Robust live migration using shared filesystem

• US 8,671,238 B2
• Filed: 07/13/2011
• Issued: 03/11/2014
• Est. Priority Date: 07/13/2010
• Status: Active Grant

• Current Assignee: Vmware LLC (Broadcom, Inc.)
• Original Assignee: VMware, Inc. (Broadcom, Inc.)
• Inventors: Mashtizadeh, Ali, Tarasuk-Levin, Gabriel
• Primary Examiner(s): SCHNEE, HAL W
• Application Number:
  US13/182,303
• Publication Number:
  US 20120017031A1
• Time in Patent Office: 972 Days
• Field of Search: 711/6, 711/147, 711/E12.065, 718/1
• US Class Current: 711/6
• CPC Class Codes: G06F 12/08 in hierarchically structured memory systems, e.g. virtual memory systems
• CPC Class Codes: G06F 15/17331 Distributed shared memory [DSM], e.g. remote direct memory access [RDMA]
• CPC Class Codes: G06F 2009/4557 Distribution of virtual machine instances; Migration and load balancing
• CPC Class Codes: G06F 2009/45583 Memory management, e.g. access or allocation
• CPC Class Codes: G06F 3/0647 Migration mechanisms
• CPC Class Codes: G06F 9/45545 Guest-host, i.e. hypervisor is an application program itself, e.g. VirtualBox
• CPC Class Codes: G06F 9/45558 Hypervisor-specific management and integration aspects
• CPC Class Codes: G06F 9/4856 resumption being on a different machine, e.g. task migration, virtual machine migration G06F9/5088 takes precedence

Abstract

A method for transferring guest physical memory from a source host to a destination host during live migration of a virtual machine (VM) involves creating a file on a shared datastore, the file on the shared datastore being accessible to both the source host and the destination host. Pages of the guest physical memory are transferred from the source host to the destination host over a network connection and pages of the guest physical memory are written to the file so that the destination host can retrieve the written guest physical pages from the file.

Claims

1. A method for transferring guest physical memory from a source host to a destination host during live migration of a virtual machine (VM) to which the guest physical memory is associated, the source host being a physical computer hosting the VM prior to the live migration and the destination host being a physical computer hosting the VM after the live migration completes, the method comprising:

creating a file on a shared datastore, the file on the shared datastore being accessible to both the source host and the destination host;

transmitting pages of the guest physical memory from the source host to the destination host over a network connection that places the source host and destination in communication with each other; and

writing pages of the guest physical memory to the file so that the destination host can retrieve the written guest physical pages from the file, wherein the writing of pages of the guest physical memory occurs concurrently with the transmitting of the pages over the network connection.

2. The method of claim 1, wherein the transmitting of the pages of the guest physical memory over the network occurs in a first order of guest physical memory pages and the writing of the pages to the guest physical memory occurs in a reverse order, the reverse order being a reverse of the first order.

3. The method of claim 1, wherein a load balancing algorithm is employed to balance transferring of guest physical memory over both a first datapath comprising the network connection and a second datapath comprising the file, the file being a multiwriter file enabling a bidirectional communication between the source host and the destination host.

4. The method of claim 1, wherein the file is a multiwriter file enabling a bidirectional communication between the source host and the destination host.

5. The method of claim 4, further comprising:

writing the pages of guest physical memory to the multiwriter file when a network failure or network degradation is detected.

6. The method of claim 4 wherein the multiwriter file is one of a plurality of files enabling the bidirectional communication between the source host and the destination host.

7. The method of claim 1:

wherein the file is a multiwriter file;

wherein the network connection is a first datapath for the transmitting of the pages of the guest physical memory from the source host to the destination host and the multiwriter file is a second datapath for the writing of the pages of the guest physical memory to the multiwriter file; and

wherein the pages of guest physical memory are transferred to the destination over the first and second datapath concurrently using a load balancing algorithm.

