The Cirrus


Published on

Published in: Technology, Business
  • Be the first to comment

  • Be the first to like this

No Downloads
Total Views
On Slideshare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • ICAC 2007 © Hewlett-Packard Company
  • The Cirrus

    1. 1. The Cirrus & Cumulus Project: Build a Scientific Cloud for a Data Center Lizhe Wang and Marcel Kunze Steinbuch Centre for Computing (SCC) Karlsruhe Institute of Technology (KIT) Germany
    2. 2. Karlsruhe Institute of Technology (KIT) <ul><li>Cooperation between Research Centre Karlsruhe und Karlsruhe University </li></ul><ul><li>Largest scientific center in Germany </li></ul><ul><li>8.000 scientists, 18.000 students </li></ul><ul><li>Annual budget: > 500 Million Euro </li></ul><ul><li>R&D focus: Energy research and Nanotechnology </li></ul>> +
    3. 3. Google Trends: Grid Computing vs. Cloud Computing | Autor | Steinbuch Centre for Computing | 06.09.2007
    4. 4. OpenCirrus Cloud Computing Research Testbed <ul><li>An open, internet-scale global testbed for cloud computing research </li></ul><ul><ul><li>a tool for collaborative research </li></ul></ul><ul><ul><li>focus: data center management & cloud services </li></ul></ul><ul><li>Resources: </li></ul><ul><ul><li>Multi-continent, multi-datacenter, cloud computing system </li></ul></ul><ul><ul><li>“ Centers of Excellence” around the globe </li></ul></ul><ul><ul><ul><li>each with 100–400+ nodes and up to ~2PB storage </li></ul></ul></ul><ul><ul><ul><li>and running a suite of cloud services </li></ul></ul></ul><ul><li>Structure: a loose federation </li></ul><ul><ul><li>Sponsors: HP Labs, Intel Research, Yahoo! </li></ul></ul><ul><ul><li>Partners: UIUC, Singapore IDA, KIT, NSF </li></ul></ul><ul><ul><li>Members: System and application development </li></ul></ul><ul><li>Available towards end of the year </li></ul><ul><li>Great opportunity for Cloud R&D </li></ul>May 21, 2010 © Hewlett-Packard Company
    5. 5. Physical and Virtual Resource Sets (PRS, VRS) experiment save/restore
    6. 6. A possible Definition of Cloud Computing <ul><li>Definition </li></ul><ul><ul><li>A computing Cloud is a set of network enabled services, providing scalable, QoS guaranteed, inexpensive computing platforms on demand, which could be accessed in a simple and pervasive way . </li></ul></ul><ul><li>Functionalities </li></ul><ul><ul><li>SaaS: Software as a Service </li></ul></ul><ul><ul><li>HaaS: Hardware as a Service </li></ul></ul><ul><ul><li>DaaS: Data as a Service </li></ul></ul><ul><ul><li>PaaS: Platform as a Service </li></ul></ul><ul><ul><li>IaaS: Infrastructure as a Service </li></ul></ul>
    7. 7. Anatomy of Cloud Computing <ul><li>Key features </li></ul><ul><ul><li>User-centric access </li></ul></ul><ul><ul><li>On-demand service provisioning </li></ul></ul><ul><ul><li>QoS guaranteed offer </li></ul></ul><ul><ul><li>Autonomy </li></ul></ul><ul><ul><li>Scalability & flexibility </li></ul></ul><ul><li>Enabling technologies </li></ul><ul><ul><li>Virtualization </li></ul></ul><ul><ul><li>Service flow/workflow orchestration </li></ul></ul><ul><ul><li>Web service & SOA </li></ul></ul><ul><ul><li>Web 2.0 </li></ul></ul><ul><ul><li>World-wide distributed storage & file system </li></ul></ul><ul><ul><li>Parallel & distributed programming model </li></ul></ul>
