Today, computing grids, that are shared and heterogeneous platforms, are ubiquitous. Furthermore, the need to execute complex parallel applications is growing. The aim of this thesis is to schedule applications modeled by moldable task graphs (the number of processors allocated to each task is fixed by the scheduler) onto computing grids to exploit maximum parallelism, use efficiently the resources and manage the sharing of resources. We chose to design pragmatic heuristics because, although they do not guarantee performance, they can lead to good average performance while computing schedules in relatively short times. Almost all existing heuristics only schedule mixed parallel applications onto homogeneous platforms and they sometimes use inefficiently the resources. Thus, we first studied different scheduling heuristics in the case of homogeneous platforms and propose improvements to have a good compromise between the completion time and the efficiency. We then introduced management of heterogeneity in the proposed heuristic and compared its performances with those of a guaranteed algorithm. Finally, we have taken into account the fact that grids are shared by managing competition between applications. Our approach is to limit the amount of resources that
each application can use to build its schedule. We also proposed
different strategies to determine that resource constraint.

EXPLOR_STEP=$WICRI_ROOT/Wicri/Lorraine/explor/InforLorV4/Data/Main/Exploration

HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 003717 | SxmlIndent | more

HfdSelect -h $EXPLOR_AREA/Data/Main/Exploration/biblio.hfd -nk 003717 | SxmlIndent | more

{{Explor lien
   |wiki=    Wicri/Lorraine
   |area=    InforLorV4
   |flux=    Main
   |étape=   Exploration
   |type=    RBID
   |clé=     Hal:tel-00385329
   |texte=   Scheduling Parallel Tasks on Shared Heterogeneous Platforms
}}