Measuring Parallel Performance: Optimizing A Concurrent Queue

摘要

How would you write a fast, internally synchronized queue, one that callers can use without any explicit external locking or other synchronization? Let us count the ways...or four of them, at least, and compare their performance. We'll start with a baseline program and then successively apply three optimization techniques, each time stopping to measure each change's relative performance for queue items of different sizes to see how much each trick really bought us.