Docstoc

Virtual Machine Introspection Grid Virtualization Engine Design

Document Sample
Virtual Machine Introspection Grid Virtualization Engine Design Powered By Docstoc
					Grid Virtualization Engine: Design,
Implementation, and Evaluation


  Lizhe Wang, Senior Member, IEEE, Gregor
 von Laszewski, Member, IEEE, Jie Tao, and
                            Marcel Kunze


                                          報告人:倪丞頤
                Distributed System Lab.             1
Outline
  Introduction
  Purpose
  Detail about GVE
  Implementation
  Experiments
  Conclusion


              Distributed System Lab.   2
Introduction
  Virtual machines offer various
  advantages which can help Grid
  communities to solve research issues.
  The work proposed in this paper
  focuses on building a Web service
  based virtual machine provider for Grid
  infrastructures.


               Distributed System Lab.   3
Difficulties of Grid Computing
  Qualities of service (QoS) of resource
  provision and performance isolation.
  Configurations of customized runtime
  environment for Grid applications need
  administration privileges




              Distributed System Lab.      4
Virtualization’s Benefit for Grid
Systems
  On demand creation and customization
      Users can create a customized VM
  Performance isolation
      VM’s Users can expect dedicated computing
       environments
  Legacy software support
      Customized VM are compatible
  Easy management
      VM can offer users with the “root” access


                     Distributed System Lab.       5
Challenges for VM-based Grid
Systems
  Site Autonomy
     The host resources are commonly owned and controlled by
      different institutes or organizations.
  Hierarchy
     A virtual machine-based Grid system is hierarchical in nature.
  Heterogeneity
     A VM based Grid system includes various virtual machine.
  Large scale distribution
     Computing and data centers frequently build Grid across
      distributed sites.




                       Distributed System Lab.                   6
Grid Virtualization Engine (GVE)

  The GVE is light weight distributed
  middleware, which enables users to
  operate virtual machines and provides
  virtual distributed environments across
  distributed computing centers.




               Distributed System Lab.      7
Aims & Unique Contribution
  To work in a large scale distributed Grid
  infrastructure.
  Defines a hierarchical virtual machine based Grid
  system model.
  Builds a scalable and efficient GVE service that can
  provide virtual machine resources to Grid system
  based on popular VMMs.
  Makes a performance evaluation on GVE and virtual
  machines with a nontrival application.
  Provides our experience of employing virtual machine
  resources for Grid infrastructures.

                   Distributed System Lab.          8
VM-Based Grid System Model




          Distributed System Lab.   9
VM-Based Grid System Model
  Access Service
     Allows remote users to access resources of the computer
      center.
     The GVE service is developed and integrated in this level.
  Virtual Machines
     GVE service operates on virtual machines in this level.
  Host Resources
     Which are installed with virtual machine hypervisors.
     The GVE Agent Service is implemented in this level with aids
      of VMM APIs and SDKs.




                       Distributed System Lab.                     10
Architecture of GVE




           Distributed System Lab.   11
GVE
 GVE Site Service
     Which resides on the access point of a Computer
      Site
     Communicates with GVE Agents
 GVE Agents
     Which are responsible for accepting requirements
      from GVE Site service.
     Organizing the requirements and passing the
      requirements to proper underlying VMMs with
      corresponding APIs, commands, and messages.
 Virtual Machine Disk Database

                    Distributed System Lab.         12
GVE Site Service
  It manages host resources inside the center by communicating
  with the GVE Agent Services
  GVE Web Service
     It decides the Agent Service to which virtual machine requests
      should be sent and defines the policies of resource allocation.
  User Information Service
     Make an inventory of the data that the GVE Site Service needs to
      store and create an interface that provides methods to store and
      manage data.
  User Information Database
     Records the management policies and accounting information of
      virtual machine usage.
     The place where the user information and the virtual machine
      information is stored.



                         Distributed System Lab.                        13
GVE Agent Service
  Run on the host resources and is virtualization
  technology dependent.
  All GVE Agent Services implement the same Web
  Service interface.
  The Registry Service provides the following functions
  to access the Registry Database, which stores the
  state of virtual machines, the GVE Site information,
  the state of the virtual machine request and
  management.
  The Virtual Machine Disk Database may not be
  directly implemented by the GVE developer.

                   Distributed System Lab.           14
Use Case of GVE




          Distributed System Lab.   15
Implementation of GVE
  To define the services provided by describing them
  using Web Service Description Language (WSDL).
  Java interfaces and Web Service artifacts are
  generated using JAXB.
  To create a database and a Java interface to access
  and manage the database.
  For the GVE Site Service, the User Information
  Database is created.
  The Java interfaces and classes are implemented
  according to their functions and roles with VMM APIs.


                   Distributed System Lab.          16
GVE Site Service Implementation
  The User Information Service is a Web service used by the GVE
  Site Service to access and to manage the data stored in the
  User Information Database.
  The User Information Database is defined with the data entities
  of Agent services, virtual machines and user information.
  The GVE Site Service defines various Web Service portTypes in
  the WSDL file.
  Job, RequestVirtualMachine or GetRequestVirtualMachineResult,
  means a virtual machine request or a virtual machine
  management operation.
  The JobState is the object returned to GVE clients




                      Distributed System Lab.                 17
Implementation of the GVE Agent
Service
  The Agent Web service implements a lot of
  operations about VM with underlying Virtual Machine
  Monitor APIs.
  Implementation of Registry Service and Disk
  Database Access
     The data structures are defined firstly.
     The interface of the Registry Service is then defined in the
      WSDL file, which contains the portTypes.
     The implementation of the interface termed as DataProvider.
     The Registry Database tables are generated at runtime.
  VMware Infrastructure SDK does not provide APIs
     Solved by InstallService, deploying an extraWeb Service
      component in VM


                       Distributed System Lab.                  18
GVE instance start overhead




     The max overhead of starting time is around 17%
                    Distributed System Lab.            19
GVE network performance




  The throughputs between virtual machines can reach around 90% of
  those between real machines.
                        Distributed System Lab.                      20
CMS Performance Evaluation



   OSCAR test results                                ORCA-digi test results




   ORCA-dst test results                         Pythia test results

                           Distributed System Lab.                            21
CMS Performance Evaluation



                                                  ROOT test results




                                        In General, VM can attain
                                        70%-95% performance of
                                        real machines.
   OSCAR test results


                        Distributed System Lab.                       22
Service composition of workflow
system




            Distributed System Lab.   23
Execution process of the
workflow services




            Distributed System Lab.   24
Conclusion
  GVE is a software layer which resides
  between users and various Virtual
  Machine implementations.
  The GVE is designed and implemented
  in modularity.
  The GVE is designed and implemented
  in the hierarchical flavor.

              Distributed System Lab.   25
Future Work
  Development of industrial security
  control mechanism.
  Accommodation of more virtual
  machine technologies.
  Moving the GVE work into the Cloud
  computing context.



              Distributed System Lab.   26

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:12
posted:8/23/2011
language:English
pages:26