Reactive Scheduling of DAG Applications on Heterogeneous and Dynamic Distributed Computing Systems

Emerging computational platforms enable a set of geographically distributed computers with different capabilities to be linked together and used in a coordinated fashion to solve a parallel application at the same time. Effective scheduling mechanisms are essential to exploit the tremendous potential of computational resources offered by such platforms. We consider the problem of scheduling parallel applications which are often abstracted as directed acyclic graphs (DAGs), in which vertices represent application tasks and edges represent data dependencies between tasks. The core scheduling issues are that the availability and performance of resources, which are already by their nature heterogeneous, can be expected to vary dynamically, even during the course of an execution. This thesis summary presents the main results of the Global Task Positioning (GTP) mapping method, which is based on the cyclic use of a static mapping method over time. We place strong emphasis in three key aspects, which we believe are central to address the dynamic nature of the problem: reactivity, data-aware components and fault tolerance.