32
Wes Showfety Open Source Database & HPC strategist, North America [email protected] 770-617-7377 LinkedIn: https://www.linkedin.com/in/wes-showfety-2399444 Twitter: @Wes_Show IBM HPC & Kinetica

Kinetica + ibm + nvidia customer presentation_showfety-v1

Embed Size (px)

Citation preview

OpenPower in 2015

Wes ShowfetyOpen Source Database & HPC strategist, North [email protected]: https://www.linkedin.com/in/wes-showfety-2399444Twitter: @Wes_Show

IBM HPC & Kinetica

#1

Portfolio of HPC SolutionsDeployment tools, integrated managementCompilers: gcc, IBM XLC, LLVM OpenMP4, PGI Fortran/C/C++, Java, OpenACC, OpenMPDebuggers, Profilers, Math libraries, MPI & HPC appsProcessors & SystemsHPCSoftwareHigh PerformanceFile System & StorageHigh Performance Processors & SystemsAccelerator, networking, storage integration via NVLink & CAPIHighest memory throughputHighest Performance HPC Storage: Elastic Storage ServerHigh Performance Spectrum Scale (GPFS) Parallel File SystemData centric designHigh Speed InterconnectHigh speed interconnect / network fabric from Mellanox TechnologiesMPI acceleration in the IB fabric, reducing CPU overheadSupport for GPUDirect, NVMe over fabric

#Two Major New Product Announcements

Novel Embedded NVLink Interface to Tightly Integrate with NVIDIA Tesla GPUsNew ChipPOWER8 with NVLinkNew Power LC Linux Servers3 New Linux Servers Built for Big Data, Analytics, Machine Learning, and HPC

#System Details2-socket, 2UUp to 20 cores (2.86-3.26Ghz)1 TB Memory (32 DIMMs)230GB/sec memory bandwidth2x SFF (HDD/SSD), SATAUp to 4 integrated NVIDIA Pascal GPUs3 PCIe slots, 3 CAPI enabled, IB Add-inAir or water cooled New Power Linux Servers

System Details2-socket, 2UUp to 20 cores (2.9-3.3Ghz)512 GB Memory (16 DIMMs)115GB/sec memory bandwidth12 SFF/LFF (HDD/SSD) 96 TB storage 5 PCIe slots, 4 CAPI enabled2 NVIDIA K80 GPU capable

System Details2 socket, 1UUp to 20 cores (2.09-2.32Ghz)512 GB Memory (16 DIMMs)115 GB/sec memory bandwidth4 SFF/LFF (HDD/SSD), 32 TB Storage4 PCIe slots, 3 CAPI enabled1 NVIDIA K80 GPU capable

S822LC for High Performance ComputingS822LC for Big DataS821LC

New POWER8 with NVLink Processor

New Tesla P100 with NVLink

#

5

250+ OpenPOWER Foundation Members

Implementation, HPC & ResearchSoftwareSystem IntegrationI/O, Storage & AccelerationBoards & SystemsChips & SoCs

#

5

2300+ Linux Applications on POWERBig Data & Machine Learning

Cloud

Mobile Enterprise

Major Linux Distros

HPCminiDFTCTHBLASTBowtieBWAFASTAHMMERGATKSOAP3STAC-A2SHOCGraph500IlogCHARMMGROMACSNAMDAMBERRTMGAMESSWRFHYCOMHOMMELESMiniGhostAMG2013OpenFOAM

#

6

Emerging technologies drive business transformationCloud60% of banks process most transactions in cloud by 2016Collaboration60 million US households conducting P2P payments Big Data2.5 billion gigabytes of data generated every dayIntelligent/Connected Systems7.9 million in U.S. adopted NFC e-WalletsMobile 35% transaction growth driven through mobile annually through 2017AnalyticsMoving to real time

+7.6% in customer lifetime value for firms using engagement analytics

$226B annual cost of health care fraudSecurity$5.65 million average cost of a security breach in the US

#

And yet client expectations remain daunting 24x7x365applicationavailabilityUsers expect 4 Billion TransistorsFollow-on to Power8 CPUs18

Introducing 822LC Power System for HPCFirst Custom-Built GPU Accelerator Server with NVLink2.5x Faster CPU-GPU Data Communication via NVLinkNVLink80 GB/sGPUP8

GPU

GPUP8GPU

PCIe32 GB/sGPUx86

GPUGPUx86GPUNo NVLink between CPU & GPU for x86 Servers: PCIe Bottleneck

NVIDIA P100 Tesla GPUMinskyPOWER8 NVLink Serverx86 Servers with PCIeCustom-built GPU Accelerator ServerHigh-Speed NVLink Connections between CPUs & GPUs and among GPUsFeatures novel NVIDIA P100 Tesla GPU accelerator

