Distributed futures for efficient data transfer between parallel processes

Abstract

This paper defines distributed futures, a construct that provides at the same time a data container similar to a distributed vector, and a single synchronization entity that behaves similarly to a standard future. This simple construct makes it easy to program a composition, in a task-parallel way, of several massively data-parallel tasks. The approach is implemented and evaluated in the context of a bulk synchronous parallel (BSP) active object framework. CCS CONCEPTS• Software and its engineering → Parallel programming languages; Concurrent programming structures.

Promise Plus: Flexible Synchronization for Parallel Computations on Arrays

“…Distributed futures. Distributed futures [7] provide efficient data transfer when coupling data-parallel kernels in a task-parallel way. Spawning a task creates a distributed entity that works as a future that can be partially resolved by each process of the data-parallel computation.…”