Study Of Marine Monitoring Data Flow Processing Stategy For Optimization

As the ocean decision support system has been completed the overall framework,which include the data sources, such as data from the marine stations and satellitesreal-time monitoring; the transmission networks, such as far and near ocean datatransmission lines and3G wireless networks; the data receiving stations,such as theocean central stations. The historical marine monitoring data, whose capacity wasalmost to2TB had been stored into the databases.As the historical data is an importantrecord of the marine’s evolution, these data will support the system to generate thedecisions. When the marine disaster appeares, it is essential to match these historicaldata with the real-time monitoring data for generating the decisions rapidly, and anotherhot issue is about how to save the cost of the storage and matching while ensure that thesystem will run in high quality.The idea of this paper was from the study of the National Digital Ocean Research inShanghai-"Digital Ocean"908special supported research projects, which would sovlethe marine issues with the computer science. The current system could not analyse thesemassive marine data nor generate the decisions in a short time when disaster occurs.Private cloud storage subsystem will both save manpower and financial while store thereal-time ocean monitoring data into the directed databases. By the GPU’s multicoreparallel computing’s superiority, the GPU collaborate with CPU parallel processingsubsystem will optimize the data flow matching process. The paper studies the digitalocean data flow in order to neaten the characteristics of ocean. The study will supportthe system to be a high-performance.The paper describes the development of Chinesse and overseas scholars’ study ofdigital ocean decision support system firstly, then explores the possibility of framing theprivate cloud storage mechanism with the GPU-CPU parallel computing theory togetherfor the "Digital Ocean" decision support system secondly, the paper demonstrates and explains the relevant theories and technology, the system was applicated in Shanghaidigital ocean project and was consummated architecture during the research. The newarchitecture optimized both the real-time data storage mechanism and the parallelcomputing mechanism, so the system’s parameters such as reliability, throughput andresponse time are better than the previous system’s.The study shows that the current data storage subsystems are generally consideredonly with the rates of the disk storage, but the characteristics of real-time monitoringdata are uninterrupted and contain multi-attributes, after calculating with the alreadyestablished marine data transmission networks, this paper proposes to inspect themultiple parameters of the computers,such as inspect the rate of memory, the rate ofdisks and the rate of CPUs, the private cloud storage subsystem will use the dynamicload balancing mechanism to allocate the storage, the target of the system is going tostore the real-time data into the leisure servers.After studying with the current digital ocean systems and decision support systems,the research found that the systems’ response times were too long than the ideal, thesystem’s efficiency of generating the decisions became system’s bottleneck.Byanalyzing the characteristics of real-time monitoring data, the paper decided to establishthe frameworks of GPU collaborate with CPU parallel computing. As the technologydeveloping, GPU has become a multi-cores processing unit, which has high bandwidth.It will solve the data stream’s problems perfectly. This paper proposed a GPUcollaborate with CPU parallel computing theory: the GPU will pretreatment themonitoring data stream with its bandwidth, after the normalized process, the GPU willusing DTW algorithm to match national standards with the sliding windows.Experimental results show that the new system’s performances meet the needs ofdata storage and generate decisions. Because there are no international standards for theperformances of decision support systems, so the contributions of this study is not onlyto optimize the processing of marine monitoring data flow, but also could share theideas with other scholars, who study the digital ocean decision support system.