高级检索
    徐 懿, 李 丽, 杜高明, 张宇昂, 张 冰, 高明伦. 一款基于多处理器片上系统的动态自适应仲裁器[J]. 计算机研究与发展, 2008, 45(6).
    引用本文: 徐 懿, 李 丽, 杜高明, 张宇昂, 张 冰, 高明伦. 一款基于多处理器片上系统的动态自适应仲裁器[J]. 计算机研究与发展, 2008, 45(6).
    Xu Yi, Li Li, Du Gaoming, Zhang Yuang, Zhang Bing, Gao Minglun. An Adaptive Dynamic Arbiter for Multi-Processor SoC[J]. Journal of Computer Research and Development, 2008, 45(6).
    Citation: Xu Yi, Li Li, Du Gaoming, Zhang Yuang, Zhang Bing, Gao Minglun. An Adaptive Dynamic Arbiter for Multi-Processor SoC[J]. Journal of Computer Research and Development, 2008, 45(6).

    一款基于多处理器片上系统的动态自适应仲裁器

    An Adaptive Dynamic Arbiter for Multi-Processor SoC

    • 摘要: 随着深亚微米工艺技术的发展,同一芯片上集成多个处理器得以实现.通信架构是多处理器片上系统的瓶颈,而高效的仲裁器可以解决多个处理器同时访问共享资源引起的冲突和竞争,从而防止系统性能的下降.提出一款算法简单的动态自适应仲裁器.它可以自动调节各个处理器占据的总线带宽,避免饥饿现象.基于多处理器仿真平台的实验结果显示它比传统的仲裁器减少了68%的任务完成时间,缩短了78%的总线等待时间,并且能更好地控制各处理器的总线带宽.

       

      Abstract: As the development of deep submicron technology, the integration of more than one processor on a chip is becoming possible. Multi-processor system-on-a-chip (MPSoC) provides a viable solution to highly parallel computations and communications required by high-end applications. The communication architecture is becoming the bottleneck for MPSoC and a determined element of performance, and efficient arbiter is able to solve the contentions due to simultaneous request accesses in shared bus systems to prevent system performance degradation. A simple adaptive dynamic arbiter that can adjust the bandwidth proportion assigned to every processor automatically to avoid starvation problem in MPSoC environment is presented. The proposed arbiter uses random number generator to select winning processor to solve the problem of fairness in competition. To evaluate the adaptive dynamic arbiter, a gate-level four-core simulation environment is set up and two popular bus arbiters—time division multiplexed arbiter and Lottery bus are implemented. The experimental results show that the proposed one could reduce 68% task execution time, decrease the bus request latency of a processor by 78% and provide better control of the communication bandwidth allocated to individual processor than convention arbiters.

       

    /

    返回文章
    返回