I am currently testing the new doParallel package and I compared the performance to the old doSMP backend. I observed that the doParallel backend consumes significantly more memory. It almost looks as if the memory consumption of the RGUI accumulates the memory consumption of the corresponding slaves. I tried the following simple example and the memory consumption went up to ~800MB (compared to <100MB if one uses doSMP).
library(doParallel)
workers=makeCluster(3)
registerDoParallel(workers)
Runs=c(1:1000)
Output_parallel=foreach(i=Runs, .combine='+') %dopar% {
result=rnorm(100000)
}
stopCluster(workers)
Is there an explanation for this issue?