Send function calls with different arguments to different processors in R using parallel package

Question

I am trying to use the parallel package in R to send four different function calls to four different processors but am really lost as to how to assign different cores to do different work. I've read through the documentation for the parallel package, doParallel, Rmpi, and foreach in R. I've seen many posts using mclapply for calling different functions with the same argument. I'd like to call the same function with different arguments.

This is pseudocode of what I'd like to accomplish:

BEGIN parallel (core)
if(core == 1)
   foo(5, 4, 1/2, 3, "a")
if(core == 2)
   foo(5, 3, 1/3, 1, "b")
if(core == 3)
   foo(5, 4, 1/4, 1, "c")
if(core == 4)
   foo(5, 2, 1/5, 0, "d")
END parallel

This seems to be a perfect application to parallel computing since these four separate function calls can act independently to solve the problem I am working on. I don't know how to do this in R though.

Answer 1

You could use the clusterApply function from the parallel package:

library(parallel)
cl <- makeCluster(4)
clusterExport(cl, "foo")
cores <- seq_along(cl)
r <- clusterApply(cl[cores], cores, function(core) {
  if (core == 1) {
    foo(5, 4, 1/2, 3, "a")
  } else if (core == 2) {
    foo(5, 3, 1/3, 1, "b")
  } else if (core == 3) {
    foo(5, 4, 1/4, 1, "c")
  } else if (core == 4) {
    foo(5, 2, 1/5, 0, "d")
  }
})

This is very similar to your pseudocode and demonstrates how you can direct particular tasks to particular cluster workers using clusterApply. Note that by changing the value of cores, you can execute on any subset of the cluster workers that you choose.

If a "core ID" isn't really important, you can pass different arguments to the function by iterating over vectors for each of the arguments using the foreach package:

library(doParallel)
registerDoParallel(cl)
r2 <- foreach(a1=c(5,5,5,5), a2=c(4,3,4,2), a3=c(1/2,1/3,1/4,1/5),
              a4=c(3,1,1,0), a5=c("a","b","c","d")) %dopar% {
  foo(a1, a2, a3, a4, a5)
}