OPTIMIZING multiprocessor systems

摘要

Optimizing a system means more than getting as close as you can to 100% usage. It also involves getting close to 0% inefficiency. Designing systems using multiple processors, each with multiple cores running multiple threads, presents a difficult optimization challenge. Higher order languages, framework environments, APIs, resource managers, and a slew of other abstractions enable you to step back from the complexity of a problem and view it from a higher, simpler, and more comprehensible level. Abstraction's great advantage is that you can quickly sketch out your design and fine-tune it where you need better performance. In other words, optimizing a system means determining where abstraction introduces the most inefficiencies and then selectively replacing those abstractions with more concrete descriptions to remove those inefficiencies.