Parallel Computing with Remote Procedure Calls
スポンサーリンク
概要
- 論文の詳細を見る
In this talk we give a survey on the most important of our recent activies in distributed computing. The talk will concentrate on the design, implementation and application of Sciddle and the Remote Computation System. Sciddle is a portable communication environment for distributing computations over heterogenous networks of UNIX computers. It is based on the client-server model. It was designed to support the distribution of large scale numerical computations and to keep its usage as simple as possible. All interprocess communication is done via remote procedure calls. Explicit message passing is not required. The user defines the interface between communicating processes in a simple declarative language. Parallel programming is supported by what we call asynchronous RPCs. For performance reasons, the newest version of Sciddle is based on the Parallel Virtual Machine (PVM) message passing library. We discuss the usefulness of the Sciddle system by means of two important applications from Computational Chemistry on internet-connected workstations and supercomputers. The goal of the Remote Computation System (RCS) is to provide easy access to modern parallel algorithms on (super)computers for the inexperienced user. RCS has an easy-to-use mechanism for using computational resources remotely. The computational resources available are used as efficiently as possible. The user's view of the RCS is that of an ordinary software library. However, in order to minimize the response time, a RCS service need not be provided by the workstation from where it has been requested but is dynamically allocated on an arbitrary machine in a given pool of computers. Furthermore, the RCS routines are generic in the sense that the method with which a problem is solved is not predetermined. The system tries to solve the problem by the best possible algorithm according to the input arguments. We will discuss the performance of a prototypical implementation of the RCS.
- 一般社団法人情報処理学会の論文
- 1997-03-06