Programming the grid: component systems for distributed applications

Abstract

Summary form only given. The traditional model of software design for large scale scientific problem solving is outdated. The emphasis is now on large teams that must build simulation software that integrates physical systems from multiple scientific disciplines. In addition to the problem of multi-disciplinary physics, the computational environment is now a grid of distributed resources consisting of large supercomputers, databases and networked instruments connected by high speed networks. Consequently, a new model for programming these applications is required. What is emerging is a programming style where the individual components of a large simulation can be located on remote systems and the application is built by composing these elements together to form a single distributed system. Building applications from software components is not a new idea. It has a rich history in the visualization and object oriented software design communities. More recently, the desktop top software industry has embraced this concept with systems like Java Beans, ActiveX and CORBA-Enterprise Beans. The author examines a new project to design a component architecture for scientific programming. He examines some of the design trade-offs that make this problem different from the conventional CA environment.