Metacomputer architectures are able to offer high computation performance to parallel scientific applications with weak synchronisation. Current metacomputer-systems are using dedicated high speed networks for applications with fine-grain communication, or they are limited to loosely coupled parallel applications with coarse-grained communication in order to deal with architecture-dependent high communication latencies and varying communication throughputs. In this work an analysis of methods of existing metacomputer shows the weakness for commercial users. Their existing applications need cost-intensive adjustments, in order to be able to use the performances of complex metacomputers. High adjustment costs imply a dependency of the software to the metacomputing environment. This economic risk reduces the acceptance of metacomputers in the industry. As a result of this analysis, a lightweighted metacomputer environment in the programming language Java has been developed which supplies a solution to these problems. Therefore a new and efficient method of process distribution has been applied to common communication patterns. This method is based on an optimization of communication performance of parallel applications by a smart placement of processes. Together with the ability to react to the varying communication performance of the connecting network an application-transparent optimization, regarding runtime, has been realized. With this threadbased system and with the interfaces offered for external controlling, adjustments for existing applications are made very easy. The performance of the system has been verified with synthetic benchmarks. This concept has been evaluated within the industrial environment of the company Tecoplan AG and their product 'virtual workshop ', a application for the design and construction process. This application, based on a voxeltechnology, profits from the speedup of parallel execution and from the smart process distribution of the metacomputing system.     «

Metacomputer architectures are able to offer high computation performance to parallel scientific applications with weak synchronisation. Current metacomputer-systems are using dedicated high speed networks for applications with fine-grain communication, or they are limited to loosely coupled parallel applications with coarse-grained communication in order to deal with architecture-dependent high communication latencies and varying communication throughputs. In this work an analysis of methods of...     »