clusterOpts: Cluster and large dataset management utilities.

In benilton/oligoClasses: Classes for high-throughput arrays supported by oligo and crlmm

Description

Tools to simplify management of clusters via 'snow' package and large dataset handling through the 'bigmemory' package.

Usage

ocSamples(n)
ocProbesets(n)

Arguments

n	integer representing the maximum number of samples/probesets to be processed simultaneously on a compute node.

Details

Some methods in the oligo/crlmm packages, like backgroundCorrect, normalize, summarize and rma can use a cluster (set through the 'foreach' package). The use of cluster features is conditioned on the availability of the 'ff' (used to provide shared objects across compute nodes) and 'foreach' packages.

To use a cluster, 'oligo/crlmm' checks for three requirements: 1) 'ff' is loaded; 2) an adaptor for the parallel backend (like 'doMPI', 'doSNOW', 'doMC') is loaded and registered.

If only the 'ff' package is available and loaded (in addition to the caller package - 'oligo' or 'crlmm'), these methods will allow the user to analyze datasets that would not fit in RAM at the expense of performance.

In the situations above (large datasets and cluster), oligo/crlmm uses the options ocSamples and ocProbesets to limit the amount of RAM used by the machine(s). For example, if ocSamples is set to 100, steps like background correction and normalization process (in RAM) 100 samples simultaneously on each compute node. If ocProbesets is set to 10K, then summarization processes 10K probesets at a time on each machine.

Warning

In both scenarios (large dataset and/or cluster use), there is a penalty in performance because data are written to disk (to either minimize memory footprint or share data across compute nodes).

Author(s)

Benilton Carvalho

Examples

   if(require(doMC)) {
       registerDoMC()
       ## tasks like summarize()
   }

benilton/oligoClasses documentation built on Dec. 27, 2019, 12:13 a.m.