Insightful Workflow for Grid Computing

The management of distributed scientific computing tasks in Grid environments, such as those operated by the DOE national laboratories, has not been completely addressed by (and sometimes is complicated by) existing workflow management software. The ability to continuously identify critical changes in the condition of grid resources, and to alter workflow accordingly, is required; yet, this capability is not available in existing workflow management solutions. This project will develop a Grid workflow management system that is adaptive, robust, and scalable. The approach will make use of technologies in machine learning, automated planning, and agent-based system design. The system will continoulsy adapt its workflow execution in response to the state of relevant resources; further, it will continously improve its model of when and how to adapt workflow execution by learning models of workflow behavior. Phase I will develop realistic, scientific-computing workflow scenarios in conjunction with Argonne National Laboratory (ANL). Learning and replanning algorthms for workflow analysis and management will be developed, and a limited evaluation of the software will be conducted. Commercial Applications And Other Benefits as described by the Applicant: The workflow management sysem should allow DOE scientists and their collaborators to make better use of computing resources that are distributed throughout the globe.. The system also should be of value to corporations and other organizations that manage computing tasks distributed over a global enterprise