Optimized C++

Chapter 12. Optimize Concurrency

It’s tough to make predictions, especially about the future.

Yogi Berra (1925–2015), baseball legend and inadvertent humorist

This quip appeared in English-language physics and economics journals prior to becoming a “Yogi-ism.” It is also attributed as a Danish proverb. It seems unlikely, though, that Berra misappropriated it from any of these sources.

All but the smallest modern computers process multiple execution streams concurrently. They contain multiple CPU cores, graphics processors with hundreds of simple cores, audio processors, disk controllers, network cards, and even keyboards with separate computing power and memory. Like it or not, developers live in a concurrent world, and must understand how to program concurrent activities.

Software practices for concurrency evolved in a world of single-core processors. Since 2005, the advent of multicore microprocessors providing true (versus time-sliced) concurrency has altered the development landscape, informing best practices with new rules. These rules may be unfamiliar even to developers experienced with concurrency issues that arise in single-processor systems.

If the future direction of processor development leads to commercial devices with dozens or hundreds of cores, programming best practices will continue to change. Several competing tools promoting fine-grained concurrency have arrived that anticipate this future. However, general-purpose hardware with many cores has yet ...