Statistical considerations in simulation on a network of microcomputers

Biles, William E.; Daniels, Cheryl M.; O'Donnell, Tamilea J.

doi:10.1145/21850.253416

Cited by 16 publications

(11 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Nevertheless, in many simulation applications, the primary benefit offered by PADS is the execution speedup of running many replications on parallel processors. Early work can be found in Biles et al (1985) in which different computer architectures for carrying out a large number of simulation replications in a parallel computing environment were examined. Subsequent work was done by Heidelberger (1988), who proposed a parallel replications environment equipped with more advanced statistical methods for supporting replication analysis.…”

Section: Literature Reviewmentioning

confidence: 99%

Factory flow design and analysis using internet-enabled simulation-based optimization and automatic model generation

Ng¹,

Bernedixen²,

Moris³

et al. 2011

Proceedings of the 2011 Winter Simulation Conference (WSC)

View full text Add to dashboard Cite

Despite simulation offers tremendous promise for designing and analyzing complex production systems, manufacturing industry has been less successful in using it as a decision support tool, especially in the early conceptual phase of factory flow design. If simulation is used today for system design, it is more often used in later phases when important design decisions have already been made and costs are locked. With an aim to advocate the use of simulation in early phases of factory design and analysis, this paper introduces FACTS Analyzer, a toolset developed based on the concept of integrating model abstraction, automatic model generation and simulation-based optimization under an innovative Internet-based platform. Specifically, it addresses a novel model aggregation and generation method, which when combined together with other system components, like optimization engines, can synthetically enable simulation to become much easier to use and speed up the time-consuming model building, experimentation and optimization processes, in order to support optimal decision making. INTRODUCTIONReal-world systems in manufacturing, supply chains and public services are too complex to be modeled by analytical techniques. Therefore, discrete event simulation (DES) are very useful for performing modeling and analysis on these systems. However, DES models are by nature evaluative -instead of suggesting any optimal solutions, a DES model evaluates a given set of design variables and generates the required performance measures. For a decision maker, the process of finding a sufficiently good design setting could be too time-consuming and in many cases impossible if the search space is huge. Simulation-based optimization (SBO) is a relatively new technique applied to seek the "optimal" setting for a complex system based on one or multiple performance measures generated from simulation by using various searching methodologies. SBO is a technology that offers huge potentials to solve real-world problems and have been successfully applied in many different domains (April et al. 2004). Nevertheless, until now, virtually all of today's commercial SBO packages still possess several major limitations: (1) they work in a deterministic mode, without taking into account the stochastic outputs from DES; (2) they do not explicitly address multi-objective problems, and (3) similar to most of the DES packages, the majority of SBO tools available is traditional software that need to be installed and run locally on the users' computers. With the vision that Internet and Web technologies could enable the explosive growth in research and commercial opportunities (Fu et al. 2000;Boesel et al. 2001;Miller et al. 2001), many efforts have been paid on Web-based simulation (WBS) since the 1990s. However, as summarized recently by Byrne 2176 978-1-4577-2109-0/11/$26.00 ©2011 IEEE

show abstract

Section: Literature Reviewmentioning

confidence: 99%

Factory flow design and analysis using internet-enabled simulation-based optimization and automatic model generation

Ng¹,

Bernedixen²,

Moris³

et al. 2011

Proceedings of the 2011 Winter Simulation Conference (WSC)

View full text Add to dashboard Cite

show abstract

“…In a DES, parallelism can be exploited in various ways [23,26], including 1) applying a parallelizing compiler to the sequential simulation implementation, 2) separating independent simulation runs on multiple processors [4], 3) running subroutine calls in the simulation on different processors, 4) maintaining a global event list and having multiple processes access and process the events in the list simultaneously [17], and 5) decomposing the simulation into multiple components in time or space domain and running the components on multiple processors at the same time [6,21]. In this paper, we parallelize the simulation program using the last approach, dividing events in the system in the space domain, through which we may have the greatest potential of exploiting parallelism as the system being simulated increases in size and complexity.…”

Section: Parallel Discrete Event Simulationmentioning

confidence: 99%

Parallelizing a discrete event simulation application using the Habanero-Java multicore library

Xiao

Zhao

Sarkar

2015

Proceedings of the Sixth International Workshop on Programming Models and Applications for Multicores and Manycores

View full text Add to dashboard Cite

Discrete event simulation (DES) has been widely adopted for simulating communication systems such as computer networks. As the network size and complexity of communication patterns increases, the complexity of simulation tools and the execution time of DES also increases. Parallelizing DES programs using multiple processing units reduces the overall execution time; however, unlike regular programs, data dependencies in DES are usually determined at runtime, which makes exploiting potential parallelism in the program very challenging. In this paper, we build a parallel version of a DES program written in Java using the Habanero-Java library (HJlib), which is a lightweight and programmer-friendly parallel Java 8 library. While the DES problem benefits greatly from HJlib's support for lightweight tasks and efficient parallelism based on work stealing, it also pushed the boundaries of the standard primitives available in HJlib. In particular, our study motivated the addition of fine-grained locking to the Habanero execution model, in a manner that still preserves Habanero's deadlock freedom guarantees. This extension in turn led to additional optimizations of the DES implementation relative to the original Galois implementation. Our initial results are encouraging, and point to further opportunities for exploiting parallelism in challenging applications like DES on manycore hardware platforms in the future, especially as the system being simulated increases in size and complexity.

show abstract

“…If the simulation is largely stochastic and one is performing long simulation runs to reduce variance, or if one is attempting to simulate a specific simulation problem across a large number of different parameter settings, an alternative (and probably preferred) approach is to execute independent, sequentialsimulation programs on different processors [5,24]. This replicated trials approach can be very effective, though it is only useful if each processor contains sufficient memory to hold the entire simulation.…”

Section: Introductionmentioning

confidence: 99%