Shin, Dong In Distributed Computing System Laboratory 2005. 11. 21

1. Migration of Computing Environments Shin, Dong In Distributed Computing System Laboratory 2005. 11. 21

2. Introduction • Multiple work-site environment • User want some way in which he or she can see an identical environment everywhere he or she works. • Portable devices • Such as laptop computers. • Disadvantages • The user must carry a physical device to transport the environment. • Physical security of the portable device at all times. • Mainframe servers • Provide users with simple terminals as the only interface to the system. • Virtual Machine • Provides flexibility and security

3. Replication • The state of the machine • The state of the resources used by the operating system. • The applications running on the machine. • The code and data belonging to the operating system and applications. • A capsule • Capture of the state of a running machine and information about the processes currently active on the system. • Replication of the environment • Encapsulation of the state of the machine • Transmission across a network • Installation on the other computer before the user can take over operation at a second computer

4. Virtual Computers • Virtual machine technology • provides the ability to capture the entire state of a computer system. • Facilitates migration of a full computing environment. • Traditional machine vs Virtual machine migration • Virtual machine is better at smoothing out subtle configuration difference. • The process of environment encapsulation is simplified on a virtual machine. • A capsule does not contain any information about the mapping of the virtual resources.

5. Environment Migration Using VM Apps1 Data Data Apps2 Traditional Data Migration OS1 OS2 Hardware1 Hardware2 VM Migration Virtual Machine Virtual Machine Guest Apps Guest Apps Guest OS Guest OS Virtual Machine Monitor 2 Virtual Machine Monitor 1 Hardware 2 Hardware 1

6. Virtual Computer • The issues in developing a solution for migration problem • The time taken to migrate the large state of computer • Transmit a small part of the state at first, transmitting additional parts of the state as needed. • Packaging and secure transmitting the information • Compression and encryption techniques • The different ISA between the user’s VM and the host computers • Binary translation and optimization techniques

7. Migration Using a Distributed File System (ISR) • Internet Suspend/Resume (ISR) • The system allows a user to suspend operation on one machine, travel to another machine, and resume execution on the other machine. • Use the VMware GSX Server • MobileIP technology • To access the network, it change the IP address and all references in the saved environment.

8. The Internet Suspend/Resume Scheme • Suspend/Resuming • “pull” model, with the destination computer reading in the environment state file on demand, when the user logs in to the destination machine. • The pattern of usage • Incremental loaded module makes the migration more responsive to the user. • Optimization • The suspension of the environment at a site need not completely eliminate the environment at the site • Reusing portions of the environment from the environment of other users.

9. State Encapsulation in the Stanford Collective • Collective System • “push” model • The user can have instantaneous access to his or her environment at the destination site. • Intel IA-32 platform running on the VMware GSX server. • Virtual Private Network (VPN) • The collective system enhances the virtual machine monitor to enable tunneling of network packets travelling to and from the capsule’s old network over VPN. • The migration takes about 20 minutes using standard 384 kbps DSL lines.

10. State Encapsulation in the Stanford Collective • Reducing Memory State Before Migration • Which pages do or do not belong to the current working set at suspend time without intruding into the guest OS. • Using Ballooning Program of VMware. • It is triggered by the VMM when it receives a suspend request. • The pages of the inactive processes are released in the guest operating system. • Issues • How may pages are recovered through ballooning program?

11. Inflating a balloon When the server wants to reclaim memory Driver allocate pinned physical pages within the VM Increase memory pressure in the guest OS, reclaim space to satisfy the driver allocation request Driver communicates the physical page number for each allocated page to ESX server Deflating Frees up memory for general use within the guest OS Ballooning in VMware

12. State Encapsulation in the Stanford Collective • Reducing the Size of the Transmitted Packet • The state represented by the disk is formidable and the disk hold many types of data. In typical user migration, however, the required disk image is usually not a big change from the image that was left on a platform when last visited. • Incremental Disk technology (capsule hierarchy) • Copy-on-write concept. • Whenever a copy of a disk is needed, the entire disk contents are not copied. Rather, the image of the disk is represented by a set of pointers to unique copies of segments of the disk. • Before changes are made to the disk image, all nodes in the path from the root to the leaf corresponding to the image have to be transferred.

13. Example Capsule Hierarchy for a University Root node : the state of the disk is saved in its entirety University Capsule Department1 Capsule Department12 Capsule Student1 Capsule Student2 Capsule Student3 Capsule Department2 Updated Capsule Student4 Capsule Student1 Working Capsule modified Segments Child node : Save the incrementalstates of the parent andthe child The state of the disk at any nodecan be derived from the disk infocontained in the nodes along thepath from the root.

14. State Encapsulation in the Stanford Collective • Reducing Transmission Time and Bandwidth by Exploiting Redundancy in Disk Blocks • Sample situations when a needed block is already available on the system. • A trail of disk blocks when a user moving back and forth between two systems can potentially be reused. • Memory resident blocks of program files. • Memory resident blocks other than program files. • Different user’s program files or data files. • Hashed copy scheme • Each block is associated with a hash value to identify uniquely. • SHA-1, which has a very low collision probability.

15. Experiment Result

16. Migration of Virtual Machines In VMotion • VMotion • The part of VirtualCenter infrastructure management software developed by VMware ESX Server. • It manages a cluster of Intel IA-32 virtual machine system connected in a LAN . • VMotion characteristics • Load balancing • improve the response time of the system through better utilization of resources. • Security • quarantine a virtual machine that has been attacked • Collocation • Bring communicating virtual machines close together • Fault tolerance • Move a failing host to another processor • Power management • Move the load away from an overheated processor • Maintenance • Move the load away from some processor while it is upgraded.

17. Overview of the VM VritualCenter VC Client (User 1) VC Client (User 1) VC Client (User 1) VC Client (User 1) VC Management Server VC Database VCagent VCagent VCagent VM1 VM2 VM3 VM1 VM2 VM3 VM1 VM2 VM3 SAN Data Store Host A Host B Host C

18. Migration of Virtual Machines In VMotion • Restriction of VMotion • The source and destination computers must be in the same server cluster managed by the same VirtualCenter manager • The file systems on the src and dest computers must be identical and located on shared disks in a SAN. • The processors must have the same architecture. • Gigabit Ethernet adapter. • The virtual machines should be running only stand-alone applications.

19. Thank You!