Modern Computational Finance

CHAPTER 3Parallel C++

In this chapter, we introduce the C++11 Standard Threading Library, discuss its main components, and develop a basic thread pool. We reuse that thread pool to perform concurrent calculations in the rest of the book.

In order to understand concurrent programming and make sense of this chapter, we must first appreciate what are the different pieces involved, how they interact, and who is responsible for their management.

• Cores are the hardware processing units that execute sequences of instructions. Typical laptops and PCs have two to six physical cores today, typical workstations have eight cores and up.

  Intel's hyperthreading technology runs multiple (two) hardware threads per core. Hardware threads, also called logical cores, accelerate execution by switching logical cores on physical cores when beneficial, in accordance with a logic hard coded on the chip. From the application's point of view, hardware threads are seen and treated as physical cores.

  Performance benefits are entirely dependent on the number of cores on the executing machine. Multi-threaded applications run no faster (actually, slower) on single-threaded computers.¹

• Threads are the sequences of instructions executed on the machine's available cores. Every application creates at least one thread, called main thread, which starts when the application launches. Multi-threaded applications are those who explicitly create additional threads, with a syntax explained in Section 3.2.

  The threads ...