSSaCS’98, 1st Conference on Foundations of

Software Science and Computation Structure, LNCS

1378, 140-155.

Bauer, B., Muller, J. P., Odell, J., 2001. Agent UML: A

Formalism for Specifying Multiagent Software

Systems. In Proceedings of the First International

Workshop on Agent-Oriented Software Engineering

(AOSE-2000) held at the 22

International

Conference on Software Engineering, Limerick,

Ireland, Springer Verlag.

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

522