رویکردهای ترکیبی عامل گرا مبتنی بر خدمات جریان کار متقابل سازمانی

With the sophistication and maturity of distributed component-based services and semantic web services, the idea of specification-driven service composition is becoming a reality. One such approach is workflow composition of services that span multiple, distributed web-accessible locations. Given the dynamic nature of this domain, the adaptation of software agents represents a possible solution for the composition and enactment of cross-organizational services. This paper details design aspects of an architecture that would support this evolvable service-based workflow composition. The internal coordination and control aspects of such an architecture is addressed. These agent developmental processes are aligned with industry-standard software engineering processes.

Online businesses are beginning to adopt a developmental paradigm where high-level component-based services and semantic web services [37] are becoming sufficiently modular and autonomous to be capable of fulfilling the requirements of other businesses. We use the term, services-based cross-organizational workflow (SCW) [5], to describe the workflow interaction that occurs when one business incorporates the services of another within its own processes (also described as business-to-business (B2B)). This term is sometimes associated with the idea of a third-party organization that composes the services of multiple businesses, which is also described as virtual enterprise [15]. In general, the major problems in this domain relate to the dynamic and distributed nature of the Internet environment. In this environment, business processes and the underlying services are constantly removed and updated. It is a major problem to create systems that operate with respect to these dynamic conditions. A second major issue is related to the distribution of services. Since services are distributed across physical and geographical boundaries, any solution architecture must support an equivalent degree of distribution. Although there are several related projects that define solutions to the problems of cross-organizational workflow (which will be described in more detail in Section 2.3), a distinguishing innovation of this work is the use of the autonomy of agent technologies. In applying agent technologies to the problems related to dynamism and distribution, an agent architecture and design is introduced which places emphasis on the specific roles, responsibilities, and actions of the individual agents. A further contribution of this work is the specification and programming of the control mechanisms internal to the agents. A unique feature introduced here is the integration of this specification approach with current, industry-standard software engineering processes and methodologies. Considering the dynamism of the Internet environment, these specification-driven approaches are essential for the dynamic reconfiguration that results from distributed service and process changes. The paper proceeds in Section 2 with an overview and motivation of the cross-organizational workflow domain with respect to the integration of distributed services. In Section 3, the Workflow Automation through Agent-based Reflective Processes (WARP) is introduced to support the SCW domain. In Section 4, there is a discussion of the integration of independent services and, in Section 5, the agent-based modeling approach to support the community of services in the SCW domain. Section 6 contains details of the general agent interaction protocols that support this environment. Finally, Section 7 discusses the WARP prototype and its performance.

8. Discussion and conclusion The WARP approach to service composition extends the state of the art discussed in Section 2.3. The work of Mennie, CHAIMS, and Chakroborty [9] and [33] successfully support base-level composition approaches, however consideration of complex workflow interactions is not evident. These approaches tend to be formal, but consider composition at a relatively atomic level. The WARP approach extends the work in these projects by investigating complex workflow interactions. Another variation is their sole use of text-based specification, while the WARP approach uses both visual (software engineering standard) and text-based specifications. The work of Casati and Benatallah [1] and [8] consider more complex workflow-oriented interactions with visual and text-based specification approaches. Both adopt formal approaches to the service composition, but have non-agent-oriented interactions in the operation of their respective internal architectures. The WARP approach extends the approaches of Casati and Benatallah by considering both the complex workflow-oriented interactions among the services, and the complex interactions of agents internal to our architecture. The work of Helal et al. [20], Chen et al. [10], and Singh [30] concentrate on agent architectures for service composition and discovery. Though these approaches are promising with respect to their problem domains, their service composition specification languages are neither text-based nor were they intended to address the complex workflow configuration issues supported in the WARP approach. Another innovation in the WARP approach is the software developmental process (as in Fig. 14) in the creation of these composite systems that respect workflow protocols. This software development lifecycle is consistent with industry-standard software engineering lifecycles. In fact, the use of industry-standard modeling methodologies (i.e. UML) assists the integration into industry processes. In experiments of the WARP prototype, we determined the system latency of the WARP approach was comparable to related middleware approaches. These performance results verify the appropriateness of the WARP approach considering the major increase in operational flexibility with overhead comparable to related approaches. The WARP approach is one of few projects that use an industry-standard developmental process and modeling techniques. By using multiple UML-based views and software agents to extract the operational data, service evolution and process evolution can occur independently. Furthermore, a workflow designer can control this change by visually modeling the workflow design in available object-oriented software engineering tools. This work also presents a systematic method of combining the UML-based workflow representations to configure an agent-based architecture and relational database for the management of workflow. However, there were several issues encountered in this approach that are the source for future investigations. As briefly discussed in earlier sections, we have started investigating agent approaches for the major problem of data integration and modeling. Since input/output messages can be represented in heterogeneous formats, such as plain text, concrete objects, or XML, modeling approaches and agent support have a formidable challenge to create operational patterns to deal with these formats. In future work, we intend to investigate DAML-S as a solution to the data management issues. Another problem is support for error-handling. An inefficient approach taken in this work is to create new models for each error-handling case. These cases can be more efficiently wrapped into the general operational models (i.e. Control Model and Role Collaboration Model). Future work would consist of further evaluating this approach with advanced workflow patterns and interactions while also incorporating UDDI and SOAP technologies.