|
Pleiades ( or ) is a petascale supercomputer housed at the NASA Advanced Supercomputing (NAS) facility at NASA Ames Research Center located at Moffett Field near Mountain View, California. It is maintained by NASA and partners Silicon Graphics (SGI) and Intel. As of June 2015 it is still ranked the world's eleventh fastest computer on the TOP500 list with a LINPACK rating of 4.09 petaflops (4.09 quadrillion floating point operations per second)〔(【引用サイトリンク】title=TOP500 June 2015 List )〕 and a peak performance of 5.35 petaflops following a large hardware upgrade.〔 The ultimate goal of the partnership between NASA and SGI is to attain a peak supercomputing performance of 10 petaflops, or 10 quadrillion floating point operations per second, in order to "increase the computational capabilities for research, () modeling and simulation work at the NASA Advanced Supercomputing (NAS) facility."〔(【引用サイトリンク】title=NASA Selects Next-Generation SGI ICE HPC Platform for Technology Upgrade )〕 ==History== Built in 2008 and named for the Pleiades open star cluster, the supercomputer debuted as the third most powerful supercomputer in the world at 487 teraflops.〔(【引用サイトリンク】title=TOP500 Pleiades 8200 System Ranking )〕 It originally contained 100 SGI Altix ICE 8200EX racks with 12,800 Intel Xeon quad-core E5472 Harpertown processors connected with more than 20 miles of InfiniBand double data rate (DDR) cabling.〔(【引用サイトリンク】title=NASA Supercomputer Ranks Among World’s Fastest )〕 With the addition of ten more racks of quad-core X5570 Nehalem processors in 2009, Pleiades ranked sixth on the November 2009 TOP500 with 14,080 processors running at 544 teraflops.〔(【引用サイトリンク】title=NASA Supercomputer Ranks Among World’s Fastest )〕 In January 2010, the scientists and engineers at NAS successfully completed a “live integration” of another ICE 8200 rack by connecting the new rack’s InfiniBand dual port fabric via 44 fibre cables while the supercomputer was still running a full workload, saving 2 million hours in productivity that would otherwise have been lost.〔(【引用サイトリンク】title=’Live’ Integration of Pleiades Rack Saves 2 Million Hours )〕 Another expansion in 2010 added 32 new SGI Altix ICE 8400 racks with Intel Xeon six-core X5670 Westmere processors, bringing up to 18,432 processors (81,920 cores in 144 racks) at a theoretical peak of 973 teraflops and a LINPACK rating of 773 teraflops.〔(【引用サイトリンク】title=NASA Supercomputer Doubles Capacity, Increases Efficiency )〕 NASA also put an emphasis on keeping Pleiades energy efficient, increasing the power efficiency with each expansion so that in 2010 it was three times more power-efficient than the original 2008 components, which were the most power-efficient at the time. The integration of the six-core Westmere nodes also required new quad data rate (QDR) and hybrid DDR/QDR InfiniBand cabling, making the world's largest InfiniBand interconnect network with more than 45 miles of cable.〔 After another 14 ICE 8400 racks containing Westmere processors were added in 2011, Pleiades ranked seventh on the TOP500 list in June of that year at a LINPACK rating of 1.09 petaflops, or 1.09 quadrillion floating point operations per second. InfiniBand DDR and QDR fiber cables are used to connect the all of nodes to each other, as well as to the mass storage systems at NAS and the hyperwall visualization system, creating a network made up of more than 65 miles of InfiniBand fabric, the largest of its kind in the world. Pleiades is built in a partial 11-D hypercube technology, where each node has eleven connections to eleven other nodes, with some making up to twelve connections to form a 12-D hypercube.〔(【引用サイトリンク】title=NAS/HECC User Knowledge Base Pleiades Interconnect )〕 In 2012, NASA and partners SGI and Intel began working on the integration of 24 new Altix ICE X racks with Intel Xeon eight-core E5-2760 Sandy Bridge processors to replace 27 of the original Alitx 8200 racks containing quad-core Harpertown processors. With a total of 126,720 processor cores and over 233 terabytes of RAM across 182 racks, the expansion increased Pleiades' available computing capacity 40 percent.〔(【引用サイトリンク】title=Pleiades Supercomputer Gets a Little More Oomph )〕 Each new Sandy Bridge node has four networking links using fourteen data rate (FDR) InfiniBand cable for a total transfer bandwidth of 56 gigabits (about 7 gigabytes) per second.〔(【引用サイトリンク】title=HECC Project Hardware Overview: Sandy Bridge Processors )〕 In early 2013, work began on a larger hardware refresh for Pleiades, ultimately removing all of the original 4-core Harpertown processors and adding 46 SGI ICE X racks with 10-core Intel Xeon E5-2680V2 (Ivy Bridge) processors. When installation was complete in August 2013, the system's overall peak performance increased 62% from 1.78 petaflops to 2.87 petaflops.〔(【引用サイトリンク】title=NASA's Pleiades Supercomputer Upgraded, Harpertown Nodes Repurposed )〕 The system was slowly upgraded again between January and April 2014, adding another 29 racks of Ivy Bridge nodes and increasing the system's theoretical computational capability to 3.59 petaflops.〔(【引用サイトリンク】title=Pleiades Upgrade Completed )〕 To make room for the expansion, all of the system's remaining Nehalem nodes and 12 Westmere nodes were removed. In late 2014, more Westmere nodes were removed to make room for new Intel Xeon Haswell processors, increasing the theoretical processing power by one petaflop to 4.49 petaflops.〔(【引用サイトリンク】title=NASA's Pleiades Supercomputer Upgraded, Gets One Petaflops Boost )〕 In January 2015, additional Haswell nodes were installed and released to users, giving Pleiades a new peak theoretical processing capacity of 5.35 petaflops.〔(【引用サイトリンク】title=Pleiades Supercomputer Performance Leaps to 5.35 Petaflops with Latest Expansion )〕 抄文引用元・出典: フリー百科事典『 ウィキペディア（Wikipedia）』 ■ウィキペディアで「Pleiades (supercomputer)」の詳細全文を読む スポンサード リンク
|