DGSI - D-Grid Scheduler Interoperability

The Project

Most Service Grids share the ability to efficiently distribute user workload to the resources available. This issue, usually generalized under the term Grid Scheduling or Meta Scheduling, is already very diverse within a community: both submitted jobs and available resources differ considerably, to the extent that coordination has to handle specialized knowledge about usage scenarios and infrastructure. This leads to very different, community-specific approaches for the development of Grid scheduling services. The resulting incompatibility on the meta-scheduling level, however, proved to be a major hurdle for the coordinated cooperation of different Service Grids especially when focusing on the overall goal of better resource utilization. Moreover, cooperation on a scientific, cross-disciplinary level, are being impaired as well. As such, two major use cases for temporarily including alien resources into the own community arise: first, the need to cover peak demand, and second, the usage of specialized resources.

The D-Grid Scheduler Interoperability (DGSI) is targeting these use cases with the conception and development of a standards-based interoperability layer for Grid level scheduling in service Grids. By allowing the users of a community to distribute the workload among resources within the management domain of another community while keeping the individual, specialized scheduling solutions being run by the communities, it offers new perspectives for community collaboration, resource sharing, and efficient utilization. For sharing the resources between different communities we follow two different approaches: Activity delegation and resource delegation.

In activity delegation, a Grid scheduler hands over an activity and the management of its execution to the domain of the scheduler of another community. In this way, an incompatibility of the basis middleware of two communities can be bridged orthogonally to the first approach introduced. This approach also requires negotiation between both Grid scheduling services.

Our Mission: Delegation of Resources

In resource delegation, a scheduler can place a resource available within its domain under the exclusive planning authority of another scheduler outside of the community, for a previously negotiated time window. This approach is particularly suited where the partners participating in the negotiation and delegation are using the same or similar base middleware systems, and when the schedulers involved exhibit differences in the area of activity description or management functionality. Examples of this are, for example, the unilateral support of workflows, but also activities that require access to community-specific services.

During the definition phase of the DGSI project two alternative technologies for resource delegation have been identified. Both will be developed and implemented in DGSI by PC2.

The first approach focuses on the management of delegated resources and is exclusively realised on top of the base middleware. The result of a negotiation between two community schedulers is a proxy of the middleware interface to the resources with the following tasks:

• Encapsulation of the delegated resources
• Monitoring of the agreed terms of usage
• Hiding of the security-relevant adaptation and of the delegation of rights

The second approach primarily considers the configuration of the resources to be delegated. Besides the base middleware layer it includes also activities on the layer of the local RMS systems. Here, the resources become available only after the requested middleware has been deployed for these resources, which might include a lightweight local RMS. No modifications are required for the existing middleware systems. The approach is flexible since the resources can be prepared virtually for every middleware. Access rights for the resources can be configured as part of the deployment.