The main objective of this project was to enable the knowledge-based support of workflow construction and execution in a Grid computing environment. In order to achieve this objective the consortium members developed a system that enables users to:

• semi-automatically compose a workflow of Grid services,
• execute the composed workflow application in a Grid computing environment,
• monitor the performance of the Grid infrastructure and the Grid applications,
• analyze the resulting monitoring information,
• capture the knowledge that is contained in the information by means of intelligent agents,
• and finally to reuse the joined knowledge gathered from all participating users in a collaborative way in order to efficiently construct workflows for new Grid applications.

Existing Grid infrastructures support mostly single-site Grid applications and parameter studies. Little attention has been given to truly distributed Grid workflow applications. Even more, hardly any system supports Grid workflows based on the widely accepted Web services standard. In order to go beyond existing ad-hoc approaches, the K-Wf Grid system assists its users in composing powerful Grid workflows by means of a rule-based expert system. All interactions with the Grid environment will be monitored and evaluated using a new monitoring infrastructure that fills the gaps in Grid performance monitoring. The knowledge about the Grid itself is mined and reused in the process of workflow construction, service selection and Grid behaviour prediction. In contrast to much existing work, workflows will be dynamic in nature and faulttolerant.

Common state-of-the-art Grid systems do not use the Grid monitoring information in order to assist the user in constructing workflows for Grid applications. The effort will focus on properties typical for next generation Grids: easy to use, easy to program, based on standards for software and protocols, person-centric, and easy to configure and to manage. This will result in the development of software, which is transparent and reliable, open to wide user and provider communities, persistent, and scalable.

K-Wf Grid closes the currently existing gap between monitoring and workflow construction, assisting users in workflow composition and execution by employing a semantic description of existing services and by recording and evaluating all performed activities for future use. Hence, the system is able to learn how to perform tasks (parts of applications) and subsequently will employ this knowledge in workflow construction.
The system monitors, measures, and analyzes the Grid services of a workflow application. During runtime, important information about performance and fault behaviour is collected. This information is carefully examined in order to improve future composition of Grid workflow applications and their runtime behaviour (e.g. with regard to performance or fault tolerance). The K-Wf Grid system is highly user-friendly in order to simplify the development of Grid applications and to unleash the full potential of Grid infrastructures.

The main scientific and technological achievements of the project are:

1. Development of advanced techniques with the aim of supporting online performance monitoring and analysis for workflow Grid applications. The performance monitoring and analysis services interpret performance data, thus going beyond existing systems, which commonly yield vast amounts of performance data without further analysis.

2. Development of a comprehensive system for monitoring and performance analysis that covers Grid sites, networks, and applications, in contrast to existing work, which concentrates mostly on specific aspects (e.g. monitoring of networks only).

3. Definition and employment of a specification language for performance problems with the goal of identifyingimportant issues that merit the attention of a user or some Grid service (e.g. scheduler).

4. Development of an environment for an automatic and fault-tolerant workflow execution service that maps the workflow descriptions onto available Grid resources and that invokes and controls the processes that are specified in each workflow. The workflow mapping is dynamic and just-in-time.

5. Definition of ontologies for the description of Web and Grid services, workflow jobs and their results.

6. Development of a set of intelligent components, able to use observations, monitoring results and other sources of information in workflow creation, resource usage and Grid application execution.