片上网络的访存延迟均衡性

Latency equalization of memory access in network-chips

对片上网络访存延迟均衡性展开了研究,提出基于总延迟预测的访存报文仲裁技术。首先,依据访存报文后续路径的拥塞信息预测访存报文未来等待延迟,并计算出总延迟。其次,基于预测的总延迟对竞争同一链路的访存报文进行仲裁。在Mesh片上网络路由器中,对该技术进行了设计和实现。实验结果表明:在不同的网络规模和报文注入率下,与经典Round-Robin仲裁机制相比,本文技术能够极大减少片上访存的最大延迟和延迟标准差,减少平均延迟,证明能够获得更佳的访存延迟均衡性。

A novel arbitration technique for memory access packets is proposed,which is based on the round-trip latency prediction.First,the congestion information in the subsequent path of memory access packets is used to predict the waiting latencies of the memory access packets in the future,and then the round-trip latencies are calculated.Second,the predicted round-trip latencies are used to decide the arbitration for the memory access packets contending for the same link.The proposed technique is designed and implemented in the routers of mesh-based NoCs.Experimental results show that,under different network sizes and packet injection rates,compared with the classic Round-Robin arbitration mechanism,the proposed technique can greatly reduce the maximum latency,the average latency and the latency standard deviation of on-chip memory accesses,and it is proved to achieve better latency equalization of memory access.