Management von verteilten ingenieurwissenschaftlichen Anwendungen in heterogenen Grid-Umgebungen

Deploying Grid technologies by distributing an application over several machines is widely emerging for scientific simulations, which have large requirements for computational resources. While in the last couple of years distributed simulations were often executed on a pure demonstration level, currently ongoing work is focusing on taking the steps necessary to use the technology in an every day production environment. One of the major problems end-users are facing is the diversity in the available Gridsoftware environments. This work presents the concept and implementation of a Grid Configuration Manager (GCM), a tool developed to hide some of the complexity of Grid environments from the end-user. The central objective of GCM is to ease the handling of computational Grid environments for end-users and application developers, by abstracting the steps necessary for submitting computational jobs and implementing them for various Grid software environments. Up to now, GCM supports Globus, UNICORE and ssh based environments. The GCM assists users during three phases of the execution of an application: defintion of a Grid configuration, job submission and monitoring. Furthermore, GCM supports the generation and management of distributed multi-site jobs on the basis of the communication library PACX-MPI, by generating the according configuration files and keeping them consistent on all machines participating in the current simulation. The Grid Configuration Manager integrates a performance based resource brokering mechanism. The basic idea is, that given a pool of machines, GCM can decide based on the prediction of the application performance on various machines or combination of machines, which one to use. The performance prediction step is based on Dimemas. Dimemas can estimate the performance of an application based on an initial tracefile for varying network and machine parameters like network latency or bandwidth. The Grid Configuration Manager was tested and deployed in various scenarios. It has been shown, that the handling of scientific, computational jobs in heterogeneous Grid environments was significant easier and the determination of the execution environment was facilitated for users by utilization of the GCM.