PMC: A shared short-cut to portable parallel power
Monte Carlo particle transport has been shown to be very well suited to exploit the dramatic speedups available from Multiple Parallel Processors (MPPs). Other portions of large-scale codes that simultaneously simulate a variety of physical processors are less easily adapted to MPPS. PMC (Parallel Monte Carlo) is a package of generic interface routines that allow distributed processing where Monte Carlo is performed on an MPP while the other physics is simulated on a conventional, serial supercomputer. PMC handles passing of messages between the serial machine and the MPP, passing of messages between nodes on the MPP, distributing work between nodes, and providing independent, reproducible sequences of random numbers. Reproducibility is ensured through multi-tasking on nodes. As long as the number of tasks remains constant from one problem execution to another, the Monte Carlo simulation will yield identical results regardless of the number of nodes used. A task is a portion of work such as advancing a collection of particles from the beginning of a time step to its end. PMC handles this multi-tasking. PMC has been designed to allow re-use of existing Monte Carlo transport packages with minimal rewrite.
- Research Organization:
- Lawrence Livermore National Lab., CA (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 10120908
- Report Number(s):
- UCRL-JC--112311; CONF-9211128--6; ON: DE93005815
- Country of Publication:
- United States
- Language:
- English
Similar Records
Parallelizing Monte Carlo with PMC
Experiences with different parallel programming paradigms for Monte Carlo particle transport leads to a portable toolkit for parallel Monte Carlo