High Performance Python, 2nd Edition

Chapter 8. Asynchronous I/O

Questions You’ll Be Able to Answer After This Chapter

• What is concurrency, and how is it helpful?

• What is the difference between concurrency and parallelism?

• Which tasks can be done concurrently, and which can’t?

• What are the various paradigms for concurrency?

• When is the right time to take advantage of concurrency?

• How can concurrency speed up my programs?

So far we have focused on speeding up code by increasing the number of compute cycles that a program can complete in a given time. However, in the days of big data, getting the relevant data to your code can be the bottleneck, as opposed to the actual code itself. When this is the case, your program is called I/O bound; in other words, the speed is bounded by the efficiency of the input/output.

I/O can be quite burdensome to the flow of a program. Every time your code reads from a file or writes to a network socket, it must pause to contact the kernel, request that the actual read happens, and then wait for it to complete. This is because it is not your program but the kernel that does the actual read operation, since the kernel is responsible for managing any interaction with hardware. The additional layer may not seem like the end of the world, especially once you realize that a similar operation happens every time memory is allocated; however, if we look back at Figure 1-3, we see that most of the I/O operations we perform are on devices that are orders of magnitude slower than ...