EVALUATION OF RECENT RESEARCH IN MOBILE AGENT

PLATFORMS FOR MOBILE DEVICES

Qusay H. Mahmoud, Zhujun Xu, Xining Li

Department of Computing & Information Science,University of Guelph, Guelph, Ontario, Canada N1G 2W1

Keywords: Mobile agents, mobile devices,

wireless devices, security, scalability

Abstract: Most mobile devices such as cell phones suffer from limited computational resources as well as a small

display size in which to present information to the user. In addition, the wireless environment that devices

operate in suffers from limited bandwidth and frequent network disconnections. Mobile agents, which as

software entities that are capable of roaming the network, do not need much bandwidth. This makes mobile

agents a promising mechanism for the mobile devices. In this paper, we report on recent research that has

been carried out in the area of mobile agents for mobile devices; we classify the several agent platforms that

have been developed according to their underlining approaches and we discuss their advantages and

disadvantages as well as limitations and concerns for embedded mobile agent platforms. In addition, we

present directions for future direction in this area. To conclude, limitation and concerns for embedded

mobile agent platforms are discussed.

1 INTRODUCTION

Mobile devices, such as cell phones, Personal

Digital Assistants (PDAs) and pocket PCs have

gained popularity for their distinguished advantages

in today’s information society. The capability of

wireless access to Internet resources particularly

indicates the potentially explosive growth of

wireless applications on mobile devices. On the

other hand, mobile devices have limited

computational resources such as memory and

processing power, low bandwidth, and a greater

tendency for network errors. Such disadvantages

inevitably have an impact on the response time and

throughput of applications or services that mobile

devices are accessing.

To meet with the demanding

requirements of the

users, wireless applications will have to be: (1) Self-

adaptable in various environments which include the

users, the mobile devices and other resources (e.g.

network); (2) Interoperability with the users and

other mobile devices; (3) Able of proactively

providing and arranging rich personalized

applications or services to the users; and (4) Capable

of overcoming the low bandwidth and slow network

connections;

Mobile agents represent a promising mechanism

hat can help with the development and deployment

of wireless applications with the above capabilities.

Mobile age

nts are autonomous pieces of software

that act on behalf of their creators to proactively

access the applications or services, or solve the

problems for their users across the network. Mobile

agents may have the capability of interacting (i.e.

communicating, cooperating, negotiating) with other

agents. Integrated with the artificial intelligence

technology, agents may easily obtain the ability of

learning from the changes of the environment or

reasoning about the existing organizational

structures. The biggest advantage of mobile agent

technology that may benefit the wireless

environment is the requirement of low bandwidth

and support for disconnected operations. Mobile

agents normally move to remote hosts where they

can collect and process data locally without the risk

of network disconnection. After completing their

tasks, agents return to their home.

Given these advantages, finding how to deploy

agent

platforms on mobile devices is becoming a big

challenge considering that not all mobile devices are

capable of running an agent platform since they do

not have enough processing power.

The rest of this paper is organized as follows:

Sectio

n 2 introduces several approaches to the

development and deployment of mobile agent

331

H. Mahmoud Q., Xu Z. and Li X. (2004).

EVALUATION OF RECENT RESEARCH IN MOBILE AGENT PLATFORMS FOR MOBILE DEVICES.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 331-336

DOI: 10.5220/0002642103310336

 SciTePress

platforms for mobile devices, and discusses their

architectural styles and implementations. The

approaches are compared and contrasted in Section

3. Security concerns and requirements for protecting

mobile devices from malicious agents are presented

in Section 4. Section 5 discusses some scalability

issues. Finally, concluding remarks are discussed in

Section 6.

2 AGENT PLATFORMS FOR

MOBILE DEVICES

A mobile agent platform is a software infrastructure

that implements the mobile agent paradigm. The

platform provides an environment supporting

agents’ creation, migration and execution. In the

application level, the mobile agent platform should

provide sufficient APIs for agent application

developers or programmers to create and test agents

running on it. From the user or agent point of view, a

collection of services or applications should be

provided in a mobile agent system.

The creation, execution, migration and deletion

of agents are costly. This means that running mobile

agent on mobile devices is very resource demanding.