#Detailed Diagram of 822LC for HPCP100GPUPOWER8CPUGPUMemorySystem Memory

P100GPU80 GB/s

GPUMemory

NVLink115 GB/sP100GPUPOWER8CPUGPUMemorySystem Memory

P100GPU80 GB/s

GPUMemory

NVLink115 GB/s

NVLink between CPUs and GPUs enables fast memory access to large data sets in system memoryNVLinkNVLink720GB/sUp to 0.5 TB16GB HBM2

#Kineticas CTO on IBM Power servers with NVlink vs. x86

#

Kinetica is already seeing a three- to four-times performance increases using Power chips with NVLink as well as significant gains in other areas.

Amit VijCEO & Co-FounderKinetica Accelerated Database Company

#Kinetica: 10x Faster Relational Database w/ AnalyticsRetail / Supply Chain Use Case ExampleFuse real-time data from multiple sourcesPoint-of-Sale (POS) dataDistribution centers inventoryHistorical buying patternsDemographicsWhats trending on TwitterWeather data Logistics Use Case ExampleDelivery route planningMonitor delivery / collectionContingency planning (traffic, accidents, employee sickness)Telcos: Analyze log information from OTA cell phone updates

Many more use cases for Finance, Defense, Healthcare, Ad-tech, Insurance, etc.

#Lets take GPUdb as an example. They have a relational database that is massively parallel and runs 10 to 100x faster than Oracle, by using the high performance of GPU accelerators. You can do SQL queries and also run analytics using GPUdb.The GPUdb database has uses across many domains. In retail, it enables real-time fusing of data from multiple sources, like xxx(listed above). This enables retailers to track their inventory in real-time and keep their stores stocked based on changing consumer demands. Likewise, use cases in logistics & telcos23

High Performance Data Analysis KineticaUltrafast ingest and analysis of billions of objects using GPUs

Advantage Performance, Cost, Scale

Logistics United States Postal Service was billed $100M by Oracle for Exadata and could only get 20% of their 220,000 mail carriers online with real-time Geospatial before it would crash. GPUdb does 100% at 95% less cost.

Retail Large retailer estimates $3B in lost sales last year due to empty shelves and lost purchase opportunities. They spend $100M on HANA, and could ingest 1B purchase records per hour. With Kinetica they did 4.5-6B records per minute.

#

Performance Leadership Roadmap for HPC / HPDA

201520162017POWER8POWER8 with NVLinkPOWER9OpenPowerCAPI InterfacePowerAccel Interfaces:NVLink, CAPI, PCIe Gen3PowerAccel:Enhanced CAPI, NVLink Next Gen, PCIe Gen4

Connect-IBFDR InfinibandPCIe Gen3 ConnectX-4EDR InfinibandCAPI over PCIe Gen3 ConnectX-5Next-Gen InfinibandEnhanced CAPI over PCIe Gen4MellanoxInterconnect Technology

IBM CPUsNVIDIA GPUsKeplerPCIe Gen3VoltaNVLink Next GenPascal &TeslaNVLinkIBM Nodes

#

25

After: NoSQL POWER8 + CAPI Flash

Load Balancer500GB Cache NodeBackup Nodes500GB Cache Node500GB Cache Node500GB Cache Node500GB Cache Server NodeBefore: NoSQL In-Memory (x86)

24U4UAccelerating NoSQL Databases with CAPI-Attached Flash

POWER8 ServerFlash Array w/ up to 40TBCAPI Device

Flash Acts As Extension of System Memory3x Lower Cost

#Today we are going to show you a revolutionary technology that we think is going to change the way services are consumed on the cloud.We all know the prevalence of NoSQLs and there seems to be flavors for every task at hand. For example, if you need large massive data storage there is Hadoop. If you need lightning quick response time there is in memory redis. However, there doesn't seem to be a way to combine the lightning quickness with massive capacity. Well, today we have a technology in out POWER 8 Systems that introduces a new tier of memory. This new tier lets our POWER 8 System perform like they are stacked with TBs of memory.

