Concurrency and parallelism in MC and MD simulations in physics

• Published in: Computer physics communications 1985-01, Vol.37 (1), p.251-260
• Main Authors: Pawley, G.S., Bowler, K.C., Kenway, R.D., Wallace, D.J.
• Format: Article
• Language: English
• ISSN: 0010-4655; 1879-2944
• DOI: 10.1016/0010-4655(85)90160-2

Certain problems lend themselves admirably to SIMD architecture computers. A number of such problems in physics will be discussed and their implementation on the ICL DAP analyzed. Monte Carlo (MC) problems such as the 3-D Ising model and quantum chromodynamics (QCD) calculations need care and attention with respect to maintaining detailed balance on a SIMD architecture, whereas molecular dynamics (MD) problems do not pose such problems. The efficient use of a DAP requires the development of parallel algorithms designed to make use of the particular architecture and size of the machine, though further software developments must be made which remove the user one stage farther from the particular details of the computer. Experience on the DAP, especially those with only 2 Mbytes of store such as the Edinburgh DAPs, has led to thought about the detailed precisions necessary for the various calculations being done—many computers operate at an accuracy which is in excess of requirement whereas the bit-serial DAP can have its program finely tuned in accuracy. The present DAPs are far from the modern state of the art in fabrication, and future engines of similar design could be made with present technology with a rating well over a Gflop, providing very cost-effective computing.