Introduction to Multiprocessing

At any given time, a technological limit exists on the speed with which a single processor chip can operate. If a system's workload cannot be handled satisfactorily by a single processor, one response is to apply multiple processors to the problem.

The success of this response depends not only on the skill of the system designers, but also on whether the workload is amenable to multiprocessing. In terms of human tasks, adding people might be a good idea if the task is answering calls to a toll-free number, but is dubious if the task is driving a car.

If improved performance is the objective of a proposed migration from a uniprocessor to a multiprocessor system, the following conditions must be true:

• The workload is processor-limited and has saturated its uniprocessor system.
• The workload contains multiple processor-intensive elements, such as transactions or complex calculations, that can be performed simultaneously and independently.
• The existing uniprocessor cannot be upgraded or replaced with another uniprocessor of adequate power.

Although unchanged single-thread applications normally function correctly in a multiprocessor environment, their performance often changes in unexpected ways. Migration to a multiprocessor can improve the throughput of a system, and can improve the execution time of complex, multithreaded applications, but seldom improves the response time of individual, single-thread commands.

Getting the best possible performance from a multiprocessor system requires an understanding of the operating-system and hardware-execution dynamics that are unique to the multiprocessor environment.

This chapter includes the following major sections:

• Symmetrical Multiprocessor (SMP) Concepts and Architecture
• SMP Performance Issues
• SMP Workloads
• SMP Thread Scheduling
• Thread Tuning
• SMP Tools