ABSTRACT
In this chapter, we will describe the background to multicore processors, describe their architectures, and lay the groundwork for the remaining of the book on programming these processors. Multicore processors integrate multiple processor cores on the same integrated circuit chip (die), which are then used collectively to achieve higher overall performance. Constructing a system with multiple processors and using them collectively is a rather obvious idea for performance improvement. In fact, it became evident in the early days of computer design as a potential way of increasing the speed of computer systems. A computer system constructed with multiple processors
that are intended to operate together is called a parallel computer historically, and programming the processors to operate together is called parallel programming. Parallel computers and parallel programming have a long history. The term parallel programming is used by Gill in 1958 (Gill 1958), and his definition of parallel programming is essentially the same as today. In this chapter, we will first explore the previous work and will start by establishing
the limits for performance improvement of processors operating in parallel to satisfy ourselves that there is potential for performance improvement. Then, we will look at the different ways that a system might be constructed with multiple processors. We continue with an outline of the improvements that have occurred in the design of the processors themselves, which have led to enormous increase in the speed of individual processors. These improvements have been so dramatic that the added complexities of parallel computers have limited their use mostly to very high-performance computing in the past. Of course, with improvements in the individual processor, so parallel computers constructed with them also improve proportionately. But programming the multiple processors for collective operation is a challenge, and most demands outside scientific computing have been satisfied with single processor computers, relying on the ever-increasing performance of processors. Unfortunately, further improvements in single processor designs hit major obstacles in the early 2000s, which we will outline. These obstacles led to the multicore approach. We describe architectural designs for amulticore processor and conclude with an outline of themethods for programming multicore systems as an introduction to subsequent chapters on multicore programming.
