Parallel Computing for Data Science

32 CHAPTER 2. OBSTACLES TO SPEED

Figure 2.2: Run Time Versus Number of Cores

By the way, for each of our nc workers, we had one invocation of R running

on the machine. There was also an additional invocation, for the manager.

However, this is not a performance issue in this case, as the manager spends

most of its time idle, waiting for the workers.

2.9 “Big O” Notation

With all this talk of physical obstacles to overcome, such as memory access

time, it’s important also to raise the question as to whether the application

itself is very parallelizable in the first place. One measure of that is “big

O” notation.

In our mutual outlinks example with an n ×