JP5992245B2 - Virtual machine migration system and method - Google Patents

Description

  The present invention relates to VM migration in which a virtual machine (VM) on a server is moved (migrated) to another server, and in particular, a VM on a server is moved to another server in a different data center. It relates to VM migration.

  In a data center, it is a common practice to perform VM migration to move a VM on a server to another server for server load balancing, power saving measures, and the like.

  Recently, not only VM migration between servers existing on the same LAN (Local Area Network), VM migration between servers existing on different LANs, but also different LANs moving VMs between servers existing on different LANs There is also a demand for VM migration between them.

  The present invention is intended for VM migration between different LANs.

  VM migration between different LANs requires that the networks connecting the LANs be linked together with the VM migration. As an example of a system that performs VM migration between different LANs, a VM migration system disclosed in Patent Document 1 can be cited.

  FIG. 7 is a diagram for explaining the configuration and operation of a conventional VM migration system disclosed in Patent Document 1. In FIG.

  Referring to FIG. 7, in the conventional VM migration system, on the LAN (first LAN) of the data center 10 that is the first data center, a GW (Gateway) 11 that is the first GW, And a server 12 that is a first server connected to the GW 11.

  On the LAN (first LAN) of the data center 20 that is the second data center, a GW 21 that is the second GW and a server 22 that is the second server connected to the GW 21 are provided. It has been.

  Further, on the client side LAN (third LAN), it is connected to the GW 11 of the data center 10 via the VLAN 1 which is the first VLAN (Virtual Local Area Network), and the second VLAN is the VLAN 2. And a PC (Personal Computer) 32 which is a client terminal connected to the GW 31 is provided.

  In the conventional VM migration system, a VCenter (Virtual Center) server 40 connected to each of the servers 12 and 22 and a control device 50 connected to each of the GWs 11, 21, and 31 are provided. The VCenter server 40 corresponds to the server management server (DC Manager) 162 of Patent Document 1, and the control device 50 corresponds to the management server (Mng. Server) 150 of Patent Document 1.

  In FIG. 7, assuming that the VM 13 on the server 12 of the data center 10 is moved to the server 22 of the data center 20, the VM after the VM 13 is moved is indicated as a VM 23. Therefore, the VMs 13 and 23 do not exist at the same time.

  Hereinafter, an operation of moving the VM 13 on the server 12 of the data center 10 to the server 22 of the data center 20 in the conventional VM migration system will be described.

  The VCenter server 40 monitors the states of the servers 12 and 22.

  When the VM 13 on the server 12 detects that the server 12 is heavily loaded, the VCenter server 40 instructs the server 12 to start moving the VM 13 to the server 22 of the data center 20 in step A1. To do.

  When receiving the VM movement start instruction, the server 12 cooperates with the server 22 in step A2 to copy the data on the memory used by the VM 13 to the server 22 (VM movement process). At this time, the VM 13 is kept operating.

  When the VM migration process ends, the server 12 stops the VM 13 in step A3.

  Here, since the operation of the VM 13 continues even during the execution of the VM migration process in step A2, the contents of the memory of the VM 13 change and a difference is generated from the VM 23 after the migration. Therefore, in step A4, the server 12 transmits the difference to the server 22 and eliminates the difference between the VM 13 and the VM 23 (VM migration final process). As a result, the VM 23 can be started from the state where the VM 13 is stopped.

  When the VM migration final process is completed, the server 22 activates the VM 23 in step A5. The VMs 13 and 23 have the same MAC (Media Access Control) address and the same IP (Internet Protocol) address.

  After starting, the VM 23 notifies (broadcasts) an ARP (Address Resolution Protocol) packet or an RARP (Reverse ARP) packet to the LAN in the data center 20 in step A6. Note that ARP is a protocol for resolving a MAC address from an IP address, and RARP is a protocol for resolving an IP address from a MAC address.

  At this time, since the control device 50 does not exist on the LAN in the data center 20, it cannot directly receive the ARP packet or the RARP packet from the VM 23. Therefore, an auxiliary program 24 for receiving ARP packets and RARP packets is provided on the LAN in the data center 20. When receiving the ARP packet or the RARP packet from the VM 23, the auxiliary program 24 notifies the control device 50 that the VM 23 has been activated in Step A7.