Currently, there are several approaches on the

implementation of deploying agent platforms on

mobile devices [Cosmin, 2003]. In order to run

agents on resource limited mobile devices, most

mobile agent platforms that runs on mobile devices

must have a light-weight architecture which enables

mobile agents to completely run on the devices;

whereas, some platforms are placed on remote

stationary server and mobile devices use remote

mobile agents using a delegation model through

wireless communication. A compromised approach

is also proposed in [Wang, et al, 2003]. From the

agent point of view, these approaches can be

classified into three categories:

Remote Mobile Agent Platforms

In this kind of approach, agents are not executed

on mobile devices. Instead, there is a light-weight

client side application installed on mobile devices

interact with a mediator running on the agent server

located on a stationary computer. The client

represents the user and communicates with mediator

via is an interface for the agent platform. Through

the client, users create and dispatch mobile agents

running or migrating on remote agent servers. After

accomplishing their tasks, states of the agents and

data will be returned to the mobile devices by the

client side application.

Customizable Mobile Agent Platforms

This approach compromises between the outside

agent platform approach and embedded agent

platform approach discussed above. The mobile

devices are categorized based on evaluating the

capability of the computing resources available. The

device is configured with a thick agent client that is

able to execute the mobile agent locally. Devices

with limited resources use a thin agent client which

only manipulates the agent’s data.

Now, we discuss the basic characteristics of the

three approaches as well as their advantages and

disadvantages.

Embedded Mobile Agent Platforms

This approach makes the mobile agent platforms

entirely deployed on the mobile devices. Due to the

limited resource available, the mobile agent platform

must have a light-weight architecture to provide

limited services or applications. To meet the

requirement, many techniques for downsizing the

platform such as code profiling and romizing have

been developed .

2.1 Remote Mobile Agent Platforms

This approach is based on the idea that mobile

devices do not have enough processing power to run

an agent platform locally. Users of such devices

would be able access, use, and dispatch remote

mobile agents running on wired networks. The

MobiAgent system [Mahmoud, 2002] follows this

approach.

The MobiAgent system architecture consists of

four main components: the Clients, the

Communication Manager, the Agent Gateway, and

the MobiAgent Services. These components are

connected by either wireless or wired links. The

Communication Manager acts as a mediator between

the mobile devices and the wired lines. It is

responsible for the interaction between the mobile

devices and other parts of the MobiAgent System.

The Agent Gateway provides an interface to the

Communication Manager and MobiAgent Services

to create and update user profiles. When a user sends

requests to the mobile agents though the Agent

Gateway, the mobile agent will perform its task and

go back to its host, passing the results to the Agent

Gate after accomplishing the job.

In this approach, both mobile agents and

platforms are placed on remote stationary servers.

The only application running on the device is the

Java 2 Micro Edition (J2ME) based Mobile

Information Device Profile (MIDP) MIDlet [J2ME],

which configures the mobile agent. Therefore, any

mobile device supporting J2ME will be able to use

this approach regardless of the power of the

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

332

computing resources available. The connection

between the Clients and the Mediator is HTTP over

a wireless link. The main advantages of this

approach are:

– A light-weight client in the mobile device: By

moving the very resource-demanding mobile

agent platform to the resource-rich server, the

major cost in the device is spent on interfaces.

– Few security issues are required to be

considered: Since mobile agents are running on

remote stationary servers, security issues such as

protecting hosts are the responsibility of the

security manager of the mobile agent system.

– It overcomes low bandwidth and network

disconnection: Once the agent has been

dispatched, the agent repository records the state

changes and related results. When the user re-

connects, she/he can freely check the results.

– The system architecture is not tied to any

agent platform: In fact, any agent platform can

be used in this approach.

The system architecture is suitable for resource-

limited mobile devices, especially the less powerful

cell phones. This approach, however, requires a

stationary bridge between the mobile device and

mobile agents.

2.2 Customizable Agent Platforms

The Mobile Agent Architecture for Heterogeneous

Devices [Wang et al, 2003] is an example of this

approach. The system architecture is composed of

four main parts: the agent system repository, the

general agent sever services, the mediator and the

agent client. The agent system repository is the

central part of the architecture which stores the

information both for agents and their clients. The

repository controls the services of agents such as

registration, locating and management of the agent

lifecycle. The system here evaluates the device

capability and sets up the thick agent client for the

device with powerful CPU and sufficient memory.

The thick agent client can execute the mobile agent

locally. While in the resource-limited device, the

thin agent client is configured that is only able to

manipulate the agent’s data. The mediator consists

of the agent initiator, the agent joiner/splitter and the

mobile agent dispatcher. The agent initiator initiates

the agent clients first time or reconfigures it for

some changes on the resources of device or the

software installed. The mobile agent dispatcher is

responsible for dispatching agents between clients

and servers. The agent joiner/splitter splits or joins

the data and code of the mobile agents during their

migration.

