Runtime scheduling on multiple nodes
Distributing efficiently computations on multiple nodes of a cluster is an ongoing research topic for the last several years. Using a runtime system to assign tasks to different processors of a single node has shown significant benefits in using efficiently the resources of that node. We now need to go one step forward and use the runtime to efficiently use multiple nodes of a cluster. We plan to use StarPU, a task-based runtime system to schedule tasks on multiple nodes based on a distributed shared memory view of the system.