Study how to develop generic scheduling and co-allocation models, mechanisms, and policies for inter- operating federated grids and clouds that is reliable, scalable, cost-aware, and power-aware.
Large-scale distributed systems such as multi-cluster systems, grids, and clouds evolve along multiple dimensions: more heterogeneity in processor types, different ownership and cost models, different types of middleware, larger scale and federations of systems, new faster networks, etc. In order to schedule the resources in these systems, new generic mechanisms and policies for resource discovery, resource selection, and resource (co-) allocation are needed. No single scheduler can provide all this functionality on its own; so multiple cooperating schedulers are required in the large environments of tomorrow.
Develop generic scheduling and co-allocation models, mechanisms, and policies on the basis of the current KOALA scheduler that can deal with resource heterogeneity, multiple types of systems (grids, clouds) in single federations, and that are reliable, cost-aware, power-aware, and can deal with dynamic resource availability. These models and policies will be studied at a more fundamental level, by means of the design of extensions to KOALA, and by means of controlled performance experiments on the DAS and other grids and cloud systems.