NWO grid project. Contract No. NWO-RFBS 047.016.007

NWO-GRID - joined project in High Performance Simulation on the Grid.

Partners:

• Peter Sloot (University of Amsterdam, Section Computational Science)
• Alexander V. Bogdanov (Institute for High-Performance Computing and Information Systems. Technical University of St. Petersburg)
• Guenri E. Norman (IHED-IVTAN, Russian Academy of Sciences, Moscow)
• Victor Malyshkin (Institute of Computational Mathematics, Siberian Branch of Russian Academy of Sciences)

E-Science is maturing through technological push and application pull. The current developments in Grid-based distributed computing is mostly data driven, meant for High Throughput. We want to develop theory, methods and tools to support High Performance Simulations on the Grid. This is essential for the acceptance of Grid Technology in Computational Science and Engineering. Support for long running simulations with complex a-synchronous communications and user interaction needs to be developed.

First we will install a wide area Grid testbed between the major partners in the project. Here we will make use our previous expertise obtained through the Dutch-Grid, the CrossGrid and the DataGrid projects. Next we will develop (theory, methods and implementations) Grid-based check-point restart mechanisms for distributed fail-save interactive computing. Finally we will validate the research through experiments with a virtual reactor, simulating and visualizing 3D flow, reaction and discharge

The objectives of the project.

On a technological level the objective is to create a distributed scientific workbench (a Problem Solving Environment) to support computational science and engineering. This collaborative environment will make available in a transparent way, computational resources, advanced visualization and distributed access and data archiving. This environment should be accessible by all partners in a user-friendly way and support collaboration in advanced distributed computing.

The major part of the research will be devoted to the development of dataflow and control mechanisms in a Grid environment by redesigning High Level Architecture (the de-facto standard in distributed simulation: HLA) for the Grid.

In addition to that, high performance on the Grid will be obtained by adding support of checkpoint, migration and restart for PARALLEL tasks, like Condor does high-throughput for whole sequential Jobs. We will develop 'Dynamite-G' a system for advanced task (rather than Job)- migration running under Globus and OGSA: Dynamite-G will be a library supporting fault tolerance for parallel MPI programs.

Inspection of 3D datafields will be done using a Personal Space Station (PSS), a commodity hard/software Virtual Reality system linked to the Grid. In the Cross-grid project we have already developed a prototype of such as system for dedicated clusters, now we will extend that to wide-area Grid environments. For this the software developed for the Amsterdam DRIVE system will be integrated in the PSS

Computational Science is driven by application requirements, therefore we will study an extreme time- and resource demanding problem: Plasma Enhanced Chemical Vapour Deposition, where 3D complex flows, reactions and plasma discharges are coupled in one challenging interactive computational experiment. The development of the Grid based high performance simulation environment is driven by and validated through this Virtual Reactor. The objective here is to provide a user-friendly compute environment to do interactive studies on reactor parameters for multi-objective optimization through Pareto-front analyses.

The expected results

Since we have a solid existing joined basis to start with, we expect that already in the first year we will be able to publish in various computational science Journals such as IEEE's CiSE and Elseviers FGCS and PARCO, in addition to conference proceedings (the International Conference on Computational Science: ICCS, PPAM and Grid based conferences). The aim is to have enough material for 3 already experienced PhD students to graduate (each with at least 3-4 peer-reviewed papers in international journals in addition to proceeding papers).

The interest in our PSE middleware as well as the Dynamic Task-migration software will result in public domain professional software.

The research in the application fields addressed through the case studies will be published in the domain specific journals (e.g. Phys Rev E)

We expect that the collaborative VR system based on our DRIVE expertise and the PSS, will really boost data-exchange and collaborative data analyses over the Grid. In addition it can be of general use in distributed computational steering experiments.

Up to now there is no free and easy access to middleware for dynamically connecting cluster resources with user friendly interfaces to support high performance simulations. The system that we will build as our experimental environment, our scientific workbench so to say, will provide inexpensive high performance resources both for academia and even small (computational) business. By adhering to the de facto standards in Grid-technology and through the primitives provided, as well as the tools for collaborative simulation and visualization, we will open up possibilities for further scientific collaboration between Russia and Europe.

Finally, from previous initiatives we have learned that writing down the major results of such novel general purpose methods and technologies in a book, in addition to regular journal and proceeding papers, is a very elegant way to exploit and disseminate the results. Therefore we have started negotiation with Springer Verlag Lecture Notes on Computational Science series to publish in an integrated way our work on High Performance Simulation on the Grid.