Performance Evaluation of HPCN Applications

The performance attained by parallel programs executed on multiprocessor sys tems is largely in uenced both by the characteristics of the code and by those of the system architecture Indeed parallel machines are typically message passing systems consisting of hundreds of processing elements Implementing high perfor mance parallel applications is a di cult task and understanding sources of poor e ciency represents a key factor in every evaluation process Performance analysis studies can be carried out by means of an integrated use of statistical and numerical techniques together with visualization methods In this paper the potentialities of this approach are presented by means of a case study dealing with a real kernel code used in weather forecasts