Department of Defense High Performance Computing Modernization Program

Department of Defense High Performance Computing Modernization Program Update Presented by Larry Davis, Deputy Director September 2009

HPC Forum Briefing – l. Davis Outline l Introduction l Program Structure 0 HPC Centers 0 Networking 0 Software Applications Support 0 Resource Management 5/4/2009 -2. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis HPC Modernization Program 5/4/2009 -3. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis HPC Modernization Program Goals 5/4/2009 -4. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis A Quick History l Program initiation — 1992– 1993 0 HPC Modernization Plan 0 Initial program structure established, 0 0 l S&T Initial HPC capabilities provided HPCMP established as an ACAT 1 l Major contract awards — 2001– 2003 l Operations focus — 2003– 2006 Program formalization — 1994– 1995 0 Program office established 0 Do. D oversight process implemented 0 Program structure and customer base 0 l 0 Major acquisitions — 1995– 1996 0 Four major shared resource centers 0 Defense Research and Engineering 0 Network (DREN) Programming Environment and Training (PET) contract established Operations focus — 1997– 2000 0 Continuous upgrades at HPC centers 0 Selection of new distributed centers 0 5/4/2009 -5. 1 0 New DHPIs, Institutes and Portfolios 0 expanded, include T&E l 0 DREN contract awarded 0 HPC Centers contract awards 0 New PET contract awarded (DCs) Do. D Challenge Projects established l selected Capability Applications Project (CAPs) established Minority Undergraduate Education and Research Initiative established Value to the Do. D study–Return on Investment (ROI) project Major contracts & expansion — 2007– 2009 0 Next Generation Technical Services 0 0 (NGTS) contract awarded New DHPIs, Institutes and Portfolios selected Computational Research and Engineering Acquisition Tools and Environments (CREATE) program initiated New PET 3 contract awarded New archival storage contract awarded Solving the hard problems. . .

HPC Forum Briefing – l. Davis Do. D HPC Modernization Program Army HPCMP Participation ARL & ERDC DSRCs 1, 309 Users/22 Organizations/102 Projects 56 DREN Sites 14 Challenge Projects/2 DHPIs 5 Institutes Navy HPCMP Participation NAVY DSRC 1, 031 Users/16 Organizations/198 Projects 38 DREN Sites 12 Challenge Projects/3 DHPIs 1 Institute Air Force HPCMP Participation AFRL & MHPCC DSRCs 1, 361 Users/24 Organizations/175 Projects 24 DREN Sites 12 Challenge Projects/5 DHPIs 3 Institutes Defense Agencies Participation DARPA, DTRA, JFCOM, MDA, PA&E & OTE 644 Users/4 Organizations/21 Projects 35 DREN Sites 3 Challenge Projects Other ARSC ADC 68 DREN Sites 1 DHPI 6/492009 -6. 3 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Do. D Supercomputing Resource Centers (DSRCs) Six Large HPC Centers 5/7/2009 -7. 2 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Do. D Supercomputing Resource Centers (DSRCs) FY 09 Capability Location System AFRL HP Opteron Cluster SGI Altix 4700 ARL Linux Networx Cluster (Classified) Cray XT 5 (Classified) SGI Altix ICE 8200 (Classified) ARSC SUN x 4600 Cray XT 5 ERDC Cray XT 3 Cray XT 4 SGI Altix ICE 8200 MHPCC Dell Power. Edge 1955 NAVY IBM 1600 Power 5 Cluster (Classified) Cray XT 5 IBM Power 6 Processors Memory Capacity (Cores) (GB) (Habus) 2, 048 9, 216 Total As of: July 2009 7/13/2009 -8. 4 FY 09 HPC Systems shown in GREEN FY 08 HPC Systems shown in RED FY 07 HPC Systems shown in BLUE Older HPC Systems shown in BLACK Solving the hard problems. . . 4, 096 3, 584 20, 480 20 18 84 4, 528 3, 464 10, 540 11, 384 7, 088 8, 976 6, 736 40, 128 34, 152 21, 264 55 26 138 254 161 2, 320 3, 504 9, 280 14, 784 18 55 8, 320 8, 760 16, 160 16, 640 17, 696 48, 480 95 100 360 5, 120 10, 240 61 3, 072 1, 920 12, 872 5, 312 6, 144 3, 840 26, 592 8, 448 33 17 202 125 117, 676 301, 560 1, 822

