dclone (version 2.3-2)

clusterSplitSB: Size balancing

Description

Functions for size balancing.

Usage

clusterSplitSB(cl = NULL, seq, size = 1)
parLapplySB(cl = NULL, x, size = 1, fun, ...)
parLapplySLB(cl = NULL, x, size = 1, fun, ...)

Value

clusterSplitSB returns a list of subsets split with respect to size.

parLapplySB and parLapplySLB evaluates fun on elements of x, and return a result corresponding to x. Usually a list with results returned by the cluster.

Arguments

cl

A cluster object created by makeCluster the the package parallel.

x, seq

A vector to split.

fun

A function or character string naming a function.

size

Vector of problem sizes (approximate processing times) corresponding to elements of seq (recycled if needed). The default 1 indicates equality of problem sizes.

...

Other arguments of fun.

Author

Peter Solymos,

Details

clusterSplitSB splits seq into subsets, with respect to size. In size balancing, the problem is re-ordered from largest to smallest, and then subsets are determined by minimizing the total approximate processing time. This splitting is deterministic (reproducible).

parLapplySB and parLapplySLB evaluates fun on elements of x in parallel, similarly to parLapply. parLapplySB uses size balancing (via clusterSplitSB). parLapplySLB uses size and load balancing. This means that the problem is re-ordered from largest to smallest, and then undeterministic load balancing is used (see clusterApplyLB). If size is correct, this is identical to size balancing. This splitting is non-deterministic (might not be reproducible).

See Also

Related functions without size balancing: clusterSplit, parLapply.

Underlying functions: clusterApply, clusterApplyLB.

Optimizing the number of workers: clusterSize, plotClusterSize.

Examples

if (FALSE) {
cl <- makePSOCKcluster(2)
## equal sizes, same as clusterSplit(cl, 1:5)
clusterSplitSB(cl, 1:5)
## different sizes
clusterSplitSB(cl, 1:5, 5:1)
x <- list(1, 2, 3, 4)
parLapplySB(cl, x, function(z) z^2, size=1:4)
stopCluster(cl)
}