Multicore Software Development Techniques

Applications, Tips, and Tricks

Robert Oshana

• Publisher: Newnes
• Year: 2015
• Print ISBN: 9780128009581
• eBook ISBN: 9780128010372
• Topic: Computer Science
• Subtopic: Digital Media
• Index: Computer Science

This book provides a set of practical processes and techniques used for multicore software development. It is written with a focus on solving day to day problems using practical tips and tricks and industry case studies to reinforce the key concepts in multicore software development.Coverage includes: - The multicore landscape- Principles of parallel computing- Multicore SoC architectures- Multicore programming models- The Multicore development process- Multicore programming with threads- Concurrency abstraction layers- Debugging Multicore Systems- Practical techniques for getting started in multicore development- Case Studies in Multicore Systems Development- Sample code to reinforce many of the concepts discussed- Presents the 'nuts and bolts' of programming a multicore system- Provides a short-format book on the practical processes and techniques used in multicore software development- Covers practical tips, tricks and industry case studies to enhance the learning process

Chapter 1

Principles of Parallel Computing

Abstract

The principle of parallel computing is using multiple processors in parallel to solve problems more quickly than with a single processor, or with less energy consumption. This chapter will overview some key principles including concurrency versus parallelism, SMP and AMP systems, limits to parallel computing, and Amdahl’s Law.

Keywords

Concurrency; parallelism; SMP; AMP; Amdahl’s Law; Gustufsons Law

A multicore processor is a computing device that contains two or more independent processing elements (referred to as “cores”) integrated on to a single device, that read and execute program instructions. There are many architectural styles of multicore processors, and many application areas, such as embedded processing, graphics processing, and networking.

There are many factors driving multicore adoption:

• Increases in mobile traffic

• Increases in communication between multiple devices

• Increase in semiconductor content (e.g., automotive increases in semiconductor content are driving automotive manufacturers to consider multicore to improve affordability, “green” technology, safety, and connectivity, see Figure 1.1)

Figure 1.1 Semiconductor content in automotive is increasing.

A typical multicore processor will have multiple cores which can be the same (homogeneous) or different (heterogeneous) accelerators (the more generic term is “processing element”) for dedicated functions such as video or network acceleration, as well as a number of shared resources (memory, cache, peripherals such as ethernet, display, codecs, UART, etc.) (Figure 1.2)

Figure 1.2 A generic multicore system (left) and an example multicore device from industry (right).

1.1 Concurrency versus Parallelism

There are important differences between concurrency and parallelism as they relate to multicore processing.

Concurrency: A condition that exists when at least two software tasks are making progress, although at different times. This is a more generalized form of parallelism that can include time-slicing as a form of virtual parallelism. Systems that support concurrency are designed for interruptability.

Parallelism: A condition that arises when at least two threads are executing simultaneously. Systems that support parallelism are designed for independentability, such as a multicore system.

A program designed to be concurrent may or may not be run in parallel; concurrency is more an attribute of a program, parallelism may occur when it executes (see Figure 1.3).

Figure 1.3 Concurrency versus parallelism.

It is time to introduce an algorithm that should be memorized when thinking about multicore systems. Here it is:

$\text{High}-\text{performance}=\mathrm{parallelism}+\mathrm{memory}\mathrm{hierarchy}–\mathrm{contention}$

• “Parallelism” is all about exposing parallelism in the application

• “Memory hierarchy” is all about maximizing data locality in the network, disk, RAM, cache, core, etc.

• “Contention” is all about minimizing interactions between cores (e.g., locking, synchronization, etc.)

To achieve the best HPC or “High Peformance Computing” result, to get the best performance we need to get the best possible parallelism, use memory efficiently, and reduce the contention. As we move forward we will touch on each of these areas.

1.2 Symmetric and Asymmetric Multiprocessing

Efficiently allocating resources in multicore systems can be a challenge. Depending on the configuration, the multiple software components in these systems may or may not be aware of how other components are using these resources. There are two primary forms of multiprocessing, as shown in Figure 1.4;

• Symmetric multiprocessing

• Asymmetric multiprocessing

Figure 1.4 Asymmetric multiprocessing (left) and symmetric multiprocessing (right).

1.2.1 Symmetric Multiprocessing

Symmetric multiprocessing (SMP) uses a single copy of the operating system on all of the system’s cores. The operating system has visibility into all system element, and can ...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• Dedication
• Chapter 1. Principles of Parallel Computing
• Chapter 2. Parallelism in All of Its Forms
• Chapter 3. Multicore System Architectures
• Chapter 4. Multicore Software Architectures
• Chapter 5. Multicore Software Development Process
• Chapter 6. Putting it All Together, A Case Study of Multicore Development
• Chapter 7. Multicore Virtualization
• Chapter 8. Performance and Optimization of Multicore Systems
• Chapter 9. Sequential to Parallel Migration of Software Applications
• Chapter 10. Concurrency Abstractions
• Appendix A. Source Code Examples
• Index