Parallel Remote Rendering of Large 3D Point-Based Models on Mobile Clients

Mobile visualization allows users to visualize data anywhere, anytime, on various mobile clients connected by wireless networks. In this paper, an efficient framework is proposed for the remote rendering of large point-based 3D models represented by QSplats Level-Of-Detail (LOD) on mobile devices. As client-oblivious data model is used, rendering tasks are performed on mobile devices ranging from powerful workstations to PDAs and cell phones. On the server side, the framework renders the scenes of the 3D models via the effective utilization of multicore processor(s). The high-level requirements that guided the formulation of the parallel rendering are (a) the problem domain is highly irregular, motivating the use of low-overhead dynamic load-balancing to effectively utilize the multicore processor(s) and (b) the hidden delays encountered with the multicore processors, e.g. to maintain cache consistency. In this manner, novel dynamic load balancing schemes are introduced to reach the optimum performance of the parallel rendering. These schemes are evaluated through the processing of several 3D models with different sizes. In addition, the hidden delays encountered with the multicore processors are investigated. The proposed framework exhibits remarkable efficiency in rendering 3D models especially for the large and sophisticated ones.