So far, we have learned various ways to optimize the performance of R programs running serially, that is, in a single thread. This does not take advantage of the multiple CPU cores most computers have nowadays. Parallel computing allows us to tap into them by splitting our implementations in multiple parts that are sent to these processors independently, and it has the potential to accelerate programs when a single thread is an important bottleneck.
Parallelizing real-world applications can be a very challenging task, and it requires deep software as well as hardware knowledge. The extent of possible parallelization depends on the particular algorithm we're working with, and there are many types of parallelizations available. Furthermore, parallelization is not a yes/no decision; it involves a continuous scale. On one side of the...
