一种片上网络均衡响应访存流控技术  

Memory Access Flow Control Technique with Fair Response on Network-on-Chip

作　　者：何明勇[1] 陈小文[1] 许邦建[1] 

机构地区：[1]国防科技大学计算机学院,长沙410073

出　　处：《高性能计算技术》2013年第6期40-44,共5页

基　　金：本文获得国家自然科学基金（61070036）资助.

摘　　要：摘要在以片上网络为互连架构的众核处理器中，集成了大量的处理器核和存储资源，产生大量的访存请求。访存请求的增多对访存流控机制的设计提出了新的挑战。其中之一就是如何公平仲裁和均衡响应访存请求。本文提出了一种面向片上网络，支持均衡响应的访存流控技术，综合考虑访存请求的优先级、路由延迟和存控通路延迟。基于该技术，本文设计并实现了相应的访存流控机制结构，并进行了验证。验证结果表明，采用该技术具有不同优先级的访存请求能够得到公平仲裁和响应。Many processor cores and memory resources are integrated on NoC based many-core processors, leading to a lot of memory access requests. The rapid increase of memory access requests brings new challenges on designing memory access flow control mechanism. One of them is how to arbitrate and respond to incoming memory access requests fairly. This paper proposes a memory access flow control technique with fair response for NoC, considering memory access requests＇ priority, routing delay and waiting delay in memory controllers. Based on the technique, the memory access flow control structure is designed and validated. The results show that the proposed technique can fairly arbitrate and handle memory access requests with different priority.

关 键 词：片上网络(NoC) 流控机制 优先级 报文 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]