Testing parallel random number generators

Srinivasan, Ashok; Mascagni, Michael; Ceperley, David M.

doi:10.1016/s0167-8191(02)00163-1

Cited by 74 publications

(75 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, L'Ecuyer has formulated several key criteria for "good" RNGs: a long period, repeatable outputs, portability to different execution platforms, and the ability to split the output sequence into multiple independent blocks (what means they should implement an efficient jump-ahead strategy) and each block again into substreams with the leap frog approach (L'Ecuyer, 2007; L' Ecuyer and Panneton, 2005 Ecuyer and Simard, 2007) is considered the most comprehensive one and is preferred by many researchers (Salmon et al, 2011;Hill et al, 2013). A first requirement for a good parallel PRNG is that it also has to be a good sequential PRNG (Srinivasan et al, 2003). Obviously, when testing different substreams of one PRNG, it is mandatory to ensure that exactly the same substream can be outputted again to allow debugging and reproducibility of the test results (Hill et al, 2013).…”

Section: Quality Issuesmentioning

confidence: 99%

On parallel random number generation for accelerating simulations of communication systems

et al. 2014

View full text Add to dashboard Cite

Abstract. Powerful compute clusters and multi-core systems have become widely available in research and industry nowadays. This boost in utilizable computational power tempts people to run compute-intensive tasks on those clusters, either for speed or accuracy reasons. Especially Monte Carlo simulations with their inherent parallelism promise very high speedups. Nevertheless, the quality of Monte Carlo simulations strongly depends on the quality of the employed random numbers. In this work we present a comprehensive analysis of state-of-the-art pseudo random number generators like the MT19937 or the WELL generator used for parallel stream generation in different settings. These random number generators can be realized in hardware as well as in software and help to accelerate the analysis (or simulation) of communications systems. We show that it is possible to generate high-quality parallel random number streams with both generators, as long as some configuration constraints are met. We furthermore depict that distributed simulations with those generator types are viable even to very high degrees of parallelism.

show abstract

Section: Quality Issuesmentioning

confidence: 99%

On parallel random number generation for accelerating simulations of communication systems

et al. 2014

View full text Add to dashboard Cite

show abstract

“…However, designing the program that can actually spawn those separate generators turns out to be a dicey problem. Some successful reports have been published (Srinivasan et al, 2003).…”

Section: Where Do Random Numbers Come From?mentioning

confidence: 99%

“…However, designing the program that can actually spawn those separate generators turns out to be a dicey problem. Some successful reports have been published (Srinivasan et al, 2003).The other leading approach, due to L 'Ecuyer et al (2002), is to take the one long stream of numbers from the generator and then divide it into separate substreams. Their implementation uses the MRG32k3a.…”

mentioning

confidence: 99%

Monte Carlo Analysis in Academic Research

Johnson

2013

Oxford Handbooks Online

View full text Add to dashboard Cite

Monte Carlo analysis is a research strategy that incorporates randomness into the design, implementation or evaluation of theoretical models. It began in the 1940s, when the development of computer hardware and mathematical models made it possible to generate streams of random numbers. These random number streams are combined with mathematical models in order to create models and evaluate theories of random processes. This chapter attempts to tame this diverse, unmanageable collection of concepts and methods by dividing simulation projects into three types. The first, commonly called "Monte Carlo simulation," is used to evaluate statistical estimators. When an estimation procedure is proposed, it is standard procedure to test it against a variety of simulated research problems. A second type of project, referred to as "Markov chain Monte Carlo" (or MCMC), helps researchers to draw conclusions about complicated probability models for which conventional research strategies do not yield insights. The third 1 type of project arises in the study of complex systems, which are characterized by a large number of loosely interconnected, autonomous elements. Commonly known as agent-based models, these simulations have found enthusiastic advocates in environmental and social sciences.Keywords: Monte Carlo, Markov chain Monte Carlo (MCMC), pseudo random number generation (PRNG), Bayesian statistics, agent-based modeling Monte Carlo (MC) analysis is a general term that refers to research that employs random numbers, usually in the form of a computer model (or simulation). Although this research began in the natural sciences, computer science, and mathematics, it is now widely applied in social science as well. This essay attempts to explain the fundamental ideas that spurred the creation of these new procedures as well as their eventual adaptation for use in social science research.This chapter is not a "how to" guide for simulation, but rather it is a "what for" or "why you might want to" guide. Some of the difficulties that arise in MC research projects are considered as well. It begins with some background information on the development of computers and algorithms for random numbers. After that, the chapter takes up applications in the evaluation of proposed statistical estimators, the practice of Bayesian statistics via computer simulation, and investigation of complex systems through agent-based models. Some conclusions about the challenges that face the field are presented, along with a conclusion. A significant part of the presentation is about the exciting developments that have occurred since 1990. Rapid improvements in hardware and software have opened up opportunities for scholars to work with models that were previously prohibited by conceptual and technical barriers. At the current time, we are able to conceptualize and implement models that were simply impossible just 10 years ago. The extremely rapid progress has been driven by a fruitful interaction of substantive researchers in the natural and social science...

show abstract

“…Although Monte-Carlo computations are considered easy to parallelize, simulation results can be adversely affected by defects in the parallel pseudorandom number generator used [17]. The independence of the generated random numbers and the subsequent independence of simulations on different processes are the primary requirement for the success of stochastic simulations.…”

Section: Parallel Implementationmentioning

confidence: 99%

“…However, finding a good parallel random number generator has proven to be a very challenge problem, and is still the subject of much research and debate [6]. Based on recent empirical tests for a number of parallel random number generators, it was still suggested to use a number of different generators to run the application in order to increase our confidence on simulation results [17]. In our previous attempts, we tried to use different random seeds in different processors in the MPI environment, and obtained very good speedup and efficiency that is close to 1 [2].…”

Section: Parallel Implementationmentioning

confidence: 99%