GPU -Cluster JUDGE-
Jülich makes a crutial step towards Exascale computer with the introduction of hybrid clusters using Tesla 20-series GPU's
The use of graphics processing units (GPU's) can dramatically increase the performance of supercomputing clusters for select problem types, without corresponding increases in power consumption and capital expenditure. GPU's are now seen as important components in supercomputers of the future. "Only with these technological advances can the peak performance of supercomputers, currently set at a few Petaflops, be increased to Exascale (1000 fold increase) by the end of this decade", says Prof. Thomas Lippert, director of the Jülich Supercomputing Centre.
ParTec Cluster Competence Center GmbH, played, and continues to play, a vital role in the commissioning and day-to-day operation of the JUDGE cluster. Key parts of the installation were carried out using ParTec's ClusterTools suite which provided installation configuration management, update management and batch system configuration. These tools enabled system administrators to respond to user requests for new software packages in a quick, efficient and safe fashion.
The ParaStationMPI, already proven in challenging environments such as JuRoPA, was also demonstrated to work well in these hybrid cluster environments. By enabling users to pin processes to CPU cores, PSMPI provides a key control mechanism that enables optimal use of the GPUs attached to a given CPUs.
The monitoring facility GridMonitor reports irregularities of the system in a timely manner and enables administrators to take the appropriate steps to keep the system up and running and maintain the users productivity.
"We are proud that ParTec GmbH was chosen by the Forschungszentrum Ju?lich and IBM to contribute with ParaStation Software Tool Suite to the onsite and remote management of the GPU cluster JUDGE ," said Hugo R. Falter, Chief Operating Officer of ParTec Cluster Competence Center, GmbH.
ParTec Cluster Competence Center GmbH is specialized in the development and support of a comprehensive suite of cluster management tools and a runtime environment specifically tuned for the largest distributed memory supercomputers in existence today. ParaStation V5, an in-house developed, open-source cluster software stack provides a parallel environment for Linux clusters that provides our customers with the reliability, high availability and utilization levels they demand as they strive to meet their research goals whilst remaining within their power budgets. ParTec provides vendor-independent consultancy for a broad range of cluster technologies and forms long standing customer relationships buy offering comprehensive support services in both remote and on-site scenarios. Our approach ensures rapid deployment cycles and seamless interoperability of software components. ParTec is member of major European and worldwide research consortia (e.g. European File System Initiative EOFS, ExaCluster Laboratory ECL, PROSPECT, PRACE, D-GRID, Unicore etc.) to further contribute to the development of ExaFLOPS architectures of general purpose supercomputers towards the end of this decade. The headquarter of ParTec is located in Munich, Germany.