A Parallel Processor System for Nuclear Shell-Model Calculations

[摘要] This thesis describes the design and implementation of a dedicated parallel processor system for nuclear shell-model calculations. The purpose of these calculations is to determine nuclear energy eigenvalues by the tridiagonalisation of the nuclear Hamiltonian matrix using the Lanczos method. The Theoretical Nuclear Structure group at Glasgow University's Physics Department would normally perform this type of calculation on a high-performance main-frame computer. However these machines have limitations which restrict the number and scope of the calculations that can be performed. The Shell Model Processor system consists of a Multiple Microprocessor Unit (MMPU) driven by a highly pipelined dedicated front-end processor. The MMPU has a modular, moderately coupled, MIMD architecture based on autonomous processing modules. The elements within the system communicate via three shared buses. The front-end is responsible for determining the position of non-zero elements within the Hamiltonian matrix. Once the position of an element has been found it is passed to one of the free processing modules within the MMPU. The processing module then determines the value of the matrix element and performs the appropriate arithmetic to accumulate the resultant Lanczos vector. Two such processing modules have been developed. The most recently developed module is based on two MC68000 16/32 bit microprocessors. In addition there are two supervisory processor modules, one of which controls the front-end and also assists it in its function. The other module has privileged system capabilities and is responsible for supervising the system as a whole. The system has been successfully tested and performance figures are presented. The future expansion of the system to allow it to perform larger calculations is also discussed.

[发布机构] University:University of Glasgow

[关键词] Nuclear physics and radiation, Theoretical physics