Mobilitools ابزار قابل حمل مبتنی بر استانداردهای OMG MobiliTools: A Mobility Toolbox for Agent Interoperability Based on OMG Standards

Abstract. One of the keys to success for applications of mobile and/or intelligent agents in large-scale systems such as Internet is the ability of heterogeneous agents to cooperate and negotiate, and meet if they are mobile. This requires the adoption of standards for the underlying distributed system to support interoperability in agent management, mobile agent transport and localization, and agent communication transport. This paper introduces MobiliTools, an OMG standardsbased solution to these issues, and show how mobile agent platforms can be built from the combination of three independent components, respectively addressing mobility, communication, and activity concerns. 1 Yet Another Java Mobile Agent Platform? 1.1 A new paradigm for distributed systems Classical techniques for distributed systems are based on client/server, code on demand, and remote evaluation paradigms, which finally result in moving code, and/or data, and/or control, as described in [14]. Now, mobile agents bring everything together into a new paradigm. This paradigm was introduced by Telescript [15] under the name of remote programming, to reduce network load and latency, and to fit temporary network connectivity. As underlined in [9], there is little chance to find a “killer application” of mobile agents, but the paradigm is nice for any distributed application spread in a large-scale dynamic open system, where adaptation capability, through dynamic re-distribution of a set of cooperating agents, is a key to coping with changing hosts and network conditions, or to optimize the execution of distributed services. 1.2 A complex combination This nice anthropomorphic paradigm may not be so easy to handle practically. Besides security issues, which are critical to large-scale applications on the Internet, transparency/reliability, scalability and interoperability are other key challenges. Transparency. Typical mobile agent platforms are built on a centralized programming language, enhanced with remote communication capabilities, and finally completed with mobility features (e.g. Java-based platforms). This final add-on of mobility deeply changes the behaviour of the original programming framework. For instance, many useful JDK packages are not designed for mobility, and transparency to mobility issues arise for any access to resources such as threads, files, sockets… This is the reason why Java-based frameworks include specific models and tools for agent activity, communication and mobility, and specify programming restrictions. For instance, creating threads is discouraged (or forbidden) by Voyager [19] and Grasshopper [17], because the platform needs to tightly manage the agent activity. In some platforms, insufficient or disregarded restrictions can result in unspecified behaviour if mobility occurs while an agent is involved in communication. As a matter of fact, communication has consequences on the agent activity, and mobility has consequences on both communication and activity. Full transparency would consist in having strong mobility as defined in [6], maintaining not only the agent state, but also the state of its activity and of its bindings to resources, including on-going communications. Scalability. Both activity and communication models are of great importance for scalability. Java-based platforms that create (at least) one thread of activity per agent are examples of non-scalability if you imagine hundreds or thousands of agents needing to meet in one place. Communication tools are also determining in scalability. Agents need to communicate locally, to take advantage of the remote programming paradigm, but also remotely, as explained in [13]. Remote communication may be implemented in a number of ways, with more or less state-of-the-art properties in terms of persistence, reliability, guaranty of delivery and causality ([2], [13]). Unfortunately, these outstanding properties generally rely on distributed algorithms introducing scalability limitations. Interoperability. Last, but not least, it must also be considered that mobile agents’ specific properties are dedicated to large-scale, dynamic, open distributed systems (e.g. Internet). In such a context, heterogeneous mobile agents need a common high-level communication language to understand each other, as well as a standardised distributed execution and communication infrastructure to interoperate.