Virtual Geographical Space visualization based on a high-performance graphics cluster

The technical development from the last few years and the usage of GPU and multi-GPU systems allows the implementation of complex 3D Virtual Geographical Space (VGS) scenarios. However, the high costs involved in the acquisition and maintenance of these specialized architectures represent a major drawback, preventing large-scale access of geographical specialists to this kind of resources. In this research paper we propose a scalable solution for creating a high-performance distributed architecture specialized on spatial data modelling, data processing and data visualization. Our main objective is to use the power of multi-GPU systems and visualization clusters in order to run different complex 3D VGS scenarios, having in mind the maximization of the GPU utilization. We have also performed evaluation studies over the compatibility between the graphics cluster framework that run GPU enabled applications and parallel rendering frameworks, aiming to develop an optimal architecture at a lower cost. The performance is evaluated for various cluster configurations and different spatial data models. We will also discuss briefly about the efficiency of load-balancing in our graphics cluster taking into consideration different combinations of distributed rendering algorithms.