IBM Case Study The Paris Astrophysics Institute launches new satellite research with a high-performance solution from IBM Overview ■ Challenge To ﬁt faster, more advanced and upgradeable computing and storage equipment in a small space in order to support new satellite research ■ Solution Implementing an IBM System x iDataPlex server cluster to deliver over 10 teraﬂops of computing power, and an IBM System Storage DCS9900 with IBM General Parallel File System (GPFS) technology to manage massive amounts of data The Paris Astrophysics Institute (IAP) is “ The IBM iDataPlex ■ Beneﬁts a laboratory of France’s National Center Processes 500 billion pieces of system has for Scientiﬁc Research (CNRS), a sci- raw data, provides over 128 TB ence and technology agency of the outperformed what we of data storage capacity in a French government. Operating at the were promised.” small data center footprint, and intersection of two ﬁelds of research, — Phillippe Parnaudeau, Manager consumes 40 percent less the institute brings together astrophysi- for Physical Infrastructure, energy than traditional systems cists from the Universe Sciences and Paris Astrophysics Institute Physical Sciences divisions of CNRS. It has a staff of 160 scientists, engineers, technicians, administrators and gradu- ate students.
The institute recently set about choosing a high-performance server to handle theformidable tasks of gathering, storing and analyzing 500 billion pieces of raw datacollected by Planck—a European Space Agency satellite launched in 2009 tosearch for the origins of the universe—and Planck’s High Frequency Instrument(HFI). The only problem was that there was no data center space available tohouse the new equipment.Deploying maximum computing power in a minimal amount of spaceIn order to accommodate the computing requirements of the Planck project, IAPwas tasked with ﬁnding a solution that would increase computing power whiledecreasing the amount of space required to house it. “These technical constraintswere spelled out in the speciﬁcations,” remarks Phillippe Parnaudeau, Manager forPhysical Infrastructure at IAP. “Our computer room already contains other serversand cannot be expanded.”But size was just one part of the challenge. “We also wanted the new system tohave a thermal protective coating and use no more than 50 kW of power,” saysParnaudeau. To address both space constraints and power concerns, the Instituteturned to IBM and IBM Business Partner Serviware, a major player in HighPerformance Computing (HPC) cluster integration. IBM and Serviware came upwith a new, high-performance solution tailored to the computer engineers’ scientiﬁcand physical needs.Drawing on the strengths of IBM System x iDataPlexThe IBM System x® iDataPlex™ solution was an ideal choice for the Planck proj-ect. The system takes up an area only 7.2 meters long and 1.2 meters wide, butcan process billions of pieces of raw data. The solution delivers high computationperformance and ease of management, while saving on space, energy consump-tion, and cooling costs.The IBM System Storage™ DCS9900 disk storage system is speciﬁcally designedto address the dense storage needs of high-performance computing environments.The combination of a System x iDataPlex cluster and System Storage DCS9900disk storage system meets the Institute’s demanding performance criteria for pro-cessing capability, volumetrics and input/output speed—while at the same timeaccommodating the computer room’s space and energy consumption constraints.Realizing the beneﬁts of high processing power and storage capacityThe 140-node System x iDataPlex cluster runs the Linux® operating system andboasts 3 GHz Intel® Xeon® quad core processors delivering over 10 teraﬂops ofpower for 9 teraﬂops on demand. The DCS9900 disk storage server, which is
connected to 140 servers, has 160 disks, for a useful capacity of 128 TB—ideal for Solution Componentsheavy computation applications. A network ﬁle system (NFS) storage unit roundsout the conﬁguration by providing an additional 220 GB of disk capacity. ● IBM System x® iDataPlex™ clusterIBM GPFS™ technology helps provide swift, reliable access to data. ● IBM System x3650 ● IBM System x3850 M2Three IBM System x3650 servers running Linux were deployed to administrate ● IBM System Storage™ DCS9900the server nodes, while one IBM System x3850 M2 server is used as a largememory server. The interconnection InﬁniBand network delivers a bandwidth of2.5 GB per second and latency below 4 microseconds.The conﬁguration combines high processing power with a reduced footprint andelectrical consumption 40 percent below that of traditional systems, or 50 kW max-imum. Parnaudeau also appreciates how quiet the systems are, describing it as“another plus that improves the working environment for the systems engineer.”The solution enables two levels of storage hardware, providing a parallel architec-ture system for a read-write throughput of 2 to 3 GB and another less rapid systemfor storing intermediate results.“The IBM iDataPlex system has outperformed what we were promised,” saysParnaudeau. “IBM kept its word and Serviware met our expectations for servicequality.”A seamless solution for processing and storing massive amounts of dataThe IBM solution ﬁts into a modest-sized area while still delivering very high pro-cessing capabilities. Users and IT engineers worked closely together to set it upand ﬁne-tune it so that it would be up and running on the day Planck was putinto orbit.Now, the Planck project’s chief scientists, Jean-Loup Puget, Jean-Michel Lamarre,and Francois R. Bouchet, and their teams are using this ultra-high-performing solu-tion to measure with unsurpassed accuracy the fossil radiation that bathes the uni-verse. Aided by France’s HFI scientiﬁc instrument, which was developed by CNRSand the French National Space Center (CNES) and was loaded onboard the satel-lite, they will draw maps of the entire sky with 50 million pixels each. The maps willprovide long-awaited clues to the mysteries of the universe. Initial project data isexpected to be published in 2013.