Task-oriented engineering of coordinated software systems

In the context of Internet-based applications where heterogeneous, legacy entities should integrate and cooperate, the efficiency of the software production process is a key problem: yet, the classical development cycle and methodologies fall short, calling for ad hoc abstractions, methodologies and tools. In this work, we claim that a task-oriented approach can effectively support the design of highly-interactive applications, enabling even small development teams to afford complex software projects. By explicitly conceiving system engineering in terms of concurrent tasks plus task coordination, this approach promotes the application of innovative management techniques to improve the overall product development process, such as concurrent software engineering. Coordination models and infrastructures for agent-based systems will then be discussed as suitable means to deliver the full potential of this approach, sketching their support to concurrent software engineering techniques and to the development of a collaborative environment for concurrent engineering. 1 SOFTWARE DESIGN AND DEVELOPMENT IN THE INTERNET ERA Today, the limited amount of human resources often makes it necessary to charge a project upon few people, causing the traditional roles to overlap: the same persons can be at the same time designers, developers, and testers of their own creations. Also, the same designer can be asked to develop and implement different parts of the same project at the same time: so, following a clean development process (e.g., based on explicit contracts) may just be a dream. Moreover, in today’s Internet-based world even the smallest organisation cannot help but dealing with new kinds of applications, causing a sort of “silent revolution” to software developers, who are asked to operate on highly interactive, inherently-distributed applications, where their experience and known methodologies may fall short. This is why, apart from new languages and evolving standards, properly handling interaction is probably among the most critical issues. In legacy applications, interaction typically assumes well-known, predictable, forms: user interfaces, peripheral I/O, inter-component messages, signals, etc. There, applications are typically designed according to the classical input/compute/output scheme, which is focused on computation: interaction is assumed to occur in pre-defined, selected points – both in terms of specific code points and of time sequence with respect to computation. Switching to a world of dynamically interacting entities (components, objects, agents) spread over the Internet implies several changes to many key issues of system design. Here are some of the questions to be answered: • How should a single, interactive component be designed? In particular, how should its interaction protocol(s) be defined and tested? • How can we ensure that the whole system behaves as desired? In particular, how can we get the various interaction protocols to cooperate smoothly towards the application goals? • How can we enforce the rules coordinating the various components so as to govern the overall system behaviour and achieve the desired global system