This approach provides a flexible mobile agent

architecture which makes it possible for various

devices to access the mobile agent systems. It is

feasible to install different agent clients based on

their respective capabilities. Also separation of the

agent code and data enables either a full mobile

agent or a light-weight version of the agent (data)

running on the device.

A major disadvantage of this approach is that it

does not support advanced interactions with the user

or computations required at the mobile client. In

addition, switching the client from thick to thin and

vice-versa makes for an inconsistent experience for

the user. The thin clients have to pay expensive bill

for joining or splitting the agent code and data. But

since all these resource-demanding computations are

on the remote stationary hosts and the client is light-

weight, this is not a big issue.

2.3 Embedded Mobile Agent Platforms

In the previous two approaches, the mobile agent

platform is deployed on stationary host in a wired

network. But as mobile devices become more and

more powerful, they will have sufficient computing

resources to run an entire agent platform. Two

standardization efforts, namely FIPA [FIPA] and

OMG/MASIF which aim to help standard agent

platforms to fit mobile devices, are underway. In this

section we describe three agent platforms that are

meant to be embedded on the device.

2.3.1 Tryllian’s Agent Development Kit

Tryllian’s Agent Development Kit (ADK) [Tryllian]

is a complete development and implementation

environment. The ADK has two important

components: the Agent Foundation Classes (AFC)

and the Agent Runtime Environment (ARE).

Applications build the agents on the basis of AFC

and use the ARE to run them. Agent applications

typically consist of several Habitats spread across a

number of servers, each hosting several Rooms. A

Habitat is a collection of Rooms that share a Java

Virtual Machine (JVM). It provides services such as

agent lifecycle management, communication, inter-

Habitat travel, Room and agent persistence and the

basic security model. Agents can only exist in

Rooms. Rooms are the living environments where

agents interact. Agent communication is done

through messaging. The ADK provides a real-time,

peer to peer architecture with dynamic update, load

balancing and fault tolerance.

The significant features of this approach are the

support of strong mobility and the full security

model. Agents can move to other Habitats during

EVALUATION OF RECENT RESEARCH IN MOBILE AGENT PLATFORMS FOR MOBILE DEVICES

333

runtime, taking along with their (new) Java code and

state. The ADK implements the full Java security

model using certificates and permissions. The

system is well protected against malicious local or

foreign agents by strict control of agent permissions.

The ARE protects agents from each other.

Additionally, all the transmissions are under

protection of encryption by using a secure socket

layer (SSL). More powerful features provided in

ADK means more computing resource required. At

present only powerful mobile devices such as

PocketPCs can support the execution of the ADK.

The ADK is FIPA compliant and supports both J2SE

and J2ME platforms. The Tryllian’s Agent

Development Kit is commercially available.

2.3.2 MicroFIPA-OS

FIPA-OS [FIPA-OS] is a component-based toolkit

enabling rapid development of FIPA compliant

agents. MicroFIPA-OS [MicroFIPA-OS] is an agent

development toolkit and platform based on the

FIPA-OS agent toolkit. The target devices of

MicroFIPA-OS are non-resource-constrained

devices. To fit on resource-constrained mobile

devices, MicroFIPA-OS simplifies the FIPA-OS

components. XML parsing and heavy use of threads

are removed. Thread and resource pools are

employed. In addition, MicroFIPA-OS limits the use

of the Java reflection API and dynamic invocations.

The MicroFIPA-OS basically follows the FIPA-

OS architecture and it’s extensible by plugging in

components that either replace or supplement the

architecture [FIPA-OS]. FIPA-OS programming API

is also supported in MicroFIPA-OS with some

limitations pertaining to the supported interaction

protocols. The platform can be run in a minimal

mode, which enables the agent to be smaller by

using some low-level programming API.

Programmers or developers may create their own

task and conversation mangers in this mode, and

better performance can be achieved.

A major limitation of MicroFIPA-OS is its

limited support for mobile devices. At present it is

only suitable for some high-end PDAs.

2.3.3 LEAP

The Lightweight Extensible Agent Platform (LEAP)

[Bergenti et al, 2001] is the most well-developed

agent platform for mobile devices. LEAP is based on

JADE [Bellifemine et al, 1999], which is an agent

platform developed under a grant from the European

Commission. The JADE framework, which is fully

implemented in Java consists of a main container

and several agent containers. In an agent container,

there is a message dispatcher responsible for

delivering messages of agents to other containers

and an Agent Communication Channel (ACC)

responsible for passing messages between different

agent platforms. A main container is a special case

of an agent platform. The main container maintains

special services on the platform such as the Agent

Management System (AMS), the Agent

Communication Channel (ACC) and the Directory

Facilitator (DF) serving as the white and yellow

page directories where agents can request each

other’s address and check whether agents exists

anywhere that perform special services.