Cost and power savings will vary depending on the size of the Database and pricing variance but realistically up to 3.3X

From Technical Computing to Machine LearningHigh Performance ComputingAccelerated DatabasesMachine LearningResearch + Commercial HPC

Seismic ProcessingReservoir SimulationRisk AnalyticsOptions PricingScientific ResearchGenomics

10x Faster DatabasesKinetica, BlazeGraph

LogisticsRetailUtilitiesTelcosFinanceDefenseDeep Learning

Computer vision, Speech, NLP

Use cases: Retail, Customer Service, Text Analytics

#On Linux on Power, we started with solutions for HPC, in particular targeting both the research HPC markets and also the commercial HPC markets like finance & oil and gas.Recently, we are seeing HPC type of workloads and associated systems and storage starting to penetrate the enterprise via high-speed databases and machine learning. In particular, we are working with companies that are building GPU-accleratated databases like GPUdb, MapD, and BlazeGraph and companies doing deep learning like Minds.ai and Imagga.The solutions that these companies develop target a wide variety of enterprise markets, including retail, telcos, finance, etc27

HPC Pre-Sales Centers and Technical SupportPADC centers with IBM, NVIDIA and Mellanox focused on accelerated applications and technical collaborations IBM Systems Client CentersHPC BriefingsHPC WorkshopsHPC Benchmarks

UK Science and Technology Facilities Council (STFC) PADCIBM PADC Montpellier joint center with NVIDIA and MellanoxIBM PADC Boeblingen joint center with NVIDIAIBM Poughkeepsie POWER HPC Benchmark Center

For latest HPC information refer to the IBM Systems Client Centers HPC page IBM Austin POWER HPC Executive Briefing Center [email protected]! NVIDIA/IBM Acceleration Lab

#At the Hartree Centre we provide collaborative research, innovation and development services that accelerate the application of HPC, data science, analytics and cognitive techniques into UK industry. Backed by over 170M of funding from the Department for Business, Innovation and Skills and part of the Science and Technology Facilities Council, our mission is to enable businesses and researchers to use these tools to gain insight, value and competitive advantage.

28

IBM Research Paving the Path to Next-Generation HPC

Programming Models for Exascale

Enhancing Open InterconnectsPOWER

Data Centric System Node, & Processor Innovations

Scalable High Performance Storage & File Systems

#Power servers today are used for a wide range of applications all over the world. When we looked at the design of the hardware we had a couple of key workloads in mind. The first are the traditional systems of record. These are the large ERP systems (e.g. SAP), the electronic medical records (EMR) systems, and many of the OLTP workloads that run our clients businesses. The second key workload are Systems of Insight . Those are the workloads that include analytics, Big Data and Data Warehouse. Lastly, our designed were influenced by the needs of rapidly growing mobile workloads or Systems of Engagement. I want to spend a few minutes talking about what these workloads require to run successfully. 29

Integrated Clusters: Validated, Tested, CompleteThe Power Systems HPC Cluster is a modular solution combinesHigh performance compute nodesLow latency interconnect fabricHigh performance parallel storageSystem software In to a single, integrated solution; scale from 5 to 64 compute nodes

Assembled, tested, and provisioned in IBM manufacturing for faster time to compute

Modular and available with or without GPU and FPGA accelerators for compute intensive, data intensive, or balanced cluster performance

#HPC Cluster Definition Entry Model exampleHardwareServersComputing nodesLogin/Utility nodesManagement nodesCluster StorageESS GL2 or GL4 - optionalAttach stand alone ESS racksExisting customer storage NetworkData network: IB EDRManagement network: EthernetInfrastructure managementHMC and consoleESS Mgmt & NSD servers Cluster management sever(s)Power, Packaging, CoolingAir or water coolingMonitored PDUsCablesRacks

Compute nodesHMCKeyboard/monitorUFM applianceESS GL2Starter Rack Example

xCAT managerLogin node

SoftwareOSRHEL 7.2 LE NVFile systemSpectrum Scale (GPFS)Cluster provisioning & mgmtxCAT,Spectrum Cluster Foundation (PCM)Resource scheduler/mgmt Spectrum Computing LSFDevelopment & runtime toolsCompilers & LibrariesXLC/C++/FortranESSL, PESSLParallel EnvironmentPE RuntimeDeveloper EditionOther HPC ToolsGDB, PTP, othersCUDA

ESS mgmt. node

IB TOR switchEnet TOR switch

ServicesPre-Build / Pre-installationCluster Integration Planning Workshop (Lab Services)verify all aspects of integrating the cluster into the client's environment. Manufacturing plant & CSCProduct build and offering assemblySoftware preload, as specifiedIntegration checks/testsOn-site installation (IBM / Business Partner)Hardware setup (rack/stack/cable)

Cluster Network Integration & Verification (Lab Services)System software load verificationNetwork IntegrationCluster Checker / DebuggerSpectrum Scale /filesystem verification

|31

#

31

Questions?-Next Steps

#