-Coyrehourcq. OpenMOLE, a workflow engine specifically tailored for the distributed exploration of simulation models. Future Generation Computer Systems, Elsevier, 2013Elsevier, , 29 (8), pp.1981Elsevier, -1990Elsevier, . 10.1016Elsevier, /j.future.2013 OpenMOLE, a workflow engine specifically tailored for the distributed exploration of simulation models AbstractComplex-systems describe multiple levels of collective structure and organization. In such systems, the emergence of global behaviour from local interactions is generally studied through large scale experiments on numerical models. This analysis generates important computation loads which require the use of multi-core servers, clusters or grid computing. Dealing with such large scale executions is especially challenging for modellers who don't possess the theoretical and methodological skills required to take advantage of high performance computing environments. That's why we have designed a cloud approach for model experimentation. This approach has been implemented in OpenMOLE (Open MOdel Experiment) as a Domain Specific Language (DSL) that leverages the naturally parallel aspect of model experiments. The OpenMOLE DSL has been designed to explore user-supplied models. It delegates transparently their numerous executions to remote execution environment. From a user perspective, those environments are viewed as services providing computing power, therefore no technical detail is ever exposed. This paper presents the OpenMOLE DSL through the example of a toy model exploration and through the automated calibration of a real-world complex system model in the field of geography.
Monte Carlo simulations are increasingly used in medical physics. In scintigraphic imaging these simulations are used to model imaging systems and to develop and assess tomographic reconstruction algorithms and correction methods for improved image quantization. In radiotherapy-brachytherapy the goal is to evaluate accurately the dosimetry in complex phantoms and at interfaces of tissue, where analytic calculations have shown some limits. The main drawback of Monte Carlo simulations is their high computing time. The aim of our research is to reduce the computing time by parallelizing a simulation on geographically distributed processors. The method is based on the parallelization of the Random Number Generator (RNG) used in Monte Carlo simulations. The long serial of numbers used by the sequential simulation is split. Once the partitioning is done, a software application allows the user to generate automatically the files describing each simulation part. Finally, another software executes them on the DataGrid testbed using an API. All these steps have been made transparent for the user by providing a web page asking the user for all the parameters necessary to launch the simulation and retrieve results. Different tests have been done in order to show first, the reliability of the physical results obtained by concatenation of parallelized output data and secondly the time gained for jobs execution.
Models of emergent phenomena are designed to provide an explanation to global-scale phenomena from local-scale processes. Model validation is commonly done by verifying that the model is able to reproduce the patterns to be explained. We argue that robust validation must not only be based on corroboration, but also on attempting to falsify the model, i.e. making sure that the model behaves soundly for any reasonable input and parameter values. We propose an open-ended evolutionary method based on Novelty Search to look for the diverse patterns a model can produce. The Pattern Space Exploration method was tested on a model of collective motion and compared to three common a priori sampling experiment designs. The method successfully discovered all known qualitatively different kinds of collective motion, and performed much better than the a priori sampling methods. The method was then applied to a case study of city system dynamics to explore the model’s predicted values of city hierarchisation and population growth. This case study showed that the method can provide insights on potential predictive scenarios as well as falsifiers of the model when the simulated dynamics are highly unrealistic.
Multi-agent geographical models integrate very large numbers of spatial interactions. In order to validate those models large amount of computing is necessary for their simulation and calibration. Here a new data processing chain including an automated calibration procedure is experimented on a computational grid using evolutionary algorithms. This is applied for the first time to a geographical model designed to simulate the evolution of an early urban settlement system. The method enables us to reduce the computing time and provides robust results. Using this method, we identify several parameter settings that minimise three objective functions that quantify how closely the model results match a reference pattern. As the values of each parameter in different settings are very close, this estimation considerably reduces the initial possible domain of variation of the parameters. The model is thus a useful tool for further multiple applications on empirical historical situations.
This paper presents an incremental method of parsimonious modelling using intensive and quantitative evaluation. It is applied to a research question in urban geography, namely how well a simple and generic model of a system of cities can reproduce the evolution of Soviet urbanisation. We compared the ability of two models with different levels of complexity to satisfy goals at two levels. The macro-goal is to simulate the evolution of the system's hierarchical structure. The micro-goal is to simulate its micro-dynamics in a realistic way. The evaluation of the models is based on empirical data through a calibration that includes sensitivity analysis using genetic algorithms and distributed computing. We show that a simple model of spatial interactions cannot fully reproduce the observed evolution of Soviet urbanisation from 1959 to 1989. A better fit was achieved when the model's structure was complexified with two mechanisms. Our evaluation goals were assessed through intensive sensitivity analysis. The complexified model allowed us to simulate the evolution of the Soviet urban hierarchy.
No abstract
Monte Carlo simulations are considered as naturally parallel, because many replications of the same experiment can be distributed on multiple execution units to reduce the global simulation time. However, one needs to take care of the underlying random number streams and ensure that the generated streams do not show intra or inter-correlations. Such errors occur in naïve parallelizing approaches, they can lead to erroneous results or to a significant loss in precision. Based on a generic and documented XML format for random number generator statuses and on automatic tools to distribute stochastic simulations, the DistMe software package eases the distribution of stochastic simulations, while keeping the quality of the parallel random number streams as a critical issue. It is written in Java and has been designed to be run on any operating system and hardware with a Java virtual machine available. It has been designed using model engineering to obtain a high quality, modular and very extensible software. This toolkit, freely available on Sourceforge, is designed to speed up Monte Carlo simulations using any parallel machine based on the bag of work paradigm. It provides the user with a set of classes representing a description at a meta level of his simulation environments. Once the developer has described his simulation using DistMe classes, simulation jobs ready for runtime are instantiated. This software is released under GPL licence and the latest development sources are available online (Sourceforge CVS). This paper presents the architecture of DistMe and simulation distribution examples for Geant4 and GATE simulations. The impact of correlations is shown on the GATE application.Index Terms-Distributed stochastic simulation, GATE, Geant4, Monte Carlo, multiple replication in parallel.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.