Parallel Computing in Ido Lab.

Parallel Computing

Parallel computing is applied to particle simulations and Monte Carlo simulations for nuclear fusion and plasma engineering studies. In these simulations, parallelization shows the good performance to accelerate the calculations, nevertheless vectorization is not useful.

The purpose of parallelization is to accelerate the calculation speed in the simulations. The Amdahl's law is applicable in not only vector computing but also parallel computing, when the key factor is vectorization in the former and parallelization in the latter.

• Acceleration vs. Parallelization

• Acceleration vs. Number of Processors

Parallelization of the simulation codes were initiated in 1988 by using the trasputer network where the main cpus were T800 (20MHz). The languages for parallel computing were "Parallel FORTRAN" and "Parallel C" which have the message passing routines. By developing the interface routines, the same parallelized programs were applied to other MIMD type parallel computers. Then, bench mark tests were carried out by using the same calculation program.

• Bench Mark Tests

The parallel systems used in the simulations are

• Transputer (T-800)
• Intel iPSC860
• PVM using SPARC stations or HP9000-715

Published papers

• Personal supercomputing by using transputer and Intel 80860 in plasma engineering

• Parallel programming in MIMD type parallel systems using transputer and i860 in physical simulations