Distributed Functional Evaluation: Dynamic Load Balancing on a Beowulf Cluster

摘要

The absence of side-effects and the abstraction of functional languages can be used to build programming patterns that can be easily adapted to a multicomputer. That is the motivation for DFL, a distributed functional language designed for exploring the concurrent evaluation of functional programs on a cheap Beowulf cluster, while preserving the main properties of these languages. In the paper, a brief description of DFL is presented as well as some comments and measurements about the dynamic scheduler based on a work-stealing algorithm that balances the workload. The decision of which tasks should be carried out remotely is taken by the runtime system using annotations or programmer bints called futures, which do not change the actual semantics of programs but their runtime behaviour.