Case Study I: Fast Synchronization of Threads

🔑 The reference PCs

You should take the performance numbers with a grain of salt. I’m not interested in the exact number for each variation of the algorithms on Linux and Windows. I’m more interested in getting a gut feeling of which algorithms may work and which algorithms may not work. I’m not comparing the absolute numbers of my Linux desktop with the numbers on my Windows laptop, but I’m interested to know if some algorithms work better on Linux or Windows.

When you want to synchronize threads more than once, you can use condition variables, std::atomic_flag, std::atomic<bool>, or semaphores. In this section, I want to answer the question: which variant is the fastest?

To get comparable numbers, I implement a ping-pong game. One thread executes a ping function (or ping thread for short), and the other thread a pong function (or pong thread for short). The ping thread waits for the pong-thread notification and sends the notification back to the pong thread. The game stops after 1,000,000 ball changes. I perform each game five times to get comparable performance numbers.

🔑 About the numbers

I made my performance test at the end of 2020 with the brand new Visual Studio compiler 19.28 because it already supported synchronization with atomics (std::atomic_flag and std::atomic) and semaphores. Additionally, I compiled the examples with ...