Parallel R

Chapter 4. parallel

A new parallel programming package named parallel will be included in R 2.14.0, tentatively scheduled for release on October 31, 2011. It is derived from the snow and multicore packages, providing many of the same functions as those packages. Some of the functions derived from multicore have been renamed by adding the prefix “mc.”, and some of the arguments to mclapply() have been changed a bit, but if you have read the snow and multicore chapters of this book, you will have very little difficulty learning to use parallel.

This is an exciting development, since it makes parallel computing in R more mainstream. Hopefully the parallel package will be used from other standard packages, giving many more users the benefit of parallel computing, perhaps without knowing that they’re using it.^[39]

An important feature of parallel is its integration with the new L’Ecuyer-CMRG random number generator (RNG), also new in R 2.14.0. The seed of this generator can be easily advanced a given number of steps, making it very useful as a parallel RNG. This is accomplished using the same concepts used in the rlecuyer package, but it is a completely new implementation, so parallel has no dependency on the rlecuyer package itself.

In particular, the multicore derived functions in parallel now have true parallel RNG support, solving the biggest “gotcha” in the multicore package.

Warning

This chapter was written using an experimental version of the parallel package using the development ...