Abstract:
|
Despite the impressive improvement in computer power over the years, statisticians are trying to tackle even more difficult problems with computational intensive methods (resampling, MCMC, etc.), and more computing power is always desirable. Given the cost of new hardware, it is very attractive to go parallel and share the load among many existing commodity computers on a local area network. PVM (Parallel Virtual Machine) is a highly portable, high-level message passing system that supports parallel computing on heterogeneous compute clusters. RPVM is a R package that aims to integrate PVM with R's powerful and versatile computing environment. RPVM supports master/slave, divide and conquer, as well as single-program multiple data (SPMD) parallel programming paradigms. RPVM also include wrapper functions for SPRNG (Scalable Parallel Random Number Generator) library that provide high quality parallel random number streams. We introduce the use of RPVM through examples and discuss its applications in statistical computing.
|