Cloud-Computing Erkenntnissprünge mit fremden Rechnern Klaus Gottschalk Deep Computing Architect IBM Deutschland GmbH München, 29.04.2008
IBM Deep Blue 1996 und 1997 IBM POWER2 2
IBM Deep Computing Computing in der Formel 1 Klasse Durchbruch zu neuen Horizonten Berechnungsmodelle für Unberechenbares Simulation ergänzt heute Beobachtung, Experiment Visualisierung, Verwaltung und Bearbeitung von extremen Datenmengen Lösung von Business-Problemen Neue Lösungen über Business-Grenzen hinweg benötigen Hochleistungs-Rechnerarchitekturen Strategische Vorteile im Markt erlangen Mehr Schub durch IBMs Expertise Starke Lösungen für die herausfordernden Probleme unserer Kunden Wettbewerbsvorteile sichern durch produktionsreife HPC Produkte 3
Deep Computing Marktsegmente oder Hochleistungsrechnen ist überall Seismic Analysis Reservoir Analysis Derivative Analysis Actuarial Analysis Asset Liability Management Portfolio Risk Analysis Statistical Analysis Mechanical/ Electric Design Process Simulation Finite Element Analysis Failure Analysis Drug Discovery Protein Folding Medical Imaging Bandwidth Consumption Digital Rendering Gaming Collaborative Research Weather Analysis High Energy Physics Energie Finanz Engineering LifeScience Media Forschung 4
Beobachtungen Systems and Technology Group Anschaffungskosten der HPC Systeme treten in den Hintergrund Die Betriebskosten für Strom und Kühlung übertreffen über die Betriebszeit inzwischen die Anschaffungskosten Serverbeschaffungen ohne Betrachtung des Energieverbrauchs führen zu hohen Betriebskosten für Stromversorgung und Kühlung Energieeffiziente Systeme zeichnen sich durch eine bessere Total Cost of Ownership aus Grid Computing ist im Markt und in der TOP500 Liste angekommen Virtualisierungstechniken erlauben neue On-Demand Ansätze Mit Cloud-Computing sind neue Anbieter von Rechendienstleistungen am Markt Die Verbindung von Virtualisierungstechniken und Cloud-Computing bieten Ansätze für Virtual Private Grids und stoßen damit auf Neuland vor. 5
Energieeffizienz hat für die IT viele Seiten 100 Einheiten Rechenzentrum Server-Hardware Prozessor Server Auslastung 35. 33 geliefert* Kühlung Backup 55% IT Energie 45% Sonst. 70% Prozessor 30% Rechen- Leistung / Watt Bis zu 95% Leerlauf Ressourcen 5-20% Ø Last *Quelle: U.S. Department of Energy May 18, 2007 Steigerung der Effizienz Effizientere Kühlung und Energieversorgung Optimierte Server Hardware und Design, Energie Management Verbessertes Prozessor Design, Energie Management Verringerung ungenutzter Kapazitäten, Virtualisierung 6
Systems and Technology Group High Performance On Demand Solutions Deploy deep skills Customer Customer Collaboration Collaboration Internet Scale Computing Innovation Enablement Customer Customer Success Success Lead in best practices Create strategic assets Accelerate emerging technologies 7
Systems and Technology Group Einflüsse auf Cloud Computing Skyrocketing costs of power, space, maintenance, etc. Explosion of data intensive applications on the Internet Advances in multi-core computer architecture Fast growth of connected mobile devices Growth of Web 2.0enabled PCs, TVs, etc. 8
Systems and Technology Group Die Evolution des Enterprise Cloud Computing Enterprise Cloud Web-centric Cloud e.g. Google Dynamic Dynamic Information Information delivery delivery No mission critical No mission critical applications applications Optimized Optimized for for traffic traffic 9 Dynamic Dynamic & & request request driven driven Virtualized Virtualized Responsive Responsive to to changing changing needs needs Information Information and and transaction transaction delivery delivery Mission Mission critical critical applications applications Optimized for security, Optimized for security, auditablility, auditablility, data integrity and traffic data integrity and traffic Enterprise Data Center Consistent Consistent Transactional Transactional workloads workloads Mission Mission critical critical applications applications Optimized Optimized for for security, security, auditability and auditability and data data integrity integrity
Was unterscheidet Grid / Utility /On Demand Computing? Alle verfolgen ähnliche Ziele Effiziente und verwaltete Compute-Infrastruktur im großen Maßstab Unterstützung für hosted Service und virtualisierte Resourcen Security und Isolation für inter- und intra-enterprise Anwendungen Grid, OGSA offene Standards Fördert die Verwendung von Arbeits-Instanzen beruhend auf Middleware Modulen Geringer Fokus auf Interaktive Verarbeitung (ist möglich) Current Cloud Modell Middleware-Anwendungen sind mit dem OS Image zusammengepackt Gedacht um mehrstufigen Scale-Out Web Anwendungen zur Backend- Verarbeitung zu unterstützen Keine Abhängigkeit von einem Programmiermodell oder Job Manager Fokus auf der Optimierung von Data-Center Abläufen 10
