Achieving High Performance Parallel Statistical Computing in Tightly Coupled Network of Unix Workstations

Keywords: parallelism, TCP/IP, Socket, multi-thread, IPC

Abstract: We set up the DiPGAC (Distributed Parallel Genetics Anticipation Computing)---a multi-CPU system through a tightly coupled network of Unix workstation and try to perform the parallel and multi-threaded computing over a computational statistics problem. TCP socket served as the inter-CPU data communication and Unix IPC served as an interface for high speed C program to communicate with slow statistical functions calls in Splus. By paralleling the CPU intensive module and balanced distribution of system resource, the system can achieve high performance computing without extra cost of multi-processor computer and parallel operating system and compiler.