Problem solving environments are an attractive approach to the integration of calculation and management tools for various applications. Scientific and engineering applications are becoming increasingly demanding in terms of both problem size and problem complexity and therefore benefit from the integrated problem solving approach. However such applications often require high performance computing components in order to be computationally feasible. It is therefore a challenge to construct

more »

... ation technology, suitable for problem solving environments, that allows both flexibility as well as the embedding of parallel and high performance computing systems. Our DISCWorld system is designed to meet these needs and provides a Java-based middleware to integrate distributed computing component applications across wide-area networks. Our system supports both long running simulations as well as parameter-search style applications where many shorter "runs" are used to explore the parameter space of a scientific or engineering model. Key features of our design are the ability to access remotely stored data; compose complex processing requests either graphically or through a scripting language; and execute components on heterogeneous and remote platforms. Operators in task graphs can be slow (but portable) "pure Java" implementations or wrappers to fast (platform specific) supercomputer implementations. In addition, some operators may represent compound services that can potentially be expanded into a sub-graph of simpler operators, allowing the simulation to be reconfigured to run across multiple compute servers if this will lead to a faster execution time. Our system also supports efficient access to distributed and hierarchical data stores. User's processing requests can also be optimised by reusing data that has been created by previous requests. We describe our prototype and use it as a vehicle to discuss key issues for problem solving environments in science and engineering.