8. A non-transitory computer readable storage medium storing instructions that when executed by a computer system cause the computer system to perform a method for transferring guest physical memory from a source host to a destination host during live migration of a virtual machine (VM) to which the guest physical memory is associated, the source host being a physical computer hosting the VM prior to the live migration and the destination host being a physical computer hosting the VM after the live migration completes, the method comprising:

creating a file on a shared datastore, the file on the shared datastore being accessible to both the source host and the destination host;

transmitting pages of the guest physical memory from the source host to the destination host over a network connection that places the source host and destination in communication with each other; and

writing pages of the guest physical memory to the file so that the destination host can retrieve the written guest physical pages from the file, wherein the writing of pages of the guest physical memory occurs concurrently with the transmitting of the pages over the network connection.

9. The non-transitory computer readable storage medium of claim 8, wherein the transmitting of the pages of the guest physical memory over the network occurs in a first order of guest physical memory pages and the writing of the pages to the guest physical memory occurs in a reverse order, the reverse order being a reverse of the first order.

10. The non-transitory computer readable storage medium of claim 8, wherein a load balancing algorithm is employed to balance transferring of guest physical memory over both a first datapath comprising the network connection and a second datapath comprising the file, the file being a multiwriter file enabling bidirectional communication between the source host and the destination host.

11. The non-transitory computer readable storage medium of claim 8, wherein the file is a multiwriter file enabling a bidirectional communication between the source host and the destination host.

12. The non-transitory computer readable storage medium of claim 11, wherein the method further comprises:

writing the pages of guest physical memory to the multiwriter file when a network failure or network degradation is detected.

13. The non-transitory computer readable storage medium of claim 11, wherein the multiwriter file is one of a plurality of files enabling the bidirectional communication between the source host and the destination host.

14. The non-transitory computer readable storage medium of claim 8:

wherein the file is a multiwriter file;

wherein the network connection is a first datapath for the transmitting of the pages of the guest physical memory from the source host to the destination host and the multiwriter file is a second datapath for the writing of the pages of the guest physical memory to the multiwriter file; and

wherein the pages of guest physical memory are transferred to the destination over the first and second datapath concurrently using a load balancing algorithm.

15. A method of migrating virtual machines from a source host to a destination host over a network, comprising:

copying memory pages associated with an executing state of a virtual machine from the source host to the destination host over the network;

before all of the memory pages associated with the executing state of the virtual machine have been copied from the source host to the destination host over the network, switching execution control of the virtual machine from the source host to the destination host; and

after the execution control of the virtual machine has been switched from the source host to the destination host, receiving at the source host a request for contents of a memory page from the destination host and, in response thereto, writing the contents of the memory page in a multiwriter file that is shared by the source host and the destination host, wherein the request is written in the multiwriter file by the destination host and read from the multiwriter file by the source host and wherein the source host writes a reply to the destination host in the multiwriter file in addition to the contents of the memory page.

16. The method of claim 15, wherein the destination host reads the reply and the contents of the memory page written by the source host.

17. A method for transferring guest physical memory from a source host to a destination host during live migration of a virtual machine (VM) to which the guest physical memory is associated, the source host being a physical computer hosting the VM prior to the live migration and the destination host being a physical computer hosting the VM after the live migration completes, the method comprising:

creating a multiwriter file on a shared datastore, the multiwriter file on the shared datastore being accessible to both the source host and the destination host;

transmitting pages of the guest physical memory from the source host to the destination host over a network connection that places the source host and destination in communication with each other, wherein the network connection is a first datapath for the transmitting of the pages of the guest physical memory from the source host to the destination host and the multiwriter file is a second datapath for the writing of the pages of the guest physical memory to the multiwriter file; and

writing pages of the guest physical memory to the file so that the destination host can retrieve the written guest physical pages from the file, wherein the pages of guest physical memory are transferred to the destination over the first and second datapaths concurrently using a load balancing algorithm.