The goal of the OURAGAN project is to provide access ofmeta-computing resources to Scilab users. We present here an approachthat consists, given a Scilab script, in scheduling and executing thisscript on a heterogeneous cluster of machines. One of the most effectivescheduling technique is called clustering which consists in groupingtasks on virtual processors (clusters) and then mapping clusters ontoreal processors. In this paper we study and apply the clusteringtechnique for heterogeneous systems. We present a clustering algorithmcalled Triplet, study its performance and compare it to the HEFTalgorithm. We show that Triplet has good characteristics and outperformsHEFT in most of the cases
展开▼
