数据块并行处理的实现策略及评价方法的研究

Study on implementing and evaluating methods of parallel data block processing

提出了一种针对高速、大容量测试数据的多DSP实时并行处理方法。其中DSP采用了环形的互联方式对数据按块处理。在分析了处理系统中数据I/O、DSP的运算与通信等过程相关性的基础上,提出了如何在给定数据吞吐率下,确定系统工作参数的方法,并给出了具体的算法评价指标,最后以F IR滤波算法为例,给出了详细的实施方法、评价指标和实际验证结果。

A real-time block parallel processing implementation, using a multi-DSP system in a ring network, is proposed for high-speed and massive testing data. After analyzing the dependencies among data input/output, DSP operation and communication procedures in this scheduling model, the approach to establish system parameters is put forward when given the maximum data throughput. The detailed evaluating indexes for parallel algorithm are also presented. Finally, investigation for parallel FIR algorithm in a ring network, including its implementation, evaluating indexes and testing result , are illustrated in detail.