摘要
随着社会的发展进步,卡通动画逐渐成为一种重要的休闲娱乐和教育手段。动画作为一种独特的视觉艺术形式一直受到广泛欢迎。在进行三维创作过程中,动画设计者往往使用图形工作站来完成渲染的任务,而单机渲染视图和动画时所需时间往往让人难以忍受。在此研究背景下,本文集中研究分布式集群渲染环境下,有效地按照用户需要提供高质量的渲染服务问题。
本文首先给出研究背景和意义,分析了分布式集群渲染调度技术,讨论集群渲染调度现状和发展。在此基础之上,设计分布式集群环境下的动画渲染系统,描述了系统各部分功能,给出相关工作流程。
针对分布式集群调度问题中参数较多,集群资源属性较多,并且两组参数不存在线性关系等特点,本文给出一种基于神经网络的参数映射方法,建立参数与资源属性间的非线性对应关系,利用得到的应用资源距离对集群候选资源进行筛选,确定最佳候选资源集。
任务调度是分布式集群系统中的一个关键问题。合理的分布式系统的任务调度算法对发挥系统的并行性能、保证任务的可调度性以及保证网络的负载平衡都具有非常重要的意义。通过对常用分布式集群任务调度方法进行分析,设计应用于实际分布式集群渲染系统的基于剩余时间的动态负载平衡渲染调度方法。仿真实验验证了此方法提高了系统的性能,缩短了任务的执行时间,达到了更好的效果。为了提高分布式集群系统的可靠性,增加系统的容错能力,使系统在局部出错的情况下仍能稳定正常运行。本文建立一个容错系统模型,该模型采用两级容错机制,即节点级容错机制和任务级容错机制。此模型对分布式集群系统下的容错问题进行进一步研究建立了基础。
With the development of the society, cartoon has become an important way of entertainment and education. As a particular visual art, cartoon is always popular. In the process of building 3D cartoons, cartoon designers often use graphic workstation to finish the Rendering tasks, however, the time of single-machine Rendering views and cartoons are always unbearable. On the basis of this research background, the thesis studies the problem of providing high quality Rendering service according to users' requirements in the situation of distributed cluster Rendering.
The research background and significance are showed in the paper. The author analyses the scheduling technology of distributed cluster Rendering, discusses the status and development of cluster Rendering scheduling. Based on that, cartoon Rendering system in the distributed cluster is designed, the system's function module is described, and the process of this system is also given.
The parameters in the distributed cluster scheduling problem, the cluster resource attributes, and there is not linear relation between the two parameters. Therefore, a parameter mapping method based on Fuzzy Neural Network is proposed. This method establishes the non-linear relation between parameters and resource attributes. Using the resource distance, the candidate resource in the cluster is filtered and the set of best candidate resource is confirmed.
The task scheduling is the key problem in the distributed cluster system. The reasonable task scheduling algorithm in distributed system is very important to exert the parallel capability and guarantee the task schedulabiliy and load balance. By the analysis of common scheduling algorithms in distributed cluster, the Rendering scheduling method for dynamic load balance based on residual time applied in the distributed Rendering cluster is designed. The result of emulation experiment shows the system capability is raised, the execution time is shortened and the better effect is realized. For the purpose of improving the reliability of distributed cluster system, the fault tolerance ability is added to the system. This ensures the system will also work steadily normally on the condition of partial fault. This paper establishes a fault tolerance system model. There are two-level fault tolerance mechanism in the model, that is, node fault tolerance and task fault tolerance. This model provides the basis of the further research on fault tolerance in the distributed cluster system.
引文
[1]中国青年报创业版.我国动漫产业未来将至少有1000亿产值[EB/OL]http://chuangye.cyol.com/content/2007-O8/20/content_1865875.htm.
[2]火星时代社区.渲染农场、渲染服务解析[EB/OL]http://bbs.hxsd.com.cn/printthread.php?t=9877616.
[3]徐文莉.三维集群渲染负载平衡均衡技术研究:[硕士学位论文].湖南.湖南师范大学,2007.
[4]Kavas Avi,Feitelson Dror G.Comparing Windows NT,Linux,and QNX as the basis for cluster systems.Concurrency Computation Practice and Experience,2001,13(15):1303-1332.
[5]Hasegawa Atsushi,Matsuoka Hiroshi,Nakanishi Kouichi.Clustefing software for Linux-based HPC.NEC Research and Development,2003,44(1):60-63.
[6]Kavas Avi,Er-El David,Feitelson Dror G.Using multicast to pre-load jobs on the ParPar cluster.Parallel Computing,2001,27(3):315-327.
