一种基于微片分组的片上网络微节点交叉开关调度机制  

Crossbar Scheduling Mechanism Based on Flit Batching for Micro-nodes of Network-on-Chip

作　　者：刘亮亮[1] 韩国栋[1] 宋克[1] 张帆[1] 

机构地区：[1]国家数字交换系统工程技术研究中心,郑州450002

出　　处：《小型微型计算机系统》2013年第5期1108-1112,共5页Journal of Chinese Computer Systems

基　　金：国家"八六三"高技术研究发展计划项目(2009AA012201)资助

摘　　要：在基于微片(flit)分组的动态缓存分配基础上,提出一种基于微片分组的片上网络交叉开关调度机制.该机制与静态独立分割缓存的思想不同,首先对输入端缓存进行统一管理,对微片根据其流向进行分组,并为所分各"组"动态分配缓存,然后引入一种基于"组"规模的概率仲裁算法,通过"组"分配和开关分配实现调度过程.为进一步降低开销,还在该机制基础上提出一种各"组"共享仲裁的策略.理论分析与实验结果均表明:所提出的机制相对于传统和动态虚通道机制,可节约25%以上的硬件开销并可获得更优的网络延迟与吞吐性能;共享仲裁策略可在所提机制基础上进一步降低硬件开销,但其代价是网络性能有所下降.This paper introduces a dynamic buffer allocation scheme based on flit batching. And a crossbar, scheduling mechanism for Network-on-Chip （ NoC） is proposed based on this buffer allocation scheme. The mechanism proposed is different from the idea of partitioning the buffer space statistically and independently. It manages the input buffer uniformly, batches flits according to their flow directions and dynamically allocates buffer space for batches, then a probabilistic arbitration algorithm based on batch sizes is proposed and the scheduling process is completed through batch allocation and switch allocation. To further reduce the consumption, an arbiter sharing method is proposed as well based on this mechanism. The theoretical analysis and experimental results both indicate that the mechanism this paper proposed could save at least 25 % hardware consumption and attain better network latency and throughput performance compared with general and dynamic virtual channel mechanisms; the arbiter sharing scheme could further reduce the hardware consumption at the expense of some decline in network performance.

关 键 词：片上网络 微片分组 动态缓存分配 概率仲裁 共享仲裁 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]