Multicore Application Programming: Identifying Opportunities for Parallelism
- Using Multiple Processes to Improve System Productivity
- Multiple Users Utilizing a Single System
- Improving Machine Efficiency Through Consolidation
- Using Parallelism to Improve the Performance of a Single Task
- Parallelization Patterns
- How Dependencies Influence the Ability Run Code in Parallel
- Identifying Parallelization Opportunities
This chapter discusses parallelism, from the use of virtualization to support multiple operating systems to the use of multiple threads within a single application. It also covers the concepts involved in writing parallel programs, some ways of visualizing parallel tasks, and ways of architecting parallel applications. The chapter continues with a discussion of various parallelization strategies, or patterns. It concludes by examining some of the limitations to parallelization. By the end of the chapter, you should be in a position to understand some of the ways that a system can support multiple applications and that an existing application might be modified to utilize multiple threads. You will also be able to identify places in the code where parallelization might be applicable.
Using Multiple Processes to Improve System Productivity
Consider a home computer system. This will probably have only one active user at a time, but that user might be running a number of applications simultaneously. A system where there is a single core produces the illusion of simultaneous execution of multiple applications by switching between the active applications many times every second. A multicore system has the advantage of being able to truly run multiple applications at the same time.
A typical example of this happens when surfing the Web and checking e-mail. You may have an e-mail client downloading your e-mail while at the same time your browser is rendering a web page in the background. Although these applications will utilize multiple threads, they do not tend to require much processor time; their performance is typically dominated by the time it takes to download mail or web pages from remote machines. For these applications, even a single-core processor often provides sufficient processing power to produce a good user experience. However, a single-core processor can get saturated if the e-mail client is indexing mail while an animation-heavy web page is being displayed.
In fact, these applications will probably already take advantage of multiple threads. Figure 3.1 shows a newly opened instance of Mozilla Firefox launching 20 threads. A consequence of this is that just by having a multicore processor, the performance of the system will improve because multiples of those threads can be executed simultaneously—and this requires no change to the existing applications.
Figure 3.1 Windows Process Explorer showing thread activity in Mozilla Firefox
Alternatively, there are a number of tasks we perform on our personal computer systems that are inherently compute intensive, such as playing computer games, encoding audio for an MP3 player, transforming one video format into another suitable for burning to DVD, and so on. In these instances, having multiple cores can enable the work to take less time by utilizing additional cores or can keep the system responsive while the task is completed in the background.
Figure 3.2 shows the system stack when a single user runs multiple applications on a system.
Figure 3.2 Single user on system
It is also possible to have multiple users in a home environment. For example, on Windows, it is quite possible for one user to be logged in and using the computer while another user, having logged in earlier, has set some other applications running. For example, you may have left some DVD-authoring package running in the background while another user logs into their account to check their e-mail.