HPC Forum Briefing – l. Davis TI-09 Installation Status l Rackable’s recent purchase of SGI’s assets included all HPCMP contracts – TI-09 systems on track for delivery as scheduled l New SGI remains committed to delivering the new system and maintaining existing system l ARL – Classified SGI Altix ICE 8200 0 6, 656 core (169 HABUs) 0 Final acceptance September 18, 2009 l ARL – Unclassified SGI Altix ICE 8200 0 10, 752 core (271 HABUs) 0 Final acceptance late September 2009 l ERDC – Unclassified SGI Altix ICE 8200 0 15, 360 core (385 HABUs) 0 Final acceptance October 2, 2009 8/3/2009 -9. 3 Solving the hard problems. . .

HPC Forum Briefing – l. Davis HPC Installations (TI-01 through 09) Added HPC Capability Trend (DSRCs Only) 1, 400 Habus Procured for Each TI-XX Percent Procured from Pevious Year 2. 1 1, 000 + 30% Total Number of Habus 1, 200 800 1. 5 600 1. 7 400 1. 8 2. 2 200 0 TI-01 8/3/2009 -10. 2 4. 6 1. 1 2. 1 TI-02 TI-03 TI-04 1. 6 TI-05 TI-06 Solving the hard problems. . . TI-07 TI-08 TI-09

HPC Forum Briefing – l. Davis Performance Increases Come from Both Processor Improvement and Processor Count 18 Increase in Habus per processor for each TI-XX Increase in Number of processors for each TI-XX 16 Factor Increase 14 12 10 8 6 4 2 0 TI-01 TI-02 TI-03 TI-04 TI-05 TI-06 TI-07 TI-08 TI-09 Implication: Need Software Applications that Scale to 1, 000 s of Processors 8/3/2009 -11. 2 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Next Generation Technical Services (NGTS) Contract Support l HPC Center’s Technical Services Support contract provides 30– 50 FTEs per Center 0 Technical Expertise Customer Support System and Network Administration Application/System Performance Analysis Expansion Analysis and Integration Services Operators and Operations Support Outreach 0 Acquisition Vehicle for Small Purchases (ODCs) 5/4/2009 -12. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis NGTS (Continued) l Attributes of Contract 0 One five year contract for four DSRCs Facilitates leveraging innovations and efficiencies across multiple DSRCs Does not include ARSC and MHPCC with the exception of support for the Consolidated Customer Assistance Center 0 Cost plus type contract with a fixed award fee pool and no base fee 0 Awarded to Lockheed Martin Information Systems (LMIS) in March 2008 0$344 M contract ceiling 5/4/2009 -13. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis HPCMP Storage Initiative l Computing power grows annually—so do stored files l Archived data is hard for users to use and manage l Costs: User time, labor, hardware, software and media l Storage Initiative 0 Objective: Refresh to manage data for next 10 years 0 Goals: 10 -year architecture Leverage advances in technology Improve user productivity Improve reliability & adaptability Sustain within current storage budget 0 Funding: Reserved 1/3 of FY 09 investment dollars 0 Challenge: User awareness and behavior changes 5/4/2009 -14. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis HPCMP Data Storage Growth Single Copy Data Storage 8, 000 Single Copy of HPCMP Storage in Terabytes 7, 000 6, 000 5, 000 4, 000 3, 000 2, 000 1, 000 0 2001 2002 2003 2004 2005 Year 5/4/2009 -15. 1 Solving the hard problems. . . 2006 2007 2008

HPC Forum Briefing – l. Davis Storage Lifecycle Management (SLM) l Three Key Features 0 Utility Server to provide longer term near-line storage to store data for up to 30 days prior to archiving 0 Information Lifecycle Management (ILM) system to help user and system to associate important information with data 0 Establishment of a 10 -year contract to facilitate predictable costs (possibly fixed) for data management software l Acquisition 0 Software ILM software, archive software and development services Full and open competition RFP release in March 2009, award made in August 2009 0 Hardware Execute via NGTS 5/4/2009 -16. 1 Solving the hard problems. . .

