Study Of Ocean Data Visualization Parallel Computing Framework Based On GPU-Hadoop

"Ocean Environmental Information Visualization" is one of the important topicsof ocean charity project "Marine Environmental Information System cloud computingand cloud services framework application research"(No.201105033)(Oceanenvironmental information visualization refers to using computer graphics theory andmethod predict ocean variations). In recent years, along with advances in computingtechnology and the development of science and technology,data of ocean environmentinformation increased quickly. For large-scale visualization tasks, alone-computer isfar below standard. In reference to the international mainstream hadoop cloudcomputing platform and the features of China’s ocean environment, hadoop will beapplied to large-scale ocean environmental information visualization computing. Thisis an effective solution to solve the above problems.With the advent of the high-performance computing rea, parallel computingprovides a new technique for the study of mass data processing. Especially in thedevelopment of GPU and CUDA matures, GPU parallel computing is becoming moreand more popular. GPU belong to parallel multicore processors.The calculation andprocessing capabilities of the GPU is10times or even ten times than CPU.GPU hasmore processors and higher bandwidth, lower power consumption.Thereby providinga new solution to improve the efficiency of mass data processing.In order to improve the ocean environment information visualization andreal-time calculation of the efficiency,combing GPU with hadoop has become a verygood solution in visualization computing area.According to ocean charity project "Ocean Environmental Information cloudcomputing and cloud service system framework application research " on the oceanenvironment information visualization accuracy and computational efficiency requirements, this paper studied the solution of embedding GPU into hadoop cloudplatform efficiently and designed a GPU-Hadoop parallel framework, applied in theframework of the Ocean Environment information Visualization successfully. In orderto pursue higher speedup,this paper proposed two methods to optimize taskscheduling of GPU-hadoop parallel computing framework.In this paper, the workdone are:1.Based on the hadoop’s features and advantages of the GPU, studied thesolution of embedding GPU into hadoop.According to our experimental environmentand feasibility, proposed a complete solution, which is available in the oceanenvironment information visualization.What’s more,it is effective for other projects.2. In the GPU-hadoop parallel computing framework, proposed using dynamicdata processor to split or assemble data which is passed from CPU to GPU. In order toimprove the computational efficiency of the GPU,used stream operations to reach theasynchronous execution of GPU and CPU.This method enhanced the computingperformance of the framework.3. GPU-hadoop parallel computing framework is applied to calculate the oceanenvironment information visualization which achieved good results.At the same time,developed ocean environment information visualization platform for remote jobsubmission and display function.4. This paper proposed the daemon and optimization method based GPUprocessing capabilities, which improved the efficiency of computing tasks schedulingunder the GPU-hadoop framework.As a result,the overall computing capacity hasbeen greatly improved. GPU-based flow visualization reached15times speedup ratethan CPU mode.