Massively Parallel Recon gurable System Architecture - POPA 1

In this paper we present a massively parallel reconngurable system architecture, POPA (POSTECH parallel computer). It uses a reconngurable interconnection network which can be dynamically controlled by software, providing the users with great exibility to choose the most eecient interconnection topology for their application. It consists of massive processing elements with MIMD architecture to exploit various types of massive parallelism, and they are organized in distributed memory mode to maintain high performance/cost ratio. In order to support fast I/O, it provides a pool of input and output devices which can be directly attached to the processing elements through the interconnection network. Also, the processing elements can be partitioned into multidomains of various size so that the system can be operated in a multiuser environment. Application softwares in a wide variety of computation intensive areas have been developed under user friendly programming environment with software tools for network connguration, job assignment, I/O management.