What's the difference between Concurrency and Parallelism?

The terms "concurrency" and "parallelism" are often used interchangeably, and while both relate to handling multiple tasks, they approach the problem differently. This confusion is so common that Rob Pike, one of Go's creators, even gave a talk specifically to clarify the distinction.

You'll frequently encounter this concept in technical interviews. While the interviewer might not ask directly, "What's the difference between concurrency and parallelism?", they'll present scenarios where understanding these approaches helps you propose the right solution.

Concurrency

Let's start with a real-world analogy. A skilled chef in a busy kitchen manages multiple tasks at once. After putting rice on to cook, while waiting for it to finish, the chef preps meat. Simultaneously, one pan fries eggs while another blanches vegetables.

Despite appearing to work on everything at once, the chef is actually doing only one thing at any given moment. Instead of waiting idle while the rice cooks, the chef switches to another task during that waiting period. For instance, while vegetables need two minutes of blanching, rather than standing idle, the chef moves to the eggs. Once those are done, the chef plates the blanched vegetables.

Single-core CPUs work similarly. A single CPU core can only execute one instruction at a time, but it spends significant time waiting for other components like disk reads or network requests. While waiting for task A to complete, the CPU can switch and start task B.

This rapid task switching to efficiently use available time is concurrency's key advantage. Because of this switching, concurrency makes it feel like multiple things are happening simultaneously, but it doesn't change the fact that the CPU handles one task at a time. More precisely, concurrency is about managing multiple tasks.

Parallelism

Now consider the same kitchen scenario in a Michelin-starred restaurant with multiple chefs. The head chef oversees several sous chefs and assistants. One chef specializes in appetizers, another in main courses, and a third in desserts. Here, multiple chefs work on different tasks simultaneously.

This mirrors parallelism. The key difference from concurrency is that parallelism has multiple actual execution units working simultaneously, such as multiple CPU cores handling different tasks. While concurrency is about managing multiple tasks, parallelism is about actually executing multiple tasks at the same time.

Different Limitations Require Different Approaches

Understanding the distinction opens the door to more nuanced analysis. When a program runs slowly, we can identify whether the bottleneck is computation-heavy, consuming all available CPU power, or something else. The slowdown might come from waiting for external operations like disk I/O or network responses.

The former case is called CPU-bound tasks, while the latter is called I/O-bound tasks.

Concurrency excels at handling I/O-bound tasks. Since I/O-bound tasks mean the CPU spends most time waiting for I/O results, there are frequent idle periods where the CPU can switch to other work.

Parallelism is ideal for CPU-bound tasks. A CPU-bound task means the existing CPU is already fully utilized, and the program runs slowly because computations aren't complete. Adding more concurrency won't help since the CPU has no idle time to handle additional tasks. However, parallelism lets you add more CPU cores to break through the existing limitation. Image compression and machine learning computations are examples of such tasks that benefit from parallel execution across multiple CPUs.

Summary

Concurrency addresses the problem: "If I have only one processor that can handle one task at a time, how do I effectively organize and schedule different tasks to minimize idle time?" Parallelism, on the other hand, addresses: "How can I use multiple processing units simultaneously to speed up task execution?"

Rob Pike, the Go language creator mentioned earlier, captured this distinction perfectly in his talk: "Concurrency is about dealing with lots of things at once. Parallelism is about doing lots of things at once."

Concurrency is about dealing with lots of things at once. Parallelism is about doing lots of things at once

Concurrency and parallelism aren't mutually exclusive. On modern multi-core processors, a program can exhibit both characteristics simultaneously. From Pike's perspective, concurrency is an architectural approach for breaking a program into independently runnable tasks. Once you design a program with concurrency in mind, it can take advantage of multiple CPU cores for true parallel execution. Therefore, good concurrent design is essential for enabling effective parallel computation.