AGENT-BASED TELEMATIC SERVICES AND TELECOM APPLICATIONS.
The telecommunications market is expanding rapidly and players in that market are facing increasingly stiff competition. The key to commercial success in the telecommunications market will be the provisioning of adequate services, the focus shifting from a purely technological one to one of convenience and usefulness. An important prerequisite is the effective management of basic telecommunications infrastructure supporting the rapid deployment of new services.
These future services will determine the market shares to be gained. Not only must their time to market be reduced, but also other requirements need to be fulfilled, for example, dynamic service development and configuration. Due to a demand for permanent availability the motto to be followed is "information anywhere, anytime." Service maintenance must not interfere with continuous service usage; future services must allow for personalization, meet security demands, and provide management functionality. Furthermore, asynchronous service usage has to be supported as well as demand-driven service combination and integration. Not the least, services must allow for access independent of specific technologies and terminal equipment.
Next to the technical requirements, new business models must be developed reflecting the fact that various actors in new roles, for example, content provider or application service provider, will need to cooperate and coordinate in order to provide these future services. Companies will provide integrated solutions with their own and third-party services being bundled on their platforms. These platforms will realize required infrastructure functionality and enable various means of access by addressing influencing factors and developments of the future telecommunications market such as consumer devices (mobile phones, Internet phones, PDAs), networks (GPRS, UMTS), languages and software technologies (Java, Jini), consumer demands and trends like convenience of use, mobility, and ubiquitous computing. Each role participating in the future telecommunications world will have specific requirements to such service platforms. These demands differ in the extent of infrastructure being needed for service usage and provisioning, according to different necessities regarding aspects such as security, personalization, asynchronous usage, mobility, device-independency, and supporting tools.
The issues outlined here open up new perspectives for services and applications, especially in the domains of service and network management, electronic business, mobility supporting services, and the so-called "Intelligent Home" .
In order to actually realize such services an adequate technological approach is required. Available technologies such as middleware, content description languages, directory services, and security frameworks cover only parts of the aforementioned requirements. A better solution would be the provisioning of one comprehensive framework. Such an approach is embodied by the service development toolkit JIAC (Java Intelligent Agent Componentware),(1) which will be discussed in more detail here.
The intrinsic features of agents--interactivity, autonomy, reactivity, and intelligence--make agent-based systems a promising approach for the realization of open distributed systems, as future telecommunication applications will be. Agents provide services in a more flexible way: They may dynamically cooperate in order to share resources intelligently or to delegate subtasks for complexity reasons. In order to support the development process of agent-based systems, methodologies and languages for analysis, design, and programming are needed (for further information on agent technology, see the article "An Agent-based Approach for Building Complex Software Systems" in this issue). In our work, we focus on the concept of an integrating toolkit for the particular tasks involved in agent-based development. Such a toolkit consists of four parts: agent architecture, programming language, runtime environment, and development tools.
The agent architecture determines how agents work: It provides generic functionality for information processing, goal-oriented behavior, communication, and local coordination. Based on the architecture's representation schemes, the agent description language (ADL) shall allow for a declarative specification of the agent's specific concepts and capabilities. The compiled ADL-code combined with the agent architecture make up the agent. The purpose of the runtime environment is to provide secure execution of agents as well as additional services like registration, agent migration, and other related telecommunication-specific management functionalities. Tools are settled on top of these concepts: Graphical editors ease the effort of programming, which is further supported by debuggers and monitors, and administration tools allow for setting up, controlling, and maintaining agent applications.
The Agent Architecture
Here, we describe the agent architecture of JIAC, which has been designed to meet the requirements of telecommunications applications by offering openness and scalability by the use of a flexible component framework, allowing for the realization of different agent types by reusable components and the exchange of components at runtime. Interaction protocols allow for negotiations and flexible service usage. Additionally, the integration of telecommunication-specific functionalities by the use of dedicated components is provided, as explained later in this article.
An agent consists of a set of components managed by the component framework. Components are identified by the roles they take within an agent and interact by message passing. Messages are delivered to the component currently associated with the role of the receiver. By this way direct dependencies between components are prevented, allowing for their exchange even during runtime. The component framework is also responsible for managing the processing resources of an agent.
The default architecture realizes a generic control scheme for reactive, deliberative, and interactive behavior following a Belief-Desire-Intention model . It is divided into the component framework, the control unit controlling the behavior according to the knowledge represented in the knowledge base, and the periphery for additional and auxiliary functionalities. However, with the component framework it is also possible to implement other control architectures, for example, purely reactive ones.
