First-class synchronization barriers

Our purpose is to promote a second-class mechanism - the synchronization barrier - to a first-class value. We introduce the synchron, a novel synchronization mechanism that enables the coordination of a dynamically varying set of concurrent threads that share access to a first-class synchronization token. We demonstrate how synchrons can be used to modularly manage resources in cases where existing techniques are either inapplicable or non-modular. In particular, synchronized lazy aggregates enable the first space-efficient aggregate data decomposition of a wide range of algorithms. We also introduce explicit-demand graph reduction, a new semantic framework that we have developed to describe concurrency and explain the meaning of a synchron rendezvous.