一种异构片上网络路由算法的研究

发布时间：2018-06-04 09:08

本文选题：异构片上网络 + 无缓存路由　；参考：《计算机科学》2017年03期

【摘要】：随着集成电路工艺的迅速发展,传统的片上网络由于缓存引起芯片面积开销和能耗增加,从而使得无缓存路由技术得到了广泛关注。通过消除缓存,整体的流水线进程大大得到简化,性能得到提高。但当网络负载量较大时,数据包被多次偏转或误传,导致网络的延迟增加,系统健壮性较差。针对片上网络运行应用的多样性,异构网络作为一种相对灵活的网络结构,能有效地降低网络的传输时延,提高系统性能。文中设计了无缓存NoC和带缓存NoC两种路由方式相结合的异构片上网络,并匹配静态路由算法和动态的自适应路由算法(AFC)进行数据包的传输。同时,还提出了一种针对AFC的优化算法(AFC-LP),其通过对无缓存路由计算的二次仲裁,进一步降低了通信的平均时延,提高了网络性能。实验表明,AFC-LP算法相比于传统带缓存的维序X-Y路由算法,片上网络的平均延迟降低了28.4%,CPU每一时钟周期内所执行的指令数IPC(Instruction Per Cycle)提升了10.4%。
[Abstract]:With the rapid development of integrated circuit technology, the traditional on-chip network caused the chip area overhead and the increase of energy consumption, thus making the non caching routing technology widely concerned. By eliminating the cache, the overall pipelining process is greatly simplified and the performance is higher. Deviation or miscommunication leads to the increase of network latency and the poor robustness of the system. As a relatively flexible network structure, the heterogeneous network can effectively reduce the network transmission delay and improve the system performance. In this paper, a non caching NoC and a cached NoC are designed to combine the two routing methods. On chip network, and matching static routing algorithm and dynamic adaptive routing algorithm (AFC) for data packet transmission. At the same time, an optimization algorithm (AFC-LP) for AFC is proposed, which can further reduce the average delay of communication and improve the network performance by two times arbitration without caching routing. The experiment shows that the AFC-LP algorithm is used in the experiment. Compared to the traditional cached X-Y routing algorithm, the average latency of the on-chip network is reduced by 28.4%, and the number of instructions executed by CPU in each clock cycle IPC (Instruction Per Cycle) improves the 10.4%.
【作者单位】：北京工业大学计算机学院;
【基金】：国家自然科学基金(61202076,61202062)资助
【分类号】：TN47

【相似文献】