pbdMPI (version 0.1-5)

get job id: Divide Job ID by Ranks

Description

This function obtains job id which can be used to divide jobs.

Usage

get.jid(n, method = .SPMD.CT$divide.method, all = FALSE,
          comm = .SPMD.CT$comm)

Arguments

n

total number of jobs.

method

a way to divide jobs.

all

indicate if return all id for each processor.

comm

a communicator number.

Value

• get.id returns a vector containing job id for each individual processor if all = FALSE. While it returns a list containing all job id for all processor if all = TRUE. The list has length equal to COMM.SIZE.

Details

n is total number of jobs needed to be divided into all processors (comm.size(comm), i.e. 1:n will be split according to the rank of processor (comm.rank(comm)) and method. Job id will be returned. Currently, three possible methods are provided.

"block" will use return id's which are nearly equal size blocks. For example, 7 jobs in 4 processors will have jid=1 for rank 0, jid=2,3 for rank 1, jid=4,5 for rank 2, and jid=6,7 for rank 3.

"cycle" will use return id's which are nearly equal size in cycle. For example, 7 jobs in 4 processors will have jid=1,5 for rank 0, jid=2,6 for rank 1, jid=3,7 for rank 2, and jid=4 for rank 3.

References

Programming with Big Data in R Website: http://r-pbd.org/

Examples

# Save code in a file "demo.r" and run with 4 processors by
# > mpiexec -np 4 Rscript demo.r

### Initial
library(pbdMPI, quiet = TRUE)
init()

### Examples
comm.cat(">>> block\n", quiet = TRUE)
jid <- get.jid(7, method = "block")
comm.print(jid, all.rank = TRUE)

comm.cat(">>> cycle\n", quiet = TRUE)
jid <- get.jid(7, method = "cycle")
comm.print(jid, all.rank = TRUE)

comm.cat(">>> block (all)\n", quiet = TRUE)
alljid <- get.jid(7, method = "block", all = TRUE)
comm.print(alljid)

comm.cat(">>> cycle (all)\n", quiet = TRUE)
alljid <- get.jid(7, method = "cycle", all = TRUE)
comm.print(alljid)

### Finish
finalize()