论文部分内容阅读
可扩展并行处理机系统有两个需要解决的问题：（１）多级互连网络的延迟；（２）吞吐能力。本文根据ＳＰＰ体系结构的特点和实际应用的需要提出了解决这两个瓶颈问题的有效方法—在前端服务器与共享存储器之间设计专用的并行Ｉ／Ｏ处理机作为系统Ｉ／Ｏ设备与ＳＭ／ＳＳＭ的直接数据通路。这样不但避免了多级网络的延迟问题，而且保证了主机访问主存和ＤＭＡ控制器访问高速缓存之间的并行，从而提高系统吞吐能力。
There are two issues that need to be addressed in scalable parallel processor systems: (1) latency in multi-level interconnected networks; and (2) throughput. In this paper, according to the characteristics of SPP architecture and the need of practical application, we put forward an effective method to solve these two bottlenecks - designing a special parallel I / O processor between the front-end server and shared memory as the system I / O device and SM / SSM direct data path. This not only avoids the issue of multi-level network latency, but also to ensure that the host access and DMA controller access cache parallelism, thereby enhancing system throughput.