针对优化片上网络数据包传输延时的问题,提出了一种在缓冲资源限制条件下的缓冲区分配算法.该算法在建立二维网格结构的片上网络通信模型的基础上,依据各节点之间的通信流量,估算出节点中各输入通道的负载大小,再根据其负载情况采用模拟退火算法实现缓冲区资源的分配.实验结果表明,该算法可以更加合理地分配缓冲区资源,有效降低数据的传输延时.在均匀通信流量下,可降低传输延时11.3%～56.4%;在单热点通信流量和双热点通信流量下,可分别降低传输延时13.2%～76.9%和13.7%～79.1%.%On the finite buffer resources in network on chip, a buffer allocation algorithm is proposed to minimize the average delay of packet. Based on the model of communication performance for 2D mesh NoC( network on chip), the proposed algorithm first estimates the load on each input channel in different routers across the chip by the traffic characteristics. Then, it uses simulated annealing algorithm to allocate the resources according to the distribution of the load on all the channels. Simulation results show that the buffer allocation is more reasonable and smaller average packet latency can be achieved. For uniform traffic, this algorithm can get packet latency 11.3 ％ to 56.4％ less than those by uniform allocation. For one hotspot traffic and two hotspot traffic, 13.2％to 76.9％ and 13.7％ to 79.1％ delay can be reduced respectively.
展开▼
