Abstract

In this paper we explore the potential of functional languages in designing and implementing complex distributed systems. We focus on the areas of parallel computation and coordination, and describe the prototype implementation of a generic runtime environment (RTE) for parallel execution. In particular, we discuss the implementation of scheduling algorithms in this context. Key features of our design are modularity, a separation between basic components, and employing a hierarchy with increasing levels of abstractions. The microkernel of this system is accessed via a narrow interface and most of the coordination of the system is realised in a functional language, (Concurrent) Haskell [5], at system level. Thus, we obtain a prototype in the form of an executable specification, amenable to formal reasoning, and easier to maintain and more flexible than conventional RTEs. In particular, components can be easily replaced, and we demonstrate the flexibility of this approach by presenting different variants of the scheduling component for the parallel execution of Haskell programs, put on top of GHC [4]. In the full paper we add sophisticated work distribution policies. With our approach we target complex architectures such as Computational Grids [1]. To handle such heterogeneous, global architectures with potentially thousands of machines a modular design with easily replacable components is crucial. More generally, this work underlines the usability of a high-level language such as Concurrent Haskell as a systems programming language [2].

Citations