Load Balancing and Performance of Sort-First Rendering Clusters for Complex Urban Scenes

  • Key Laboratory of VGE, Ministry of Education, Nanjing Normal University, Nanjing 210023, China

Received date: 2013-12-17

  Revised date: 2014-01-24

  Online published: 2014-05-10


With the rapid development of modeling methods, the scale and depth complexity of the 3D urban scenes have increased significantly. One of the methods to increase the system efficiency is to build multi-level 3D models. However, the simplification algorithms of 3D models are unsatisfactory in real-time rendering because of the complexity of 3D urban models. Sort-First rendering clusters are broadly used when single PC cannot meet the requirements of real-time rendering, which is more suitable for high resolution and large screen rendering. The current researches are weak at quantitative analysis and self-adaptive load balancing. To overcome these difficulties, we studied the influencing factors of the performance by analyzing the speed-up ratio model of Sort-First rendering clusters, including node size, rendering resolution, network speed, task partitioning and so on. We also improved an existing load balancing method by using the rendering time and the transmission time as the feedbacks for adjusting the task partition instead of simply using the rendering time. Based on this, a necessary condition to achieve the highest speed-up ratio is proposed in an ideal situation that the load is completely balanced and the triangles are uniformly distributed on the screen. This quantitative expression of the condition may help us find out the bottleneck of the parallel rendering system and rationalize the configuration. We further point out that adding more rendering nodes might not help improve the performance if the rendering load is not very heavy, and the benefits of parallel system will decrease while the node size is large. A real-time parallel rendering experiment based on Sort-First was performed, which included 4 rendering nodes and used a large scale urban model. The results verify the above analysis and indicate that our improvement has increased the performance to a certain extent. Furthermore, our research could provide a reference for building Sort-First rendering clusters to reduce the abuse of the computer resources.

Cite this article

SHAO Hua, JIANG Nan, HU Bin, ZHU Jin . Load Balancing and Performance of Sort-First Rendering Clusters for Complex Urban Scenes[J]. Journal of Geo-information Science, 2014 , 16(3) : 376 -381 . DOI: 10.3724/SP.J.1047.2014.00376


[1] 杨必胜,李清泉,梅宝燕. 3 维城市模型的可视化研究[J]. 测绘学报,2000(2):149-154.
[2] 王育坚,刘治国,张睿哲,等.城市三维建模与可视化应用研究[J].北京联合大学学报(自然科学版),2007(4):49-53.
[3] Zhu Q, Hu M Y, Zhang Y T, et al. Research and practice in three-dimensional city modeling[J]. Geo-Spatial Information Science, 2009,16(1):18-24.
[4] Mao B, Ban Y F, Harrie L. A multiple representation data structure for dynamic visualisation of generalised 3D city models[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2011,66(2):198-208.
[5] Xie J H, Zhang L Q, Li J, et al. Automatic simplification and visualization of 3D urban building models[J]. International Journal of Applied Earth Observation and Geoinformation, 2012(18):222-231.
[6] Wang L Z, Chen D, Deng Z, et al. Large scale distributed visualization on computational grids: a review[J]. Computers & Electrical Engineering, 2011,37(4):403-416.
[7] 熊华.面向并行环境的绘制加速技术研究[D].杭州:浙江大学,2008.
[8] 佘江峰,陈景广,程亮,等.三维地形场景并行渲染技术进展[J].武汉大学学报(信息科学版),2012(4):463-467.
[9] Molnar S, Cox M, Ellsworth D, et al. A sorting classification of parallel rendering[J]. Computer Graphics and Applications, IEEE, 1994,14(4):23-32.
[10] 金哲凡,林海,石教英.数据分布型sort-first 并行图形绘制系统的研究与实现[J]. 计算机研究与发展,2004(2): 376-382.
[11] 彭浩宇,金哲凡,秦爱红,等.复式并行流水线在基于PC集群机的并行绘制中的应用[J].计算机辅助设计与图形学学报,2006,18(10):1581-1586.
[12] Chang H, Lei X Y, Dai S L. A dynamic load balancing algorithm for sort-first rendering clusters[C]. 2009 2nd IEEE International Conference on Computer Science and Information Technology, 2009,515-519.
[13] 何毅权.基于sort first 并行渲染系统的动态负载平衡研究[D].成都:电子科技大学,2011.
[14] 李朝奎,杨偶,吴柏燕,等.大范围3DCM场景实时并行绘制的任务划分及策略[J].地理与地理信息科学,2012(6): 24-27.
[15] Morillo P, Bierbaum A, Hartling P, et al. Analyzing the performance of a cluster-based architecture for immersive visualization systems[J]. Journal of Parallel and Distributed Computing, 2008,68(2):221-234.
[16] Staadt O G., Walker J, Nuber C, et al. A survey and performance analysis of software platforms for interactive cluster-based multi-screen rendering[C]. Proceedings of the Workshop on Virtual Environments 2003, ACM, 2003, 261-270.
[17] Eilemann S, Makhinya M, Pajarola R. Equalizer: a scalable parallel rendering framework[J]. IEEE Transactions on Visualization and Computer Graphics, 2009,15(3):436-452.
[18] 申鼎才,董必昌.并行处理中节点间通信对加速比的影响[J].合肥工业大学学报(自然科学版),2005(7):764-767.
[19] 李文石,姚宗宝.基于阿姆达尔定律和兰特法则计算多核架构的加速比[J].电子学报,2012(2):230-234.
[20] Loewen J, Wardijono B A. Using open source tool for developing 3D graphic simulations on PC-clusters[C]. IEEE International Conference on Signal Image Technology and Internet Based Systems, 2008,580-584.