HP Sponsor Talk

Mice, Elephants, Turtles Dimes & Dollars HPC & Big Data Marc Hamilton Hyperscale Business Unit HP Enterprise Group March, 2013

347,000 kaggle Pandora Music tweets SolidFire Scribd. SmugMug iHandy Amazon Finance SuperCam salesforce.com 23,148 apps AppFog Travel Joyent Plex Systems Snapfish Urban downloaded DCC Google Xactly Facebook Parse Product SAP eBay Dragon Diction PingMe GoGrid LinkedIn Reference Configurator CCC HP CRM Hosting.com UPS Mobile Atlassian Tata Communications Bromium 400,710 ads Bills of Material buzzd Lifestyle Engineering Ariba NetReach Amazon Web Services SCM Splunk requests New technology Inventory Quickbooks Scanner Pro Sport NetDocuments Zoho LimeLight Quality Control ScaleXtreme EMC box.net Alterian Foursquare Pinterest Burroughs HCM Datapipe Qvidian Taleo Education Hitachi Cost Management OpenText Hyland every IBM Manufacturing Projects access methods MRM Workscape Sage NetSuite Mobile, Social, Unisys Order Entry Mainframe Client/Server The Internet Big Data & The Cloud 60 seconds

Cash Management CyberShift PaperHost DocuSign CloudSigma Bull Yandex Baidu NEC ERP Hootsuite Games HCM Xerox Yahoo! YouTube Time and Expense Serif nebula Fijitsu Microsoft HP ePrint Atlassian Fixed Assets Workbrain Costing Accounts OpSource SLI Systems Zynga Navigation 2000 lyrics played Receivable Avid Mixi Elemica iSchedule Workday cloudability Billing ADP VirtualEdge • Change how technology is Yandex Photo & Video on Tunewiki Payroll SCM Activity PLM Corel Khan Academy Twitter Heroku Zillabyte Yammer Training Management Adobe CyberShift Renren Entertainment Sales tracking & Yahoo SuccessFactors Viber consumed & value it can bring Time & Rostering Kinaxis Marketing Microsoft News Atlassian Answers.com Attendance SugarCRM 1,500 pings Service Saba BrainPOP RightScale Social Networking Commissions PPM Sonar6 Quadrem CYworld sent on PingMe Database Kenexa Sonar6 MobileFrame.com Claim Processing Business Jive Software Saba myHomework • Open up new business models Softscape NetSuite Tumblr. Qzone Data Warehousing Intacct Toggl Fring Amazon dotCloud Cornerstone onDemand Exact Online 34,597 people Cookie Doodle Xing Mozy FinancialForce.com New Relic Softscape MailChimp PingMe are using Zinio Utilities Zynga Ah! Fasion Girl • Remove current inhibitors & IntraLinks Volusion Associatedcontent BeyondCore SmugMug MobilieIron 208,333 minutes Fed Ex Mobile Productivity unleash power of innovation Rackspace Flickr Twitter Angry Birds played Paint.NET TripIt

2 © Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice. Anywhere, anytime demand increases opportunities Hyperscale workloads growing significantly faster than overall market

$45.0 Hyperscale to be 35% of x86 market $40.0

HPC, $6.4 $35.0 HPC, $5.2 27% Hosting, $2.4 Hosting, $3.3 35% Customer demand for content $30.0 SP, $2.8 Device proliferation SP, $4.2 $25.0 Competitive edge $20.0 Enterprise $15.0 Enterprise Driving the need for new and $28.0 $26.0 purpose built products! $10.0

$5.0

$0.0 2012 2015

• The elephant in the room

• Not all big data comes from mouse clicks

• Hadoop becoming ubiquitous - good or bad? • Is storing 3 full copies of data really efficient or just easy? • Are Hadoop management tools ready for the enterprise?

• As more and more data is stored in Users select widely HDFS, disruptive innovation becomes available platforms difficult – you can’t break the APIs to solve their • So how can you drive disruptive business problems innovation?

Vendors pursue Hadoop originally sustaining innovation ran on commonly of 2-socket x86 available 2-socket systems for Hadoop x86 systems

• SSDs usage with Hadoop growing • Disk protocols too slow for SSD/Flash • Better hardware is available (PCIeGen3, etc.) • Flash nearing end of life just in time for new NVRAM • Now its time to address the software bottlenecks • HDFS on top, but what’s inside?

© Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice. Dimes and Dollars Are HPC compute nodes getting too powerful for big data use?

• Relative performance of mobile processors increasing quicker than server processors • Typical HPC compute node today, 300 watts, $300/year power • Nvidia Tegra 5 mobile processor will be the size of a dime, run CUDA and OpenGL apps • 60 5-watt mobile processors = $300/year power

100.0 73.0 Good ARM /Atom 26.8 29.3 performance ratios 22.2 23.1 25.0 25.4 18.2 15.7 17.2 11.9 10.0 10.5 11.2

4.9

1.7 1.5

1.0 Relative perforamance Relative perforamance (Higher better) is

0.1

29.92

12.00 10.00 Good ARM /Atom performance/ watt ratios

4.39 3.20 2.50

1.00 1.02

0.66 0.70

Relative performance/watt (higher is better) is (higher performance/watt Relative 0.43

12.16 10.00 8.93 5.56 4.55 4.63 5.08

4.44 3.14 3.20 3.44 2.63 2.98

1.25 1.00 1.15

0.10 Relative performance Relative performance (higher is better)

0.01

© Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice. Project Moonshot: A new era of extreme scale computing Redefining server infrastructure for rapid application optimization

Application-driven designs Project Moonshot Space Energy Costs

• 1,000s of servers per rack • Workload - tuned servers • Federated infrastructure scales seamlessly with additional servers

Device agnostic architecture scales from extreme low energy CPUs to high performance GPGPUs

15 © Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice. Supercomputers That Reuse Energy? DOE’s National Renewable Energy Lab purchases $10M HP Supercomputer - New ProLiant server platform will delivery 1 Pflop of Intel IvyBridge and Xeon Phi - http://www.nrel.gov/news/press/2012/1985.html - Power and warm water cooling system co-designed to work with NREL’s new $100M Energy Systems Integration Facility (pictured below) - Waste heat used to heat office space & melt snow in winter

16 National Renewable Energy Lab 200 TF Linpack – Feb 27, 2013

ProLiant SL Core

Customer specific Accelerators

Big Data SL270 SL250 SL230 High-efficiency computing SL2500