% \VignetteIndexEntry{Running ISA on multiple CPUs}
\documentclass{article}
\usepackage{ragged2e}

\newcommand{\Rfunction}[1]{\texttt{#1()}}
\newcommand{\Rpackage}[1]{\texttt{#1}}
\newcommand{\Rclass}[1]{\texttt{#1}}
\newcommand{\Rargument}[1]{\textsl{#1}}
\newcommand{\filename}[1]{\texttt{#1}}
\newcommand{\listentry}[1]{\texttt{#1}}

\begin{document}

\setkeys{Gin}{width=\textwidth}

\title{Running ISA in parallel with the \Rpackage{snow} package}
\author{G\'abor Cs\'ardi}
\maketitle

% \RaggedRight

% \tableofcontents

<<Roptions,echo=FALSE,results=hide>>=
options(width=60)
options(continue=" ")
@ 

\section{Running ISA in parallel}

In this document we show a little example on how to speed up ISA
analysis~\cite{sa,isa,isamod,isapackage} by running the ISA iterations
in parallel, on a computer cluster, or a multiprocessor machine.

Since a typical ISA analysis consists of using a range of row/column
thresholds and these runs are independent of each other; it is trivial to
parallelize the task by performing the iterations for different
threshold parameters on different processors or computers. Here we
show an example on how to do this easily with the
\Rpackage{snow}~\cite{snow} and the \Rpackage{Rmpi}~\cite{rmpi}
packages:
<<parallelloadpackage>>=
library(isa2)
library(snow)
library(Rmpi)
@ 

We generate some simple in-silico data.
<<paralleldata>>=
pdata <- isa.in.silico()
@

Next, we create the MPI cluster, with eight working nodes. You need to have a
working MPI installation for this. See more in the documentation of
the \Rpackage{Rmpi} and \Rpackage{snow} packages.
<<createcluster>>=
clu <- makeMPIcluster(8)
@ 

Now we load the \Rpackage{isa2} package on all nodes and distribute
the input data to them.
<<load isa2 package>>=
invisible(clusterEvalQ(clu, library(isa2)))
clusterExport(clu, "pdata")
@ 

Next, we create a big matrix in which each row is a combination of the
threshold parameters. This will be needed for the parallel run.
<<parthr>>=
thr <- seq(1,3,by=0.2)
thr.list <- expand.grid(thr, thr)
@ 

First we run the ISA on a single processor only, and measure the
running time.
<<seqrun,cache=TRUE>>=
system.time(modules <- isa(pdata[[1]], thr.row=thr, thr.col=thr))
@ 

Let us now do a parallel run, again, with measuring the running
time. If you are really running this on multiple CPUs, then it is much
faster.
<<parrun,cache=TRUE>>=
system.time(modules.par <- parApply(clu, thr.list, 1, function(x) {
  isa(pdata[[1]], thr.row=x[1], thr.col=x[2])
}))
@

Finally, stop the cluster.
<<closecluster>>=
stopCluster(clu)
@ 

\section{Session information}

The version number of R and packages loaded for generating this
vignette were:

<<sessioninfo,results=tex,echo=FALSE>>=
toLatex(sessionInfo())
@ 

\bibliographystyle{apalike}

\begin{thebibliography}{}

\bibitem[Bergmann et~al., 2003]{isa}
Bergmann, S., Ihmels, J., and Barkai, N. (2003).
\newblock Iterative signature algorithm for the analysis of large-scale gene
  expression data.
\newblock {\em Phys Rev E Nonlin Soft Matter Phys}, page 031902.

\bibitem[Cs\'ardi, 2009]{isapackage}
Cs\'ardi, G. (Apr 1, 2009).
\newblock {\em isa2: The Iterative Signature Algorithm}.
\newblock R package version 0.1.1.

\bibitem[Ihmels, 2002]{sa}
Ihmels, J., Friedlander, G., Bergmann, S., Sarig, O., Ziv, Y., 
Barkai, N. (2002).
\newblock Revealing modular organization in the yeast transcriptional
  network.
\newblock { \em Nat Genet }, page 370--7.

\bibitem[Ihmels, 2004]{isamod}
Ihmels, J., Bergmann, S., Barkai, N. (2004).
\newblock Defining transcription modules using large-scale gene
  expression data.
\newblock { \em Bioinformatics}, page 1993--2003.

\bibitem[Tierney, 2009]{snow}
Luke Tierney, A. J. Rossini, Na Li and H. Sevcikova
(2009). 
\newblock {\em snow: Simple Network of Workstations}. 
\newblock R package version 0.3-3.

\bibitem[Yu, 2007]{rmpi}
Hao Yu (2007). 
\newblock {\em Rmpi: Interface (Wrapper) to MPI (Message-Passing Interface)}. 
\newblock R package version 0.5-5.

\end{thebibliography}

\end{document}