Parallel Monte Carlo calculations with many microcomputers

In a University environment with very many microcomputers, often idle during some part of the day, we have analysed the structure of High Energy Physics Monte Carlo programs in an attempt to transform them into parallel algorithms which could run on such idle equipment. We present a theoretical analysis of the problem of managing a large number of distributed microcomputers under control of a central machine. Emphasis is on solving contingencies, overcoming failures and estimating the optimal efficiency. We also present a test case with a relative small sized program for the calculation of showers, which has succesfully been implemented on a set of DEC-Rainbows under control of a DEC LSI-11/23. The microcomputers can be used at any time by the body of students for whom they were acquired, while the Monte Carlo monitor running on the LSI-11/23 makes use of any available microcomputer as soon as released by the user. Extrapolations to a more powerful system than the 11/23 with its memory and system limitations, and to microcomputers equipped with more memory and floating point coprocessors are also discussed.