LEAP is the result of an EU project of a

consortium consisting of Motorola, BT, Siemens,

and others. The LEAP, combined with JADE,

replaces some parts of the JADE kernel forming a

modified runtime environment that is identified as

JADE/LEAP (“JADE powered by LEAP”). The

JADE/LEAP can be deployed on a wide range of

devices varying from servers to Java enabled cell

phones.

We believe that embedding agent platforms in

mobile devices is the most promising research

direction. The tight integration approach should

yield greater flexibility and be most effective way to

design and control agents. However, there’s a major

limitation in this approach: The low level of

computing resources available, which may directly

limit the agents’ competencies. Mobile agent

systems today are very resource demanding both for

the client and the server. As a result, though mobile

devices are experiencing a continuous growth in

resources and power, some of them still do not have

the capability to execute a mobile agent platform.

For example, the JADE/LEAP platform supports

agent mobility between J2SE and Personal Java

containers, while mobility is not available on MIDP.

In other words, mobile agent technology would not

be applied in the great majority of Java enabled cell

phones on which JADE/LEAP is installed.

3 COMPARISON

This section presents a comparison between the

three different approaches discussed above, namely

“remote”, “customizable”, and “embedded”. The

comparison is based on the following issues:

– Mobile agents. In the “remote” approach, users

run MIDlets on their mobile devices to dispatch

a remote mobile agent using a delegation model.

All agents execute and migrate on stationary

agent servers. The “customizable” approach

enables the entire agents (code and data) run in

Thick Clients, but in the case of Thin Clients,

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

334

only part of the agent (data) can be accessed.

When the thin agent (data) returns to the

stationary server from the mobile device, two

parts will be joined and become a full mobile

agent. In the “embedded” approach, the lifecycle

of the agent including creation, execution,

migration and deletion, is completed on the

mobile device. It’s worth noting that supporting

agent migration is very resource-demanding;

therefore, in the latter two approaches, the agent

mobility is weak in most cases.

– Security issues. Mobile agent security is an

open research area. Security issues include

securing communication, protecting hosts

against malicious agents and protecting agents

against malicious hosts. Implementing a full

security model is very expensive, especially for

the deployment of the mobile agent system on

mobile devices. The major advantage of the

“remote” approach is that it moves the security

requirements, except for protecting network data,

to the security manager of the stationary mobile

agent platform. And securing the communication

via HTTP is relatively easy and less costly. For

the “customizable” and “embedded” approaches,

a minimal security model is supported.

– Targeted devices. The “remote” approach

enables any mobile device to access and use

remote mobile agents as long as it supports

J2ME. In the “customizable” approach, thick and

thin clients are available to a variety of mobile

devices. In the “embedded” approach, however,

the more features are supported the less mobile

devices are able to run the platform.

– Implementation standard. The mobile agent

platform located on the remote stationary server

cam be FIPA or MASIF compliant, but not

mandatory. Most embedded mobile agent

platforms on mobile devices are FIPA or MASIF

compliant and some support both standards. For

example, the Tryllian’s Agent Development Kit

and JADE/LEAP are mainly FIPA compliant.

– Usability. Usability evaluation mainly relies on

the interaction methods between a user and a

wireless application such as screen resolution,

input mechanism and user interaction patterns.

Compared to “remote” and “embedded”, the

“customizable” approach is not satisfactory as it

doesn’t support advanced interactions with the

user or computations required at the mobile

client. In addition, switching the client from

thick to thin and vice-versa make the operation

inconsistent.

– Development. Since the “remote” approach may

use any mobile agent platform available, the key

points of development focus on the non-agent-

platform components such as Communication

Manager and the Gateway, which provide the

interface between the users and the MobiAgent

services. For the “embedded” approach, it’s

very important to make the platform light-weight

and suitable to run on a variety of mobile devices

including cell phones. The “customizable”

approach is complex because of the

configuration of the clients, the implementation

of the agent joiner/splitter and its interaction

with the agent server all lead to heavy

development.

4 SECURITY

The integration of mobile agents into mobile devices

will not be effectives and useful unless effective

security mechanisms for both, mobile agents and

mobile devices are adopted. For instance, current

techniques available are not able to determine

whether a foreign agent is malicious or not, unless

the important data on the mobile devices are

damaged. Security issues in a mobile agent system

can be classified into two main categories: (1)

Protecting agents against malicious platforms; and

(2) Protecting mobile devices and the network

against malicious agents. Protecting mobile agents

against each other is another problem, which might

be ignored if the agents are cooperative agents and

can trust each other.

