To me, this is a great argument in favor of building on top of products like Akka rather than implementing the same functionality yourself. It's pretty clear from Doug Lea's description (http://cs.oswego.edu/pipermail/concurrency-interest/2012-Jan...) that this set of changes would be non-obvious to people who aren't extremely knowledgeable about the internals of Java and the semantics of its memory model.
Alone, work stealing and striped locking are not really new solutions to the problem, but their implementation in JSR166y is really impressive. This is the kind of framework investment that has high leverage, and will make it much easier to write highly performant concurrent code in Java (and related languages) and reduce the need to couple business logic and threading concerns for high performance.
Alone, work stealing and striped locking are not really new solutions to the problem, but their implementation in JSR166y is really impressive. This is the kind of framework investment that has high leverage, and will make it much easier to write highly performant concurrent code in Java (and related languages) and reduce the need to couple business logic and threading concerns for high performance.