I’m often asked, if Dataflow offers true black-box components, inherent parallelism and has been around since the 1960′s, why is it not more common?
Simon Peyton Jones states the reason very clearly in a presentation about Data Parallelism (http://www.youtube.com/watch?v=NWSZ4c9yqW8 starting at about 6:13). Back in the 1980′s Simon and others built powerful Dataflow processors but Intel always came out with a much faster, sequential, processor in a few month’s time. There was no reason for developers to switch to Dataflow because soon, their slow application, will run at an acceptable rate on a new, faster microprocessor.
He continues by pointing out that we have now come to the end of Moore’s Law. The only way to get faster is by using multiple processors in parallel. Now everyone is looking around to for alternatives to threads and locks.
Dataflow is the most obvious solution.