The malicious host problem, in which a

malicious host attacks a visiting mobile agent is the

most difficult one. In order for an agent to function,

the host on which it executes must be able to interact

with it. Most mobile agent systems rely at least in

part on virtual machines to standardize the execution

environment. However, agents written in scripting

and interpreted programming languages are easily

de-compiled from bytecodes to their original

sources. Therefore, a malicious host may easily steal

private information, modify the agent code or

mislead the agent. At present the security for

protecting mobile devices against malicious agents

is a major concern.

In the remainder of this section we will discuss

the security requirements for protecting mobile

devices against potential malicious agents and offer

possible solutions.

Migration from un-trusted hosts

Mobile agents from un-trusted hosts may carry

valuable or useless information. Existing security

EVALUATION OF RECENT RESEARCH IN MOBILE AGENT PLATFORMS FOR MOBILE DEVICES

335

mechanisms assume that trusted hosts wouldn’t

dispatch mobile agents carrying useless information.

To reduce the potential attacks, the security manager

in the mobile agent platform must be able to accept

or reject migration request from un-trusted hosts.

Special mechanisms also need to be provided to

monitor and fully control the activities of foreign

agents from un-trust hosts.

Authentication

Many mobile agent systems on stationary servers

use user information/profile registry in an external

directory for authentication. However, it’s very

resource demanding and requires a reliable

communication channel, thus not suitable for mobile

devices. To be efficient and safe, the authentication

should be performed locally.

Access control

Some platforms employ strict access controls

while some prefer the flexible access control. In the

former approach all the foreign agents are

considered potentially malicious and very strict

access policy is applied. It reduces the potential

attacks from malicious agents but offers less

flexibility. Mobile agents carry on various tasks and

they may require different access permissions on

resources to accomplish their jobs. For the sake of

extensibility and flexibility, a customizable access

control mechanism is a better choice for platforms

on mobile devices.

Interaction

In order to protect the local resources, providing

some monitoring toolkit will help users deny illegal

or potential harmful behaviors of agents. Since the

mobile devices have limited resources and usability

constraints, simple user interfaces will help improve

the interaction.

5 SCALABILITY

As the number of agent and their activities increases

so does the number of threads. Such increase leads

to more consumption of system resources and may

be the instability of the platform. This may affect the

system performance. To resolve the scalability

problem, several issues need to be considered

[Cosmin et al, 2003], including the suspending of

inactive agents.

When an agent is in an inactive state, it should be

put in a repository and removed from memory. In

other words, the agent is still present and not

consuming resources such as memory and

processing power. If communication is needed with

such inactive agent, it will be automatically

deactivated.

6 CONCLUSION

In this paper we discussed, classified, and evaluated

three main approaches for developing and deploying

mobile agent platforms on mobile devices. Each

approach has its own advantages and disadvantages,

as well as domain-specific applications. It is hard to

say which approach is the most suitable for

deploying mobile agents on mobile devices given

that mobile devices continue to have varying

capabilities such as processing power, memory, and

display size. While the “embedded” approach

sounds the most ideal and promising, several

security issues need to be researched further.

REFERENCES

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

FIPA-Compliant Agent Framework. In PAAM’99,

Conference on Practical Applications of Intelligent

Agents and Multi-Agents, London, UK.

Bergenti, F., Poggi, A., 2001. LEAP: A FIPA Platform for

Handheld and Mobile Devices. In ATAL’01, 18

International Workshop on Agent Theories,

Architectures, and Languages, Seattle, USA.

Cosmin, C., Boissier, O., 2003. Multi-Agent Platforms on

Smart Devices: Dream of Reality? In SOC’2003,

Smart Objects Conference, Grenoble, France.

FIPA:

http://www.fipa.org

FIPA-OS:

http://fipa-os.sourceforge.net

J2ME:

http://java.sun.com/j2me

Mahmoud, Q.H., 2002. An Agent-based Approach to the

Wireless Internet. In Journal of Internet Technology,

Vol. 3, No. 2.

MicroFIPA-OS:

http://www.cs.helsinki.fi/group/crumpet/mfos

Tryllian Agent Development Kit:

http://www.tryllian.com

Wang, A.I., 2003. A Mobile Agent Architecture for

Heterogeneous Devices. In WOC’03, 3

IASTED

International Conference on Wireless and optical

Communications, Banff, Canada.

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

336