Monte Carlo Methods: Algorithms and Implications for High-Performance Computing

Name:    Prof. Michael Mascagni                 
Address: Department of Computer Science and
         Department of Mathematics and
	 Department of Scientific Computing and
         Graduate Program in Molecular Biophysics	
         Florida State University
         Tallahassee, FL  32306-4530  USA
         AND
         Information Technology Laboratory
         Applied and Computational Mathematics Division
         100 Bureau Drive M/S 8910
         National Institute of Standards and Technology (NIST)
         Gaithersburg, MD  20899-8910  USA

Offices: 498 Dirac Science Library/207A Love Building (FSU) Building 225/Room B154 (NIST)
Phone:   +1.850.644.3290 (FSU) +1.301.975.2051 (NIST)
FAX:     +1.850.644.0058
e-mail:  mascagni@fsu.edu (FSU) mascagni@nist.gov (NIST)

Title: Random Number Generation Tools for Distributed Simulation on Modern HPC Architectures

Abstract:
Monte Carlo and other simulation methods are a class of computations that have always been unusually suitable to parallel computation. However, the realization of efficient parallel simulation depends of the quality of the random number generation tools available. This is especially true with parallel random number generation, where issues arise in testing the quality of a group of random number streams when they are used simultaneously. Work on these problems produced the Scalable Parallel Random Number Generators (SPRNG) library. This is a very popular library that was widely adopted in the Monte Carlo community on distributed-memory high-performance computing (HPC) systems. Current HPC systems are incorporating multicore, GPU-based accelerators, and the Intel Phi to achieve ever higher performance within ever more strict power constraints. In this talk we will discuss these architectural developments in HPC, especially at the exascale, and what this requires of the next generation of random number generation tools. We then describe how SPRNG is being upgraded to meet these more stringent requirements. Of particular emphasis with be the nonlinear, multiplicative lagged-Fibonacci generator, that has many desirable properties, has been available in SPRNG for many years, and show great promise as a simple generator family that can provide for many of the modern random number requirements for simulation on HPC systems, even at the Exascale.

This is joint work with Drs. Yue Qiu of FSU and Timothy Anderson of Daniel H. Wagner, Associates.