The components of the knowledge base are responsible to store and maintain knowledge and include a fact base, a goal stack, an intention structure, a rule base, a plan library, and a service library. The control unit operates on this knowledge. Factual knowledge reflecting the current state of the world is managed by the fact maintenance. The situation assessment reacts to new situations according to rules. For deliberation, intentions, controlled by goals, are selected, coordinated, and executed. The interactive behavior is expressed in sending and receiving speech acts guided by protocols in order to provide or to use services. The periphery contains four types of component roles, comprising application-specific functionalities, speech act transport, as well as management and security tasks.
Agents interact by speech acts compliant to the specification of FIPA ACL . A common terminology is ensured by shared ontologies. All interactions between agents are guided by a generic service scheme. Thus, a service describes an act an agent performs on behalf of another agent. Services are specified by conditions, effects, and protocols. Additional parameters specify payment modalities, security requirements, and human user access. Each service usage is handled by a common metaprotocol. The customer initiates this protocol by requesting the service. If the provider agrees to cooperate--possibly after a negotiation process concerning the service parameters--embedded protocols handle service-specific communications. Finally, the provider transmits the result to the customer.
The telecommunication-specific management framework of the JIAC platform is oriented toward the ITU-T Recommendation X.701. In this standard management activities are divided into specific functional areas: being fault, configuration, accounting, and performance; referred to as FCAE Additionally, we have realized the features according to the proposed standards of , partly as management periphery components, partly as services being offered by dedicated management agents.
Management on marketplaces. JIAC agents reside on marketplaces, referred to as agent platforms in FIPA. A central role on each marketplace is played by the manager agent (MA), which constitutes the role of the FIPA agent platform management agent by providing basic functionality for the operating agents. These management activities comprise the functionalities of the Directory Facilitator, Agent Management System, and Agent Communication Channel. An application may consist of several market places being loosely coupled by the service mediation support of the MAs.
JIAC offers a strong migration concept where code and state of an agent are transmitted. Agent mobility is controlled and supervised by the MAs of the source and destination marketplaces. The process of migration is realized by a dedicated migration protocol that inhibits authentication, authorization, and safe transport of the agent. A localization service allows for tracking mobile agents. On the basis of this functionality, the MA provides a message forwarding mechanism.
Management functionality inside agents. Below the marketplace level, rather elementary FCAP management functionalities are located inside an agent as peripheral components. A wide range of payment models is possible on electronic marketplaces, examples include remittance, credit card, or cyber coins. For supporting the billing of services, JIAC provides a configurable metering model. Each service session is automatically monitored by an accounting process that collects relevant runtime data. Permanent as well as event-driven logging is possible for all JIAC concepts, including knowledge, plan activation, migration, or communication.
JIAC supports property-based configuration of an agent's structure. It is possible to exchange or add components during runtime, thus allowing for a flexible version management of services. One can also change the behavior of a running component by manipulating its property attributes. A fault component provides an error reporting service and allows for manipulation of the knowledge base for error recovery.
The security infrastructure for JIAC offers security mechanisms for agents and marketplaces on different architectural levels: Java Virtual Machine, Transport Layer (ISO/OSI Layer 4), agent, and agent communities. The mechanisms of the Java 2 security architecture are used for securing the Java Virtual Machine. These mechanisms include the use of security managers for marketplaces, class loaders for migration, and access control with permissions and policies. JIAC has no predefined transport layer component. At agent design time, the agent designer adds all needed or wanted communication components to an agent. In order to enable secure communication on the transport layer, JIAC offers an SSL component.
Security functionality inside agents. On the agent level several security features are realized as agent components. JIAC offers a protocol, similar to SSL, which can be used for encrypting and signing speech acts on a higher communication layer than TCP/IP This protocol can be used, if secure communication on the application layer is required. The security protocol precedes the metaprotocol for service usage, as discussed earlier in this article.
With JIAC's service control mechanisms, it is possible to employ and manage access control. This is handled by attaching service control lists to services. In this way, it is possible to restrict service access to a single agent or to an organizational unit specified within a X.509 certificate. Each agent may have a component for the management and handling of this kind of certificates. The certificate management component is a central security component--it is needed by the other security components. JIAC offers the possibility to manage trust relationships between agent places. Trust relationships are entries in a market place trust list and define whether migration from and to other marketplaces is allowed.
