Bringing Knowledge to Middleware - Grid Scheduling Ontology

The Grid paradigm implies the sharing of a variety of resources across multiple administrative domains. Assuming that such an environment is highly dynamic, it is essential to abstract potential drawbacks away from resource users and resource providers. One crucial aspect in designing and operating Grids to gain the respective abstraction is the provision of a sophisticated scheduling and resource management framework. Experience shows that scheduling a single resource like an HPC system is already a challenge, but the co-ordinated scheduling of multiple resources to automatically process a complex work flow is impossible if the capabilities of resources are not a priori known. We propose to make scheduling-specific parts of such knowledge exploitable by introducing a scheduling domain ontology. This ontology provides a common semantic understanding to be shared between the components involved in the scheduling process. By agreeing upon and integrating such an ontology we increase the automation level and make usage and administration of Grids easier.