Full mobile agent interoperability in an IEEE-FIPA context

Abstract

The existence of heterogeneous mobile agent systems hinders the interoperability of mobile agents. Several solutions exist, but they are limited in some aspects. This article proposes a full interoperability solution, in the context of the IEEE-FIPA agent standards, composed of three parts. The first part is a simple language-independent agent interface that enables agents to visit locations with different types of middlewares. The second part is a set of design models for the middlewares to support agents developed for different programming languages and architectures. And the third part is a method based on agents with multiple codes and a common agent data encoding mechanism to enable interoperability between middlewares that do not support the same programming languages. Furthermore two agent interoperability implementations, and its corresponding performance comparison, carried out over the JADE and AgentScape agent middlewares are presented.

Introduction

Mobile agents are proactive, reactive, and social autonomous entities with the ability to move between different locations of a distributed system. A high grade of interaction among agents and the environment is required. Nonetheless, this is hindered by the existence of different types of Agent Middlewares (AMs), Programming Languages (PLs), and Computer Architectures (CAs). In practical terms, according to Pinsdorf and Roth (2002), two Mobile Agent Systems (MASs) are interoperable if a mobile agent can interact and communicate with other agents (locally or remotely), and if the agents of one of the systems can migrate to the other, i.e., if they can leave their system and resume their execution in the next one.

Considering Pinsdorf and Roth, 2002, Ametller et al., 2006, four agent interoperability areas can be inferred. Firstly, the execution area. The existence of different PLs and CAs makes it difficult to share a unique executable code format. This is the reason why MASs generally use interpreted PLs running over Virtual Machines (VMs). Secondly, the middleware area. Each MAS uses a specific AM with a specific Application Programming Interface (API), agent management system, and agent life cycle model. Hence, agents of different AMs are usually incompatible, despite some standards that cover some of these aspects have been used and proposed, e.g., the IEEE-FIPA specifications (FIPA, 2002). Thirdly, the communication area. Different AMs usually have different communication mechanisms, each one with their own message structures and delivery methods. Therefore, the communication of agents is not always possible. And finally, the mobility area. An agent migration implies agreement on the steps of the migration process, the set of messages exchanged, and the information included within them. Otherwise the agent would not be able to migrate. In the rest of the article it is considered that full interoperability is achieved when all these areas are covered. Besides, a security (Garrigues et al., 2009) area can also be included, although it is not essential for the full interoperability.

Despite the fact that several individual proposals for the mentioned areas exist, a full interoperability mechanism general enough to cover all the possible combinations of AMs, PLs, and CAs is not easy to achieve. Too many variables are involved. Thus, full interoperability is usually reached assuming some constraints with the intended scope. The most common ones are the entailment to a specific PL and CA, usually Java and its VM (Fortino et al., 2008).

The main contributions of this article pursue a full mobile agent interoperability within an IEEE-FIPA context. They are focused on the middleware, and the execution areas and are combined with the Inter-Platform Mobility Architecture (IPMA) approach (Cucurull et al., 2009), for the mobility area, and the IEEE-FIPA specifications, for the communication and part of the middleware areas. This paper proposes three new solutions to common interoperability issues that can be used in combination or on their own:

• Common Agent Interface (CAI): Interface that sets the basis for a minimal interoperable agent compliant with the IEEE-FIPA specifications and suitable for any PL. Agents can be moved to any AM supporting CAI and the specific PL in which the agent is developed.

• Multiple Execution Environments: The availability of multiple Execution Environments (EEs) for a selection of PLs within an AM allows the local execution and interaction of agents developed in different PLs.

• Multiple Code Agents: Agents with several equivalent versions of their code, each one written in a different PL, and a common data representation mechanism to keep their state. This method enables agents to visit locations that do not support the same PLs.

When these techniques are used simultaneously MASs can achieve a high level of interoperability, although it will be more complex. However the proposal is designed in such a way that all three mechanisms can be used alone or in combination with others. In this way, MAS developers can choose the most appropriate configuration that suits their needs.

A caveat here is that these proposals are only devised to support weak agent mobility (Fuggetta et al., 1998, Bohoris et al., 2000, Cabri et al., 2000). The use of strong mobility has some potential advantages for the agent developer, but it imposes serious restrictions to the agent interoperability because of the strong dependency created with the CA (Cabri et al., 2000). Nonetheless, in practice, weak mobility is the most extended type since it is also valid for any kind of mobile agent application.

In order to complement the theoretical proposals, two interoperability implementations for the JADE (Bellifemine et al., 2006) and AgentScape (Overeinder and Brazier, 2004) AMs have been developed. A detailed performance analysis of both implementations is also presented.

During the last few years several recent and innovative projects related to agents have been initiated, e.g., new mobile AMs (Chen et al., 2008) to solve up-to-date problems, or current work of IEEE-FIPA with OMG to propose new agent standards.¹ However, a wide scale mobile agent deployment has not been ever produced so far. A number of reasons have been used to explain this (Roth, 2004, Gray, 2004, Pinsdorf and Roth, 2002), including security issues, lack of applications, and more important, interoperability problems. This article contributes to the mobile agent deployment precisely with the improvement of the interoperability mechanisms.

The article is organised in eight sections: Section 2 presents a background of the existing agent standards and AMs; Section 3 details the first proposal, a common agent interface in the IEEE-FIPA context; Section 4 refers to the second proposal, which states different models of agent infrastructure to support several PLs; Section 5, refers to the third proposal, which defines agents with multiple codes to also support several PLs; Section 6 compares the interoperability approaches presented in this article with other approaches in the literature; Section 7 details an implementation of part of the proposal and presents a set of performance tests; and Section 8 concludes the article stating the results achieved and the future research to be done.