To ensure a dynamic, service-oriented environment to perform reliably and efficiently, grid site administrators need tools to monitor and to manage those services effectively. In this process, performance modeling and analysis play key roles. The research challenges include:
We investigate how the operational numerical weather forecast system HIRLAM could be made grid-enabled, meaning what kind of adaptions are required to execute the system efficiently on a grid. As grid platform we use the DAS-2 system, which consists of five clusters located at five different universities. Technical problems were encountered when more than one cluster was used.
In iterative compilation we search for the best program transformations by profiling many variants and selecting the one with the shortest execution time. Since this approach is extremely time consuming one has to incorporate static models. We show that a highly accurate model as a filter to profiling can reduce the number of executions by 50%. We also show that using a simple model to rank transformations and profiling only those with highest ranking can reduce the number of executions even further, in case we have a limited number of profiles at our disposal. We conclude that a production compiler might perform best using the last approach.
The application driver Ctadel is extended with an advanced numerical technique: semi-Lagrangian formulations. Furthermore, communication patterns are investigated for a model using semi-Lagrangian formulations on the parallel cluster DAS-2. With a large number of processes communication time becomes the limiting factor for performance. With a dynamic data-driven approach we are able to decrease the communication costs and to increase the performance of the model.
In this project a framework is developed to run simulations of Earth Observation instruments on Computational Grids. The framework allows data processing applications that are not Grid-aware to run on a Grid. It hides the technical details of the Grid for the user, provides easy to use interfaces, and contains a workflow system that can be used for chaining of data processing applications to perform end-to-end simulations. The framework has been demonstrated with data processing software for the OMI instrument.
The Leiden Institute of Advanced Computer Science (LIACS) pioneered Grid education at Dutch universities by successfully introducing a seminar in Grid Computing (2001/2002 and 2002/2003) for higher level computer science students. In the academic year 2004/2005 we will reopen the seminar with an excitingly new design and organization, in which the students will learn the latest grid technologies using the DAS-2 system.