SLC : Symbolic Scheduling for Executing Parameterized Task Graphs on Multiprocessors
Résumé
This paper deals with allocating and executing parallel tasks on a distributed memory machine. A parameterized task graph (PTG) is a model of computation which is small and problem size independent (i.e. it requires the same amount of memory, whatever the size of the program parameters is). In this paper we present and study SLC (Symbolic Linear Clustering). SLC is an algorithm that: (1) build a linear clustering for a given parameterized task graph; (2) finds a symbolic allocation of each cluster. This means that the processor where each task is going to be executed is a function that is generated at compile time. We show and study the results of this method on various compute-intensive kernels that can be found in scientific applications.