Based Jpvm Distributed Real-time Visualization System Design

This thesis investigates issues on building system of interactive real-time visualization system on commodity clusters on the basis of JPVM. The goal is to leverage commodity clusters and we focus on real-time 3D rendering as a canonical ICM application。In particular, our goal is to exploit clusters of commodity computers, where each machine is equipped with a hardware graphics accelerator, to improve the real-time rendering performance.While there has been considerable work on the use of clusters to support response time driven computations, ICM applications place new requirements on cluster computing because of their real-time nature: the application presents a continuing sequence frames that must be completed with low latency and low jitter. Additionally, ICM applications are bandwidth intensive because they use media such as video, audio, and computer generated images to interact with users.The challenges for building a distributed 3D rendering system on the commodity platform include:1,how to structure a distributed renderer to leverage hardware-assisted rendering.2,how to control bandwidth demand while supporting the rendering of increasingly complex scenes on moderately sized clusters.3,how to schedule tasks with potentially unpredictable execution times to inimize jitter when the communication cost for dynamic load balancing can be a significant fraction of the frame time。To help meet these challenges, we first explore policies for scheduling similar tasks with unpredictable service demands on a distributed system with the goal of maximizing the probability of meeting a real-time deadline.In particular,We then design a distributed rendering framework (DRRS)targeted specifically to the commodity platform. In particular, we employ a new work partitioning and assignment technique called Image Layered Decomposition (ILD). ILD is compatible with hardware-assisted rendering and its bandwidth requirement is independent of scene.The main work include: (1) Introduce interactive continuous media to distributed real time rendering (2)Explore a novel partitioning technique, ILD, that is compatible with hardwareassisted distributed rendering, decouples communication requirement from scene complexity, and significantly reduces communication requirement. (3) Design a frame of distributed real time rendering system(DRRS).