流式处理器针对数据并行的应用,提供多个简单的处理单元及它们之间的高带宽通道,通过高计算密度掩盖传输延迟,支持高速的数据传输和处理,与通用处理器相比,性能达到了数倍乃至数百倍的提升。在流式处理器的研究中,数据流的调度是个比较重要的问题。本文提出了一种基于拓扑特征的数据流调度策略,该策略包含平分策略和选择策略两种,其中平分策略用于体系结构中固有负载较少的情况,选择策略在固有负载较多时利用拓扑权重来辅助平分策略完成数据流任务的调度。建立了该策略的性能模型,在考虑节点间传输时间的情况下进行讨论,对数据流调度策略的选择提出建议。

Stream processors are suitable for data parallel applications; the processor is supported by many simple processing units and high bandwidth routing among the units. By means of high density computing, the transfer latency is concealed, therefore the high speed data transfer and processing are able to be supported, making the performance of stream processors raise several or even hundreds of times comparing with general processors. Data stream scheduling is a relatively important problem in the research of stream processor. Methods that utilize the features of architecture topology to compute the weights on every edge are proposed. These methods are called as Stream Scheduling based on Topology Features (SSTF). SSTF mainly includes divide strategy and select strategy. Divide strategy is suitable for the architecture with light load. However, if there is heavy load, select strategy is used to assist divide strategy with the completion of data stream scheduling. A performance model of SSTF is built, which gives a consideration to the transfer time between nodes in one topology; and the way for selecting scheduling methods is suggested.