Modelling frame losses in a parallel Alternate Frame Rendering system with a Computational Best-effort Scheme

Abstract Virtual reality (VR) surgical training and presurgical planning require the creation of 3D virtual models of patient anatomy from medical scans (e.g. CT or MRI). Real-time head tracking in VR applications allows users to navigate in the virtual anatomy from any 3D position and orientation. The process of interactively rendering highly detailed 3D volumetric data of anatomical models from a dynamically changing observer׳s perspective is extremely demanding for computational resources. We propose a parallel computing solution to this problem, involving a distributed volume graphics rendering system composed of multiple nodes concurrently working on different frames of the output stream, which are later integrated to form the final animation. In this scenario, it is important to consider frame losses generated by their out-of-order arrivals in the output sequence of 2D images. This paper presents a study of frame losses for a distributed graphics rendering system consisting of multiple GPU-based heterogeneous nodes running in a best-effort rendering scheme and applying an Alternate Frame Rendering technique. We describe a mathematical model of frame losses, as well as a performance evaluation comparing model predictions with experimental results.