Research On Stream Development Tool And Dataflow Programming Optimization

In the big data age, dataflow programming language have some obvious advantages than the traditional object oriented programming languages, however, in the process of dataflow task scheduling optimization, to get the results is not easy for the division of load balancing.Especially in the field of multimedia, the dataflow characteristics in the field of program do not make the best of multimedia to optimize task partitioning, which reduce the dataflow process in the throughput rate of multi-core platform and system performance.Dataflow process as a domain specific language features can also bring new programming challenges to the programmer.According to modular programming and multimedia dataflow field optimization problem, this paper designs the dataflow task partitioning optimization algorithm and structured caching algorithm, and the realization of the modular COStream dataflow language of visual development environment.Iteration count of multimedia data dependencies and data locality is a performance optimization direction in dataflow framework, respectively for the two problems in statistic relating variable eliminate task partitioning algorithm and data flow structured caching algorithm, Statistic relating variable eliminate in iterative replication algorithm is improved on the basis of the division, to eliminate parts of iteration dependency on the state of the dataflow node;Structured caching algorithm divides into two kinds of global and actor cache implementations based on data stream state, respectively, for the whole SDF map and a single compute node for data cache.On account of dataflow language programming challenge for programmers, the design has realized the dataflow visualization development environment, fall into six modules of programming framework, the code editor and highlighting, programming interface design, dataflow chart, according to the performance analysis to solve the dataflow programming, debugging and to inconvenience the result analysis.Based on X86 multicore architecture as the experiment platform, analysis through the typical algorithms in the field of digital media optimization results, we show that the effectiveness of the task scheduling optimization and data cache optimization.