The SAM-Grid Fabric services

To enable globally distributed computing for a large HEP experiment, a collection of computing and data storage facilities, together called the Grid fabric, must be linked together in a coherent way. The standard Grid software, including most notably the Globus Gatekeeper and Meta Directory Service, provides core tools to insert a site into the Grid and for its low-level monitoring. In practice, large experiments have data and job handling infrastructures that are not governed by the core tools. For example, local job submission is seldom done directly to the batch system, but rather, through an interface that allows for pre-submission steps (such as the decomposition of a job into smaller chunks) or is tightly integrated with a data handling system such as SAM. Likewise, monitoring is seldom done in terms of individual processors or individual jobs, but rather, via cluster-wide aggregated characteristics. In this paper, we present some of the work we have done to abstract the management of the fabric facilities of the FNAL Run II experiments, in order to enable globally distributed computing.