IBM Computation Time
Autor
Patrick Chabrier
Letzte Aktualisierung
vor 10 Jahren
Lizenz
Creative Commons CC BY 4.0
Abstrakt
A very preliminary study of the vle.extension.ibm package of vle in terms of computation time.
\documentclass{beamer}
%
% Choose how your presentation looks.
%
% For more themes, color themes and font themes, see:
% http://deic.uab.es/~iblanes/beamer_gallery/index_by_theme.html
%
\mode<presentation>
{
\usetheme{default} % or try Darmstadt, Madrid, Warsaw, ...
\usecolortheme{default} % or try albatross, beaver, crane, ...
\usefonttheme{default} % or try serif, structurebold, ...
\setbeamertemplate{navigation symbols}{}
\setbeamertemplate{caption}[numbered]
}
\usepackage[english]{babel}
\usepackage[utf8x]{inputenc}
\title[IBM Computation Time]{Computation Time consideration}
\subtitle{when using vle.extension.ibm}
\author{Patrick Chabrier}
\institute{MIA}
\date{\today}
\begin{document}
\begin{frame}
\titlepage
\end{frame}
% Uncomment these lines for an automatically generated outline.
%\begin{frame}{Outline}
% \tableofcontents
%\end{frame}
\section{Introduction}
\begin{frame}{Introduction}
This is a very preliminary study, in order to introduce the subject. And we have taken the a user point of view.
\begin{itemize}
\item An empirical approach.
\item A single model Lotka Voltera
\item A single method, expected to be the same on both side (RK4)
\item ModelMaker 3.0 has been used on a Windows Seven Virtual Machine
\item vle 1.3 has been used Ubuntu 14.04
\item the laptop used is an i7(2.7GHz)
\end{itemize}
\begin{block}{Simulators}
All the simulators we have been used for this study, even those of ModelMaker are available in the package ibm.computation.time.
\end{block}
\end{frame}
\begin{frame}{The simulator}
\begin{itemize}
\item A simple LotkaVoltera model with 2 variables an 4 parameters
\item RK4 with fixed step length (100 steps per time steps)
\item 100000 steps
\item No visualisation
\item A Storage view is used on the VLE side, to store at the resolution of the simulation, not the one of the RK4 method.
\end{itemize}
\end{frame}
\section{Computation time}
\subsection{ModelMaker \& vle.extension.differential-equation}
\begin{frame}{ModelMaker \& vle.extension.differential-equation}
\begin{itemize}
\item The unit of Table~\ref{tab:MMagainstpureODE} is the second.
\item On one side we use the duration given by ModelMaker
\item On the other side we use the user time given by the unix command time
\item MM\_vectorized : the compartments of the model are vectorized
\item MM\_Duplicated : the diagrams compartments are duplicated
\item VLE\_ODE : the models are duplicated
\end{itemize}
\begin{table}
\centering
\begin{tabular}{r|c|c|c}
Quantity & MM\_Duplicated & MM\_vectorized & VLE\_ODE\\\hline
x1 & 9.4 & 9.4 & 3.539\\
x2 & 17.9 & 46.9 & 6.934\\
x10 & 84.8 & 217.7 & 34.391
\end{tabular}
\caption{\label{tab:MMagainstpureODE} ModelMaker and pure ODE models inside VLE}
\end{table}
\end{frame}
\begin{frame}{ModelMaker \& vle.extension.ibm}
\begin{itemize}
\item The unit of Table~\ref{tab:MMagainstIBM} is the second.
\item On one side we use the duration given by ModelMaker
\item On the other side we use the user time given by the unix command time
\item MM\_vectorized : the compartments of the model are vectorized
\item VLE\_ODE : the models are duplicated
\item VLE\_IBM : the models are duplicated and connected to a unsensitive controler to test the weight of the communication first
\end{itemize}
\begin{table}
\centering
\begin{tabular}{r|c|c|c}
Quantity & MM\_vectorized & VLE\_ODE & VLE\_IBM\\\hline
x1 & 9.4 & 3.539 & 20.328\\
x2 & 46.9 & 6.934 & 36.742\\
x10 & 217.7 & 34.391 & 199.835
\end{tabular}
\caption{\label{tab:MMagainstIBM} ModelMaker and IBM inside VLE}
\end{table}
\end{frame}
\subsection{Conclusion}
\begin{frame}{Conclusion}
\begin{itemize}
\item When using pure ODE, with no comunications between the models, the computation times of the solution provided by the vle.extension.differential-equation package seems to be interresting, and could be x3 faster. And even more if considering the vector approach of ModelMaker.
\item As soon as we use the communication architecture of the vle.extension.ibm, we loose much. We communicate at the resolution of integration communication. It can cost an x6 coefficient.
\end{itemize}
\end{frame}
\subsection{Prospect}
\begin{frame}{Prospect}
\begin{itemize}
\item Using the same OS on the same hardware to provide a more accurate study
\item Developping a output frequency filter for the outputs of the ODE of VLE
\item Testing with much biggers simulators in terms of individuals (x100, x1000,...)
\item Getting farther as soon as the output filter will be available
\item during the study we also find out that the way the controler does gather event from outside is not optimal, this can be also fixed.
\end{itemize}
\end{frame}
\end{document}