IBM Plattformen für HPC und Clouds Blue Gene/P MPI und 4-wege OpenMP Supercomputer Designziel: Hohe Energie-Effizienz Höchste Skalierbarkeit BladeCenter als Linux Cluster 1 2 Sockel Blades Intel Xeon, AMD Opteron, POWER6 oder Cell BE Energie-effizientere Systeme Meist das TCO Optimum (Stromverbrauch über Standzeit) IBM POWER Systems mit POWER6 Prozessoren Multi Purpose Serversysteme Cluster aus großen SMP Systemen Hohe Speicherbandbreite Höchste Single Thread Performance Mehrfache Infiniband Interconnect IBM System x Linux Cluster mit x86 Prozessoren Commodity Based Hardware Rackmontierte 2-Sockel Systeme Intel Xeon oder AMD Opteron Gigabit Ethernet oder Infiniband Interconnect Oft das Preis / Leistungs Optimum idataplex Höhere Packungsdichte 84 Knoten pro Rack Gute Energieeffizienz Effektive Kühlung Für ISP, Cloud Computing und HPC Neu 11
IBM Blue Cloud Angebot Angekündigt November 2007 Liefert ein massively scalable und flexible compute Plattform für existierende und emerging data-intensive Workloads. Apache BladeCenter Virtual Machine Virtual Machine Virtual Machine Tivoli Monitoring Agent Virtual Machine Linux with Xen Virtualized Infrastructure Based on Linux & Xen IBM Monitoring v.6 Monitoring DB2 Provisioning Manager v.5.1 Provisioning Baremetal & Virtual Machines Provisioning Management Stack WebSphere Application Server Based on open standards and open source software Includes IBM software, systems technology and services Supports Power and x86 processors in first release Web 2.0 resource reservation system 12
Cloud-in-a-Box 1.0 Specification for System x Features include Self-service Web 2.0 user interface to easily request computing resources Automated configuration of Cloud environment Scheduling and reservations of computing resources Automated provisioning and de-provisioning of virtual Linux Red Hat images by Tivoli Provisioning Manager Xen open source systems virtualization Automated provisioning and de-provisioning of parallel computing clusters running Apache Hadoop Systems management and monitoring by IBM Tivoli Monitoring Open source Eclipse-based development tools for parallel applications Hardware Requirements IBM BladeCenter 14 LS21 high-density blade servers 2 AMD Opteron processors 8 GB Memory 293 GB Storage Software Requirements IBM Tivoli Provisioning Manager v5.x IBM Tivoli Monitoring v6.x Red Hat Enterprise Linux v5 Open Source Xen Virtualization Technology Open Source Hadoop software platform IBM Dojo Toolkit 13
Cloud-in-a-Box 1.0 Specification for PowerBlades Features include Self-service Web 2.0 user interface to easily request computing resources Automated configuration of Cloud environment Scheduling and reservations of computing resources Automated provisioning and de-provisioning of virtual Linux Red Hat images by Tivoli Provisioning Manager Automated provisioning and de-provisioning of parallel computing clusters running Apache Hadoop Systems management and monitoring by IBM Tivoli Monitoring Open source Eclipse-based development tools for parallel applications Hardware Requirements IBM Power Blade 14 JS21 blade servers 2 64-bit IBM PowerPC 970MP 8 GB Memory 146 GB Storage Software Requirements IBM Tivoli Provisioning Manager v5.x IBM Tivoli Monitoring v6.x Red Hat Enterprise Linux v5 Open Source Hadoop software platform IBM Dojo Toolkit Cloud management will run on System x and endpoints will run on System p. 14
Cloud Computing for Software Development First Commercial Cloud Computing Center Accelerates transformation to a service-led economy Eleven parks created for software outsourcing IBM Blue Cloud selected as technology base Workload Solution Patterns Software Development Eclipse Rational Application Developer ClearCase Software Architect Benefits Fast deployment of software development environments Up to 100K software developers Cost efficient shared infrastructure One of the model zones to offer software outsourcing services in China Mr. Zhu Weiping, Party Secretary of Bin Hu District, Wuxi Municipal Government, February 1, 2008 A milestone in service oriented computing T.W. Liu, Chairman and CEO, isoftstone, February 1, 2008 Cloud Computing Management Services Virtualized Physical Servers (Ensembles) System z, System x, System p, BladeCenter 15