    8. 8. Clouds vs. Grids: A Comparison Difficult use and deployment Need new user interface, e.g., commands, APIs, SDKs, services … Easy to use/deploy, no complex user interface required User interface Special application domains like High Energy Physics Suited for generic applications Application Well developed, maintained and documented Proprietary, several reference implementations exist (e.g. Amazon) Middleware Publicly funded: Use for free Commercial: Pay-as-you-go Business Model Geographically distributed, heterogeneous resource, no central control, VO Research and academic organization One or few data centers, heterogeneous/homogeneous resource under central control, Industry and Business Infrastructure Resource sharing Job execution Provide desired computing platform via network enabled services Objective No Yes On-demand provisioning Little support Possible QoS HPC, Grid infrastructure, middleware, … Virtualization, SaaS, Web 2.0, Web service, … Enabling technology Grid Computing Cloud Computing
    9. 9. Cumulus: A Cloud Computing Prototype
    10. 10. Re-engineering the Globus Virtual Workspace Service (GVWS) <ul><li>GVWS Limitations </li></ul><ul><ul><li>Force all backends to install GVWS control agents </li></ul></ul><ul><ul><li>Cloud users need to select network solutions </li></ul></ul><ul><ul><li>Cloud users need to prepare VM images </li></ul></ul><ul><li>Re-engineering the GVWS for the Cumulus </li></ul><ul><ul><li>Remove GVWS control agents, GVWS frontend talks directly to LVMS, e.g., OpenNEbula </li></ul></ul><ul><ul><ul><li>Consider the similar scenario: Globus + PBS </li></ul></ul></ul><ul><ul><li>The “forward” network requirements to LVMS </li></ul></ul><ul><ul><li>Prepare VM image via OS Farm for users </li></ul></ul>
    11. 11. OpenNEbula as a Backend <ul><li>The Cumulus frontend communicates with OpenNEbula via: </li></ul><ul><ul><li>SSH </li></ul></ul><ul><ul><li>XML-RPC </li></ul></ul><ul><li>Improvement </li></ul><ul><ul><li>NIS  LDAP: shared user management - oneadmin </li></ul></ul><ul><ul><li>NFS  OCFS (Oracle Cluster File System): shared volumes for VM images </li></ul></ul><ul><li>Can develop other plug-in, embedded in the Cumulus frontend, for other LVMS, e.g., VMware Virtual Infrastructure </li></ul>
    12. 12. OS Farm as a VM Provisioning Tool <ul><li>Development of CERN OpenLab </li></ul><ul><li>OS Farm as server to generate VM images </li></ul><ul><li>Can accept HTTP requirements, or via wget: </li></ul><ul><ul><li>wget </li></ul></ul><ul><ul><li>class=slc_old&arch=i386&filetype=.tar&group=core&group=base </li></ul></ul><ul><li>A Java client is embedded in the Cumulus frontend to invoke OS Farm service dynamically. </li></ul>
    13. 13. The “forward” Network Solution <ul><li>Users do not need to specify the network configuration of a VM </li></ul><ul><li>Users only get network access to VM, e.g., IP address or hostname + domain name </li></ul><ul><li>Cumulus forwards the network requirements to backend (LVMS), e.g., OpenNEbula </li></ul><ul><li>Network solutions on Cumulus backend: </li></ul><ul><ul><li>OpenNEbula starts a VM, listens to central DHCP server, arranges an IP address to VM, returns it to Cumulus frontend </li></ul></ul><ul><ul><li>Set hostname when the virtual machine is created </li></ul></ul><ul><ul><li>Lease an IP address from the IP pool </li></ul></ul><ul><li>Cumulus frontend returns the IP address or the host name to the client. </li></ul>
    14. 14. Thank you! Question? Contact: Lizhe Wang, Marcel Kunze Steinbuch Centre for Computing (SCC) Karlsruhe Institute of Technology (KIT) Hermann-von-Helmholtz-Platz 1 D-76344 Eggenstein-Leopoldshafen, Germany [email_address] [email_address]