a logical indicating if variables should taken from
local scope(s) or only from global.
namespace
a character given a namespace where to search for the
object.
debug
a logical indicating extended information is given upon
action to be done (e.g. print exported variables, print context of
local variables etc.).
except
character vector with names of objects not to
export/remove
hidden
also remove hidden names (starting with a dot)?
sep
a character string separating elements in x
master
a logical indicating if executed on master as well
...
additional arguments to pass to standard function
package
name of the package. Check library for details.
pos
position in search path to load library.
warn.conflicts
warn on conflicts (see "library").
keep.source
see "library". Please note: this argument has only
effect on R-2.x, starting with R-3.0 it will only be a placeholder
for backward compatibility.
verbose
enable verbose messages.
version
version of library to load (see "library").
encoding
encoding of library to load (see "library").
lib.loc
a character vector describing the location of the R
library trees to search through, or 'NULL'. Check library for
details.
character.only
a logical indicating package can be assumed to
be a character string. Check library for details.
file
filename of file to read. Check source for
details
stopOnError
a logical indicating if function stops on
failure or still returns. Default is TRUE.
type
a character determine which random number generator should
be used for clusters. Allowed values are "RNGstream" for L'Ecuyer's
RNG or "SPRNG" for Scalable Parallel Random Number Generators.
para
additional parameters for the RNGs.
seed
Seed for the RNG.
prngkind
type of RNG, see snow documentation.
Details
The current functions are little helpers to make cluster programming
easier.
All of these functions also work in sequential mode without any
further code changes.
sfLibrary loads an R-package on all nodes, including
master. Use this function if slaves need this library,
too. Parameters are identically to the R-build in funtion
library. If a relative path is given in lib.loc,
it is converted to an absolute path.
As default sfLibrary stops on any error, but this can be
prevented by setting stopOnError=FALSE, the function is returning
FALSE then. On success TRUE is returned.
sfSource loads a sourcefile on all nodes, including master. Use
this function if the slaves need the code as well. Make sure the file
is accessible on all nodes under the same path. The loading is done
on slaves using source with fixes parameters:
local=FALSE, chdir=FALSE, echo=FALSE, so the files is loaded
global without changing of directory.
As default sfSource stops on any error, but this can be
prevented by setting stopOnError=FALSE, the function is returning
FALSE then. On success TRUE is returned.
sfExport exports variables from the master to all
slaves. Use this function if slaves need acccess to these variables as
well. sfExport features two execution modes: local and global.
If using local mode (default), variables for export are searched
backwards from current environment to globalenv(). Use this mode
if you want to export local variables from functions or other
scopes to the slaves. In global mode only global variables from master
are exported.
Note: all exported variables are global on the slaves!
If you have many identical named variables in different scopes, use
argument debug=TRUE to view the context the exported variable
is coming from.
Variables are given as their names or as a
character vector with their names using argument list.
sfExportAll exports all global variables from the master to all
slaves with exception of the
given list. Use this functions if you want to export mostly all
variables to all slaves.Argument list is a character vector
with names of the variables not to export.
sfRemove removes a list of global (previous exported or
generated) variables from slaves and (optional) master.
Use this function if there are large further unused variables
left on slave. Basically this is only interesting if you have more than
one explicit parallel task in your program - where the danger is slaves
memory usage exceed.
If argument master is given, the variables are removed from
master as well (default is FALSE).
Give names of variables as arguments, or use argument list
as a character vector with the names. For deep cleaning of slave
memory use sfRemoveAll.
sfRemoveAll removes all global variables from the slaves. Use
this functions if you want to remove mostly all
variables on the slaves. Argument list is a character vector
with names of the variables not to remove.
sfCat is a debugging function printing a message on all slaves
(which appear in the logfiles).
sfClusterSplit splits a vector into one consecutive piece for
each cluster and returns as a list with length equal to the number of
cluster nodes. Wrapper for snow function clusterSplit.
sfClusterCall calls a function on each node and returns list of
results. Wrapper for snow function clusterCall.
sfClusterEvalQ evaluates a literal expression on all
nodes. Wrapper for snow function clusterEvalQ.
sfTest is a simple unit-test for most of the build in functions.
It runs tests and compares the results for the correct behavior. Note
there are some warnings if using, this is intended (as behavior for
some errors is tested, too). use this if you are not sure all nodes are
running your R-code correctly (but mainly it is implemented for
development).
See Also
See snow documentation for details on wrapper-commands:
snow-parallel
Examples
## Not run:
sfInit( parallel=FALSE )
## Now works both in parallel as in sequential mode without
## explicit cluster handler.
sfClusterEval( cat( "yummie\n" ) );
## Load a library on all slaves. Stop if fails.
sfLibrary( tools )
sfLibrary( "tools", character.only=TRUE ) ## Alternative.
## Execute in cluster or sequential.
sfLapply( 1:10, exp )
## Export global Var
gVar <- 99
sfExport( "gVar" )
## If there are local variables with same name which shall not
## be exported.
sfExport( "gVar", local=FALSE )
## Export local variables
var1 <- 1 ## Define global
var2 <- "a"
f1 <- function() {
var1 <- 2
var3 <- "x"
f2 <- function() {
var1 <- 3
sfExport( "var1", "var2", "var3", local=TRUE )
sfClusterCall( var1 ) ## 3
sfClusterCall( var2 ) ## "a"
sfClusterCall( var3 ) ## "x"
}
f2()
}
f1()
## Init random number streams (snows functions, build upon
## packages rlecuyer/rsprng).
sfClusterCall( runif, 4 )
sfClusterSetupRNG() ## L'Ecuyer is default.
sfClusterCall( runif, 4 )
sfClusterSetupRNG( type="SPRNG", seed = 9876)
sfClusterCall( runif, 4 )
## Run unit-test on main functions.
sfTest()
## End(Not run)