概要GPU（Graphics Processing Unit）は，3Dグラフィックスの表示を処理するための半導体チップのことで，これまで画像処理に関連した処理を行うことで，CPUの処理量を減らすために開発・利用されてきた．これに対してGPUの性能が非常に向上したため，CPUを主体に計算処理能力を上げるのではなく，GPUを計算の主体として利用する試みが見られるようになってきた．実際に浮動小数性能はCPUよりGPUの方が高く，パラレルデータ処理能力にも秀でていることからHPCへの適用性は高い．また，一般的な言語によるプログラミングが行える環境によりGPUの役割がさらに広げられると考えられる．第46回の超並列計算研究会では，このGPUをとりあげ，HPCにおいてどの程度能力がありどのように活用かするについてご講演いただきました．また，Scali MPIやWindows コンソーシアムの現状報告などを行いました．
Hakon Ording Bugge is a founding member of Scali AS. Hakon received his MSc from the University of Oslo in 1985 and subsequently joined the research and development department at Norsk Data where he worked on the development of advanced CPU architecture. In 1989, Hakon joined Dolphin Server Technology where he was a central engineer in the development of the superscalar RISC CPU. Hakon was senior system architect responsible for the SCI development team at Dolphin Interconnect Solutions, which was formed in 1992. In 1994, Hakon joined Kongsberg Informasjonskontroll where he drove the development of the cluster technology, which today is the foundation of Scali's advanced solutions. Hakon has been published in several international journals and conference proceedings, and has worked as a part-time lecturer in computer architecture at the University Graduate Center (UniK), Kjeller, Norway.
Scali MPI Connect is the leading MPI for multicore technologies. In this talk, we will address how performance and throughput can be maximized in a cluster system, based on the unique feature set of Scali MPI Connect. Being the only MPI supporting policy-based, dynamic mapping of processes to cores, Scali MPI Connect provides a rich feature set which can be used to extract maximum performance for a single run or maximize the throughput of several jobs. We will show how this capability can be used to increase the throughput of LS-DYNA running on a cluster consisting of dual-socket, quad-core nodes.
Scali Manage is a comprehensive cluster life-cycle management system. Scali Manage provides a rich set of deployment methods, advanced reporting and monitoring, advanced alarms, and high-availability for mission critical services. Further, Scali Manage is vendor agnostics, supports a rich set of server/compute node models and supports both Novell SLES and Red Hat Enterprise Linux. Based on the Common Information Model, a DMTF standard, Scali Manage has taken many of the cornerstone technologies from datacenter management systems and optimized these for HPC and cluster administration