J. Janecek, P. Mačejko (Czech Republic), and T.M.G. Hagras (Egypt)
DAG, heterogeneous systems, static scheduling, multi-corecomputers.
This paper presents results of our study aimed at behaviour of scheduling algorithms applied to highly heterogeneous systems in which clusters of tightly connected processors, multi-core computers, communicate over the channels producing considerable delays. We restricted our attention to the static scheduling heuristics and we have adapted the standard model of the architecture to the form, which simplifies comparison of schedules for specific networks of multi-core machines. We have been interested primarily in anomalies, i.e. cases when the particular scheduling heuristic fails to pro duce an acceptable schedule. A large amount of schedules for randomly generated application graphs and for graphs which are related to real world problems have been evaluated. Our results proved the ability of task duplication to avoid many anomalies in schedules for networks of multicore machines.
Important Links:
Go Back