Security on marketplaces. Agent marketplaces may contain additional agents, which offer security services. There are three types of security related agents: a certificate authority agent (CA), a security agent, and a security service agent. The CA is employed by a human user to issue certificates on clients' requests. It should always reside on its own secure agent location. The only task of the CA is to push information to security agents. On any marketplace there can be a security agent, which has different tasks to fulfill. First, it functions as a key distribution center for an agent place or a domain of agent places. Second, it manages trust relationships between the agent places it belongs to, and other known agent places. The security service agent is able of creating time stamps, checking signatures, and acts as a trusted third party in a contract conclusion protocol for e-commerce applications.
The toolkit as described so far consists of an agent architecture with certain security and management features and a runtime environment based on a JVM with management and security agents forming marketplaces. Given these elementary components at hand, one is able to create applications that are driven by service-provisioning agents.
Because of the powerful information processing features of the JIAC default architecture, only the implementation of domain-specific knowledge and capabilities is left to the programmer, thereby relieving much of the work. For this purpose, the JIAC Agent Description Language (JADL) has been designed. Due to its declarative concepts, the use of JADL is a process of specification rather than programming. This development process is further facilitated by a set of tools: knowledge compilers, configuration editors, and debugging instruments.
The first step in programming a multiagent system is to define the domain vocabulary as a prerequisite for mutual understanding. To this end, JADL offers language constructs for specifying ontological knowledge, such as categories, attributes, and inheritance. Compilers translate ontological representations into Java code that can directly be processed by the agents' components. In the next step, the initial knowledge, reaction rules, and capabilities comprising plans and services of the agents are specified. This is done declaratively by using a predicate logic style of programming, relying on the previously declared ontologies. Subsequently the procedural functionality of the plans and services must be programmed in Java as periphery components.
Agent Development Environment. The Agent Development Environment (see Figure 1) acts as an integration tool. It manages the compilation of ontologies and knowledge, the configuration of the agents' components, and the initial arrangement of agents on marketplaces. Furthermore, it supports project management and documentation. An agent is generated by specification and configuration of a set of components. The Agent Development Environment treats marketplaces as own entities to reflect the structure of the agent system appropriately. This structure consists of the various marketplaces and the agents being assigned to them.
[Figure 1 ILLUSTRATION OMITTED]
Agent Debugger. The Debugger provides runtime information and control of an agent. Properties of the components can be examined and reconfigured during runtime without the risk of service degradation. A single-step mode allows for controlled execution.
The Debugger logs internal processes and messages of the agent. For this purpose, different levels of detail for logging activities can be set for each component of an agent. It is also possible to inspect the informational state of the agent by looking into the contents of the knowledge bases, and by monitoring the effects of executed plans and services. For controlling the agent's behavior directly, the Debugger allows for setting new goals.
Marketplace Monitor. The activities of the agents on a marketplace are visualized by the Marketplace Monitor (see Figure 2). It shows the agents currently populating the marketplace and their ongoing communications. The symbol of each agent indicates its life-cycle state and whether it's the manager agent, a stationary agent, a mobile agent, or a remote agent being involved in communications from another marketplace.
[Figure 2 ILLUSTRATION OMITTED]
Agent Navigator. The Navigator serves as a generic user interface to access the services of an agent system by a human user. Each Navigator is a manager agent constituting a user marketplace. Via the Navigator, the user can browse the Directory Facilitator by categories and keywords to find services as needed. It handles the initiation and termination of service usage and allows for simultanous service access.
A service is available to human users if an agent provides a graphical user interface (GUI) for service access. Such agents are called Alter Ego because they act in behalf of a user. When the user selects a service, the Navigator uses a generic service for human-agent interaction at the Alter Ego. First, the Alter Ego transfers the GUI to the Navigator that displays it. Thereafter, the protocol of this generic service allows information exchange between the GUI and the Alter Ego during the service usage. The Alter Ego thereby acts as a gateway by translating GUI interactions into agent communications and vice versa.
Remote administration. The administration of the agent system is done remotely by a set of administration services. The access to these services takes place via the Navigator, but is restricted to the system administrators by service control lists.
The administration comprises management functionalities for marketplaces, agents, and services. The administrator can search for available entities of these types, gain runtime information about them, and may configure them. At each marketplace, agents can be created and terminated or sent to another marketplace. For each agent, components and knowledge, including services, can be added, removed, and exchanged.