Cloud Computing for Technology Incubation Innovation Portal Leverages IT to drive innovation across IBM Currently: 101,000 users, 65 active incubations, 35 graduated projects Benefits Enables worldwide collaboration on new technologies Speeds deployment of IT resources Shortens time to market through collaboration from 180 days to 30 days launch Easier for its 360,000-member staff to work as one virtual team Working prototype in two weeks and delivered a finished product in two months BusinessWeek, 2007 Workload Solution Patterns Technology Incubation DB2 Lotus Connections Lotus SameTime WebSphere Application Server Process Server Portal Server Cloud Computing Management Services Virtualized Physical Servers (Ensembles) System z, System x, System p, BladeCenter 16
Cloud Computing for Web 2.0 Collaboration China Telecom Transforms company s image as an innovator Largest telecom in the world Encourages collaboration within the company to generate new ideas and initiatives Benefits Accelerates time-to-market for new services Facilitates the management of new ideas Strengthens market research and development linkage Increased ideas generated by 250% Workload Solution Patterns Innovation Factory Idea Management Lotus Connections WebSphere Application Server DB2 Cloud Computing Management Services "IBM s Innovation Factory solution has enabled our researchers to be more productive in generating new ideas Mr. An Ming Li, Director of STTRI, November 15, 2007 Virtualized Physical Servers (Ensembles) System z, System x, System p, BladeCenter 17
Cloud Computing for Government-led Initiatives Vietnam Innovation Portal Accelerates Vietnam s evolution to a service-led economy Innovation platform for the Ministry of Science and Technology Benefits Fosters collaboration among education, research and industry Accelerates development of next generation skills Workload Solution Patterns Innovation Enablement Innovation Factory Lotus Connections WebSphere Portal Server DB2 Cloud Computing Management Services The delivery of VIP will help foster innovation in Vietnam, Dr. Tran Quoc Thang, Vice Minister of MoST, November 14, 2007 Virtualized Physical Servers (Ensembles) System z, System x, System p, BladeCenter 18
Systems and Technology Group Cloud Computing for Data Intensive Workloads Large Scale Information Processing Google/IBM Academic Initiative University initiative to promote open standards and emerging MapReduce parallel computing model Jointly provide compute platform of the future Workload Workload Solution Solution Patterns Patterns Benefits Trains students with next generation computing skills Optimizes emerging Internet scale workloads such as search, video, audio, 3D Internet We re aiming to train tomorrow s programmers to support a tidal wave of global Web growth and trillions of secure transactions every day. Sam Palmisano, IBM CEO, October 8, 2007 "Google is excited to partner with IBM to better equip students and researchers to address today s developing computational challenges" Eric Schmidt, Google CEO, October 8, 2007 19 Apache Hadoop Eclipse Cloud Computing Management Services Virtualized Virtualized IBM IBM and and Google Google Servers Servers
Cloud Computing in the New Enterprise Data Center Workload Solution Patterns Software Development Deploys development tools for immediate use Technology Incubation Reduces time to launch new offerings Innovation Enablement Expands sources of innovation, increases competitiveness Large Scale Information Processing Optimizes emerging Internet scale workloads Cloud Computing Management Services Self-service Admin Portal Workload Pattern Templates Administration Workflows SLA and Capacity Planning Workload Management Provisioning Monitoring Virtualized Physical Servers (Ensembles) System z, System x, System p, BladeCenter 20
Cloud Centers Around the World to Serve Clients Dublin, Ireland Seattle, WA Beijing, China Seoul, S Korea San Jose, CA US, East Coast Middle East Wuxi, China Bangalore, India Hanoi, Vietnam São Paulo, Brazil Active South Africa Planned 21
IBM s Beitrag zu Cloud Computing Projekten Cloud Computing for the Enterprise Provides secure transaction processing Supports richer applications Massive scalability of Web-centric model Enables access anywhere, centralizes services and data Dynamic workload scheduling and capacity management Improves utilization, reduces power consumption Request driven virtualized resource provisioning Simplifies IT Management Accelerates deployment of IT resources http://www.ibm.com/developerworks/websphere/zones/hipods/ 22