一种面向应用的NOC缓冲区分配算法

片上互连网络是片上通信问题的有效解决方案,但其存在严重的资源限制.输入缓冲区占据片上网络总面积的显著部分,同时其容量大小对不同应用映射后获得的性能有重要影响.给出一种面向应用数据负载的NOC缓冲区分配算法,针对不同的应用映射,该算法可以根据数据流量分布特征实现各个路由器输入通道上缓冲区资源的定制分配.实验结果表明,使用该算法后,系统缓冲区资源得到了更有效的利用.与均匀分配缓冲区的NOC系统相比,采用该算法实现的缓冲区分配方案使系统在保持性能变化不大的情况下,能够节省约50%的缓冲区总容量.

An Application-specific Buffer Allocation Algorithm for Network-on-chip

The networks-on-chip(NOC) approach was proposed as a promising solution to on-chip communication problems, but it is by far more resource limited. The input buffets in a typical on-chip router take a significant portion of the silicon area of NOC, and the performance of a NOC is drastically affected by the amount of buffeting resources. In tiffs paper, an application-specific buffer planning algorithm that can be used to customize the router design in NOC is presented. More precisely, given the mapping of the target applicadon and the traffic charracteristics, the algorithm automatically assigns the buffer depth for each input channel, in different routers across the chip. The experimental results show that the system buffering resources can be utilized more effectively. In contrast with the uniform buffer allocation, about 50% saving in buffering resources can be achieved by automatic buffer allocation using our algorithm without any reduction in performance.