HLRS steps forward in Metacomputing

Based on European, Japanese and US high speed networks several applications from computational fluid dynamics, theoretical physics and molecular dynamics will exploit the capacity of this metacomputer.

Already during SC'97 and SC'98 HLRS and PSC have demonstrated their capability of linking supercomputers together to form a single metacomputer. Following these first steps the US-European cooperation has now been extended to include ETL, one of the leading Japanese Supercomputing Centers. In addition Manchester Computing Centre has joined this alliance.

The communication among the machines is realized by PACX-MPI. PACX-MPI was developed by HLRS. It is currently used in a number of metacomputing projects. Since January 1999 PACX-MPI is part of the European project METODIS that develops a bundle of tools to support metacomputing. Based on PACX-MPI users can run their code without modification on any cluster of heterogeneous machines and still benefit from the superior performance of highly tuned native message passing libraries. This helps to migrate from a single parallel machine to clusters of machines without having to rework the code. Besides being used in a number of Gigabit testbed projects in Germany, it is currently tested by European aerospace industry for usage in their distributed computing environments.

COVISE, the collaborative visualization and simulation environment developed by HLRS will visualize the metacomputing results on a virtual reality projection screen. Direct interaction will allow to analyze in an intuitive way the results of the metacomputing simulation. The transatlantic connection will be additionally used to discuss the results collaboratively with partners in Europe. In a cooperation with Sandia National Labs that also uses COVISE, the VR based analysis of a blood flow simulation in an anomaly of an artery will be demonstrated Initially developed in European Community funded projects, COVISE is in the meantime an established VR-post-processing tool at many European HPC centers. It is also commercially available via a spin-off company of the University of Stuttgart.

The network connection set up during Supercomputing '99 is a good example for international co-operations in the field. The network connections will include a number of international partners. From Manchester to Stuttgart the European Research Network TEN-155 and the National Research Networks DFN in Germany and UKERNA/JANET in the UK provide the connection. From Manchester to Stuttgart the European Research Network TEN-155 and the German Research Network (DFN) provide the connection. DFN also connects Germany to the US research network Abilene. Going through vBNS the connection reaches PSC and STAR-TAP in Chicago from where the Japanese-US research network TransPAC connects to ETL at Tsukuba.

HLRS presents an overview of its ongoing projects at the SC'99. The focus this year is on flow phenomena. In a joint project with the Universitätsklinik Ulm scientists at Stuttgart are working on the flow simulation of blood in the artery bifurcation. Aneurysms in that bifurcation may be lethal. In order to fix the problem surgeons at the Universitätsklinik implement stents in the artery. This may change the flow pattern in the bifurcation substantially. In order to optimize the flow in the artery surgeons would therefore like to predict the flow pattern introduced by their work.

In a second project flow around the crew rescue vehicle X38 of the International Space Station during the reentry into the earth's atmosphere is simulated. The flow solver URANUS from the Institute for Space Systems (IRS) has been parallelised in a joint project of IRS and HLRS. This allows to simulate bigger configurations and to increase the level of resolution at which physical phenomena can be made visible. Special emphasis was given to adapting the code for metacomputing.

A third projects is focussing on flow phenomena occurring in the molding industry. The European funded DECAST (enhanced DEsign Environment for Industrial CASTing Processes on Parallel Computing Platforms) project was set up to parallelise CFD codes and port them to clusters of PCs. Such clusters have been identified as potentially well performing but cheap parallel platforms that may help to foster the use of parallel computing in small and medium sized enterprises.

[News on Advanced IT]   [Calendar]   [Analysis]   [IT in Medicine]