[7]Brune Matthias,Gehring Joem,Keller Axel,et al.Managing clusters of geographically distributed high-performance computers.Concurrency Practice and Experience,1999,11(15):887-911.
[8]Hai Jin,Rajkumar Buyya,and Mark Baker.Cluster Computing Tools,Applications,and Australian Initiatives for Low Cost Supercomputing.Monitor,Information,Telecomm-unications & ElectronicsEngineering Society(ITEES),2001,25(4):20-26.
[9]基于曙光集群的动画渲染集群方案[EB/OL]http://article.pchome.net/content-455941-3.html.
[10]李双庆.Web服务器集群技术研究:[博士学位论文].湖南.湖南师范大学,2003.
[11]赵振宇,徐用懋.模糊理论和神经网络的基础与应用[M].北京:清华大学出版社,1997.
[12]吕昌国.基于BP算法的网格资源调度研究:[硕士学位论文].湖南.湖南师范大学,2007.
[13]蔡美龄.基于QoS的Web服务选择及组合服务运行时容错研究:[硕士学位论文].湖南.湖南师范大学,2007.
[14]余科军.分布式实时系统任务调度算法的设计和实现:[硕士学位论文].湖南.湖南师范大学,2006.
[15]钟志伟,王广芳,赵先武.分布式计算机系统(DCS)负载平衡算法20年[J].计算机工程,1995,21(6):68-72.
[16]Majeed M.Hayatl,Sagar Dhakall,Chaouki T.Abdallahl,J.Douglas Birdwell2,and John Chiasson2.Dynamic Time Delay Models for Load Balancing Part Ⅱ:A Stochastic Analysis of theEffect of Delay Uncertainty.pdf.
[17]S.Dhakal,M.M.Hayat,M.Elyas,J.Ghanem,and C.T.Abdallah Load Balancing in Distributed Computing Over Wireless LAN:Effects of Network Delay.pdf.
[18]J.Douglas Birdwell1,John Chiasson1,Zhong Tang1,Chaouki Abdallah2,Majeed M.Hayat2,and Tsewei Wang.Dynamic Time Delay Models for LoadBalancing Part Ⅰ:Deterministic Models.pdf.
[19]陈刚,金海.集群和网格环境下作业调度系统的研究:[硕士学位论文].湖南.湖南师范大学,2004.
[20]曲绍云,许曰滨.分布式异构系统中任务调度问题的研究:[硕士学位论文]湖南.湖南师范大学,2005.
[21]卢毅军.韩宗芬.刘科.一种基于集群服务器的动态调度模型研究[J].华中科技大学学报,2001,29(12):82-83.
[22]蒋江,张民选.异构集群系统中基于进程迁移机制的负载均衡算法的研究:[博士学位论文].湖南.湖南师范大学,2002.
[23]王晓川,叶超群.金士饶.一种基于分布式调度机制的集群体系结构[J],计算机工程,2007,28(2):232-234.
[24]陈颉,周智,黄刘生.负载平衡在三维渲染中的应用[J],计算机工程与应用,2005(34):65-68.
[25]嵇鹏,罗军舟.侧重任务调度的网格计算底层支撑系统模型[J].计算机工程,2003,29(21):36-37.
[26]陈刚.集群和网格环境下作业调度系统的研究:[硕士学位论文].湖南.湖南师范大学,2004.
[27]Collins David E,George Alan D.Parallel and sequential job scheduling in heterogeneous clusters A simulation study using software in the loop.Simulation,2001,77(5-6):169-184.
[28]Liu J B,Motov I V.Distributed job scheduling in clustered networks.International Journal of Computers and Their Applications,2003,10(1):43-49.
[29]张永坤,金海,唐丹.一种基于进程剩余运行时间总和的集群动态负载平衡算法[J],计算机工程与科学,2005,27(5):64.
[30]杨振兴.机群计算在可视化平台中负载均衡策略的研究:[硕士学位论文].湖南.湖南师范大学,2005.
[31]万玮.杨志义.分布式计算集群容错系统的实现与设计[J].计算机工程与设计,2002,26(10):2811-2916.
[31]朱育清,许佑辉,王加红.一种新的用于分布式系统的恢复技术[J],小型微型计算机系统,1994,15(20):45-50.
[32]刘云生.大规模分布式仿真系统容错关键技术研究:[博士学位论文].湖南.湖南师范大学,2006.
[33]秦啸,韩宗芳,庞丽萍.基于异构分布式系统的实时容错调度算法研究[J],计算机学报,2002,25(1):49-56.
[34]左德承.分布式计算机系统故障诊断技术的研究:[博士学位论文].湖南.湖南师范大学,2000.
