Performance Prediction for Distributed Virtual Reality Applications Mappings on PC Clusters

Abstract. Large and distributed Virtual Reality applications are often built from heterogeneous parallel codes. FlowVR offers a design model to easily compose a loosely coupled VR distributed application. FlowVR highly facilitates the development of this kind of applications making possible the description of their coupling and their mapping on the cluster nodes in-dependently of their codes. Nevertheless, the choice of a good mapping is leaved to the developer skill. In an other hand, VR applications may include lots of different components and clusters may also be composed of numerous and heterogeneous nodes and networks. In this case it seems very difficult to map efficiently a distributed VR application onto a large cluster without tools to help the programmer to analyse his choices. The FlowVR design particularly does not propose any kind of performance model. In this pa-per we present an approach to determine performances of VR applications from a FlowVR network and from characteristics of a cluster. We also give some advices to the developer to design efficient FlowVR mappings. Since FlowVR model is very closed to underlying models of lots of VR codes, our approach can be useful for all designers of such applications.