Representing Eager Evaluation in a Demand Driven Model of Streams on Cloud Infrastructure

Abstract

Previously, we developed our StreamComponents framework which uses distributed components and web services to facilitate control, reconfiguration and deployment of streams on both local clusters, and remote cloud infrastructure. Our stream evaluation semantics are fundamentally demand driven, a conservative view that ensures no unnecessary computation, supports flexible structures such as cyclic networks and infinite streams, and facilitates resource management. Abstract In this paper, we focus on the evaluation semantics of our stream model, and explore circumstances under which more eager evaluation is desirable, whilst retaining the fundamental semantics. We introduce the Indirected Asynchronous Method pattern (IAM), which makes novel use of futures and auto-continuations, to facilitate fully asynchronous demand propagation leading to more eager evaluation of the streams. We present an evaluation of the model on both cluster and cloud infrastructure showing that very useful amounts of pipelining parallelism can be achieved.