Parallel Computing System Design And Implementation Of Dsp-based Network

With the development of the computer technology and network technology, real-time signal processing, large-scale scientific and engineering calculations are increasingly demanding. Parallel computers provide a good solution for these issues, which is becoming highly valued. Parallel computing technology has become the important symbol of the national economy and technology levels. It is of great significance to its economic and social development, and national security. The developed countries,such as the United States and Japan, attach great importance to the development of parallel computer systems.Traditionally, the serial computing refers to a single computer to perform the corresponding operation, CPU using a series of instructions one by one to solve the problem. Parallel computing, which is relative to the serial calculations, refers to the time parallel in which pipeline technology is used, and the space parallel in which the array of processors with multiple concurrent implementation of the calculation is used. But the array processor in the calculation of feature enhancements and extensions encountered a lot of bottlenecks.This paper presents an Ethernet-based multi-DSP parallel computing systems to solve today's increasingly real-time computing problems and array processors function in the expansion of the bottleneck. The method uses ADI's Blackfin chip, Ethernet as the transport medium, UDP network protocol for real time data communications. Server-side with the powerful performance of ADI BF548 chip .The host carries out the transplant uclinux overall task scheduling and the client using ADI's VDK provides real-time operating system kernel data calculations. Using matrix multiplication and solution of eigenvalue to test .Experimental results on multi-processor parallel system and a single processor system are compared and analyzed. They show that the multi-processor system can provide much higher computing speed compared to the single-processor system in parallel efficiency and lower network latency.