Parallel Computing for Data Science

Chapter 3

Principles of Parallel Loop

Scheduling

Many applications of parallel programming, both in R and in general, in-

volve the parallelization of for loops. As will be explained shortly, this at

first would seem to be a very easily programmed class of applications, but

there can be serious performance issues.

First, though, let’s define the class under consideration. Throughout this

chapter, it will be assumed that the iterations of a loop are independent of

each other, meaning that the execution of one iteration does not use the

results of a previous one.

Here is an example of code that does not satisfy this condition:

t o t a l <− 0

for ( i i n 1