HPCMP Storage Initiative Overall Acquisition Strategy HPC Forum Briefing – l. Davis 10 years Long-term Government Partnership with Industry Hardware Refresh Cycle 5/4/2009 -17. 1 Storage Software Storage Hardware Generation 1 Storage Hardware Generation 2 Solving the hard problems. . . Storage Hardware Generation 3 Storage Hardware Generation 4

HPC Forum Briefing – l. Davis Storage Target Architecture Remote Disaster Recovery Facility Data Sharing with Other Centers HPC SYSTEM A HPC File System A SLM ILMmanaged File System HPC SYSTEM B HPC File System B DR Cache Archive Server Center Archive Cache Utility Server Medium-Term Storage Short-Term Storage 5/4/2009 -18. 1 Solving the hard problems. . . Tape ILM-Driven Long-Term Storage

HPC Forum Briefing – l. Davis User Interface Toolkit – API UIT Making High Performance Computing Easy Single Desktop Interface to Multiple HPC Systems UIT API 7/9/2009 -19. 2 • Supports Novice to Expert Users • Central Access to HPC Resources • Custom Productivity Clients • Complete Job Stream Management • Fast Large File Transfers • Secure Authentication Solving the hard problems. . .

Defense Research & Engineering Network (DREN) (pronounced: 'dē-ren) 5/11/2009 -20. 2 Solving the hard problems. . . HPC Forum Briefing – l. Davis

DREN – Two Worlds HPCMP - Buy-In Customers HPC Forum Briefing – l. Davis 111 Subscriber sites 68 HPCMP sites 6/9/2009 -21. 2 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Software Applications Support 5/4/2009 -22. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis HPC Software Applications Institutes 8/27/2009 -23. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis HPC Software Applications Institutes 8/27/2009 -24. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis User Productivity Enhancement, Technology Transfer and Training (PET 3) 5/4/2009 -25. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis PET 3 Contract Award Update l PET 3 was awarded to High Performance Technologies Inc. (HPTi) on July 31, 2009 to start September 1, 2009 l Award amount ~$147 million over ten years l The HPTi team is made up of major HPC groups around the country, including Texas Advanced Computing Center, Pittsburgh Supercomputing Center, San Diego Supercomputer Center, and 18 other partners l One major feature is the combination of Computational Environment (CE) and Enabling Technologies (ET) into Advanced Computational Environments (ACE) 8/27/2009 -26. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Computational Research and Engineering Acquisition Tools and Environments 5/11/2009 -27. 2 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Resource Management 5/4/2009 -28. 1 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Customer Needs, Priorities, and Usage History Drive HPCMP Investments 30, 000 Capacity Habu-yrs 25, 000 Aggregated Requirements Max Projected Requirements Historical Projection of Requirements 20, 000 15, 000 10, 000 5, 000 0 2007 2008 2009 2010 2011 Fiscal Year 2012 2013 2014 l Detailed requirements information collected annually through web-based requirements survey (includes all projects) l Requirements information refined through in-person interviews and quality checks l Requirements validated by senior S&T and T&E executives in each Service and Agency 7/9/2009 -29. 6 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Do. D Challenge Projects 7/24/2009 -30. 2 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Dedicated HPC Project Investments 8/27/2009 -31. 3 Solving the hard problems. . .

HPC Forum Briefing – l. Davis Summary l World-class corporate computing capability established for Do. D HPC community l High Performance Computing capabilities being employed to provide substantial contributions to Do. D mission capabilities l Successful transition to scalable, parallel computing l Leveraging national, academic and federal activities 7/24/2009 -32. 1 Solving the hard problems. . .