Debian for Computational Grids

Among academic groups, the acceptance of computational grids is steadily increasing. The motivation to join a grid is frequently the mere extension of locally available computational power. More and more though it is the internetworking of academic groups with respective locally maintained infrastructures (i.e. databases, special hardware) that is offered directly through their local native interfaces and not by web services. This particularly applies to sciences that are data driven, i.e. physics or biological sciences. Here, groups mutually grant each other access to their compute clusters (sites) and provide storage areas for the distributed and possibly shared access to data.

A computational grid executes arbitrarily heterogeneous and complex workflows. The challenge is to provide as many resources as possible, independently from the executing site, and as homogeneously as possible. The computational load is distributed as jobs, each with specific requirements for runtime environments. For the highest flexibility in assigning jobs to the executing sites and the best-possible use of the available hardware, all software should be made available at all sites in the moment when it is required by a job.

Debian is best-prepared to provide the means for an automated addition of software, particularly when combined with the recent advancements for the virtualisation of machines. With virtual Debian nodes at grid sites, users can reliably combine cross-architecture scripts with architecture binaries or even reliably compile one's code remotely.

This talk elucidates the provisioning of dynamic runtime environments for the Advanced Resource Connector (ARC) grid middleware using Debian Linux packages.