Summary and Related Work
Services in the telecommunications domain demand short time-to-market spans, a high level of configurability, support for a multitude of devices, network independency, personalization and adaptivity, as well as support for a broad range of functionalities and applications. Here, we have presented the basic concepts of the toolkit JIAC, which relys on a scalable and open component-based agent architecture for service provisioning. Management and security functionalities contribute to the requirements of the telecommunications domain, whereas a set of tools support an integrated and easy development of agent-based applications in this area.
Several implemented prototypes demonstrate the applicability of the toolkit. For example, in a realized traffic telematics scenario, a mobile user is supported before and on his trip with routing and re-routing services, up-to-date traffic information, car status like fuel level and personalized gas station information, and breakdown and emergency support.
BDI-style agent architectures are based on the work of Bratman . Similar to other BDI architectures such as , JIAC doesn't rely on belief-desire-intention logic but rather employs a pragmatic solution of information storage and processing. On the other hand, its modular structure is as well comparable to component-based architectures such as . The use of toolkits in order to support the development process of agent-based systems is widely used; see for example, the ZEUS toolkit .
Further contributions to our work stem from standardization committees, mainly FIPA. From FIPA we adopted not only the communication language ACL, but also the specifications of agent and platform management . JADE  is another Java-based FIPA-compliant agent development framework that provides powerful debugging tools. However, no support for telecommunication-specific management functionality is given in JADE.
Although security issues are commonly regarded as very important in the context of mobile agents, architectures with enhanced security features are sparse. Another mobile agent platform, known as MOLE (see mole.informatik.uni-stuttgart.de) deals mainly with life-cycle aspects, localization, and transactions. Access control lists are a well-known feature of Aglets . In the Ajanta framework  security is managed similarly to our approach by use of credentials and authentication protocols; additionally, it provides a proxy-based mechanism for protected resource access.
(1) This work presents the current state of our ongoing research, which yields significant changes and improvements compared to earlier versions of JIAC, for example, as described in .
[1.] Albayrak, S. Introduction to agent-oriented technology for telecommunications. In S. Albayrak, Ed., Intelligent Agents for Telecommunications Applications. IOS Press, 1998.
[2.] Albayrak, S., and Wieczorek, D. JIAC--An open and scalable agent architecture for telecommunication applications. In S. Albayrak, Ed., Intelligent Agents for Telecommunications Applications. IOS Press, 1998.
[3.] Azarmi, N. and Thompson, N. ZEUS: A toolkit for building multiagent systems. In Proceedings of the Fifth Annual Embracing Complexity Conference (Paris, France, Apr. 2000).
[4.] Bellifemine, F., Poggi, A., and Rimassa, G. Developing multiagent systems with JADE. In Proceedings of the Seventh International Workshop on Agent Theories, Architectures, and Languages (ATAL-2000), (Boston, MA, 2000).
[5.] Bratman, M.E. Intentions, Plans, and Practical Reason. Harvard University Press, Cambridge, MA, 1987.
[6.] Foundation for Intelligent Physical Agents. FIPA 98 Specification, Oct. 1998; www.fipa.org.
[7.] Hayes-Roth, B. An architecture for adaptive intelligent systems. Artificial Intelligence 72, 1995.
[8.] Karjoth, G., Lange, D., and Oshima, M. A security model for Aglets. IEEE Internet Computing (July-Aug. 1997).
[9.] Karnik, N. and Tripathi, A. Security in the Ajanta Mobile Agent System. Technical Report, Department of Computer Science, University of Minnesota, May 1999.
[10.] Rao, A.S., and Georgeff, M.P. BDI-agents: From theory to practice. In Proceedings of the First International Conference on Multiagent Systems (ICMAS-95), San Francisco, CA, 1995.
STEFAN FRICKE, KARSTEN BSUFKA, JAN KEISER, TORGE SCHMIDT, AND RALF SESSELER (firstname.lastname@example.org) are research assistants at the DAI-Laboratory at the Technical University of Berlin.
SAHIN ALBAYRAK (email@example.com) is the founder and scientific head of the DAI-Laboratory at the Technical University of Berlin.
This work was funded by T-Nova Deutsche Telekom Innovationsgesellschaft GmbH.
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Industry Trend or Event|
|Author:||FRICKE, STEFAN; BSUFKA, KARSTEN; KEISER, JAN; SCHMIDT, TORGE; SESSELER, RALF; ALBAYRAK, SAHIN|
|Publication:||Communications of the ACM|
|Date:||Apr 1, 2001|
|Previous Article:||AN AGENT-BASED APPROACH FOR BUILDING COMPLEX SOFTWARE SYSTEMS.|
|Next Article:||COMMUNITY-BASED SERVICE LOCATION.|