parallel foreach loops produce mclapply error

Question

Those are my first steps with parallel computing in R. The code below results in the following error. I am clueless, since there is no mclapply function in what I wrote, at least I did not put it explicitly.

Error:

Error in mclapply(argsList, FUN, mc.preschedule = preschedule, mc.set.seed = set.seed,  : 
  (list) object cannot be coerced to type 'integer'
Calls: %dopar% -> <Anonymous> -> mclapply
Execution halted

Code:

dist<-array(0, dim=c(320,500,25))

mc<-8
cl<-makeCluster(mc)
registerDoMC(cl)
opts<-list(chunkSize=10)

for(a in 1:25) {
  dist[,,a]<-foreach(x=1:500, .combine='cbind', .options.mc=opts) %:% 
    foreach(y=1:320, .combine='c') %dopar% {
      gcd.slc(crdsx[y,x], crdsy[y,x], lot[a,5], lot[a,4])
    }
}
stopCluster(cl)

On a different machine, it works nicely with

registerDoParallel(cl)

instead of

registerDoMC(cl)

Answer 1

You're getting that error because registerDoMC expects an integer argument, not a cluster object, while registerDoParallel expects either an integer or a cluster object. Basically, you need to decide which package to use and not mix them.

If you use doMC, then you never create a cluster object. A minimal doMC example looks like:

library(doMC)
registerDoMC(3)
foreach(i=1:10) %dopar% sqrt(i)

The doParallel package is a mashup of the doMC and doSNOW packages, and so you don't need to use cluster objects. Converting the previous example to doParallel is very simple:

library(doParallel)
registerDoParallel(3)
foreach(i=1:10) %dopar% sqrt(i)

The confusing thing is that on Windows, doParallel will actually create and use a cluster object behind the scenes, while on Linux and Mac OS X, it doesn't use a cluster object because it uses mclapply just as in the doMC package. I think that is convenient, but it can be a source of confusion.