  Upon receiving the VM activation notification, the control device 50 instructs the GW 31 on the client side to perform VLAN switching from VLAN 1 to VLAN 2 in step A8.

  When receiving the VLAN switching instruction, the GW 31 performs switching from VLAN 1 to VLAN 2 in step A9.

JP 2011-210032 A

  As described above, in the conventional VM migration system disclosed in Patent Document 1, switching of the network accompanying VM migration between different LANs is performed by the control device 50.

  However, in this case, since it is necessary for the operator to set detailed data of the node and the network in the control device 50, there is a problem that the operator needs to set / change the control device 50 every time the network is changed.

  In addition, when the control device 50 fails, there is a problem that the entire network cannot be connected.

  Therefore, an object of the present invention is to enable VM migration between different LANs without an operator setting detailed data of nodes and networks in the control device, and to prevent the entire network from being lost due to the failure of the control device. It is an object of the present invention to provide a technology capable of avoiding that connection is impossible.

The virtual machine migration system of the present invention includes:
A first GW and a first server provided on the first LAN of the first data center;
A second GW and a second server provided on the second LAN of the second data center;
A network switching device provided on a third LAN on the client side, connected to the first GW via the first VPN, and connected to the second GW via the second VPN; Have
When the virtual machine on the first server moves to the second server and the virtual machine is started on the second server, the second GW transmits the virtual machine via the second VPN. Send a specific IP packet to the client side,
The network switching device performs L3 learning based on the specific IP packet, and switches the flow table describing the condition of the IP packet and the processing content of the IP packet.

The virtual machine migration method of the present invention includes:
A first GW and a first server on the first LAN of the first data center;
A second GW and a second server are provided on the second LAN of the second data center;
On the third LAN on the client side, there is provided a network switching device that is connected to the first GW via the first VPN and is connected to the second GW via the second VPN,
When the virtual machine on the first server is moved to the second server and the virtual machine is started on the second server, the second GW transmits the virtual machine via the second VPN. Send a specific IP packet to the client side,
The network switching device performs L3 learning based on the specific IP packet, and switches the flow table describing the condition of the IP packet and the processing content of the IP packet.

  According to the present invention, the second GW of the second data center moves the virtual machine on the first server of the first data center to the second server of the second data center, When the virtual machine is started on the server, a specific IP packet is transmitted to the client side via the second VPN, and the network switching device on the client side performs L3 learning based on the specific IP packet, and the flow table Switch.

  In other words, the network switching device autonomously detects the movement of the virtual machine by the L3 learning function, and switches the network by switching the flow table.

  This eliminates the need for the control device used in the conventional method, prevents the entire network from being unable to be connected due to a failure of the control device, and allows the operator to connect the details of the node, network, etc. The effect that there is no need to set data is obtained.

It is a figure explaining the structure and operation | movement of the migration system of VM of the 1st Embodiment of this invention. It is a figure explaining operation | movement of OFC33 in the 1st Embodiment of this invention. It is a flowchart explaining operation | movement of OFC33 in the 1st Embodiment of this invention. It is a figure explaining the state before and behind the switching of the flow table of OFS34 in the 1st Embodiment of this invention. It is a figure explaining the structure and operation | movement of the migration system of VM of the 2nd Embodiment of this invention. It is a figure explaining operation | movement of GW21 in the 2nd Embodiment of this invention. It is a figure explaining the structure and operation | movement of the migration system of the conventional VM.

EMBODIMENT OF THE INVENTION Below, the form for implementing this invention is demonstrated with reference to drawings.
(1) First Embodiment FIG. 1 is a diagram illustrating the configuration and operation of a VM migration system according to this embodiment.

  Referring to FIG. 1, the VM migration system of the present embodiment is different from the conventional system of FIG. 7 in that the client side GW 31 is replaced with a network switching device 35 realized by OpenFlow. Further, the VM migration system according to the present embodiment does not require the auxiliary program 24 or the control device 50 in the data center 20.

  OpenFlow is composed of two, an OFC (OpenFlow Controller) 33 and an OFS (OpenFlow Switch) 34, and the OpenFlow protocol defines the protocol between the OFC 33 and the OFS 34.

  The OFS 34 determines the flow operation based on the Match Field.

  The process for the flow is called an action, and the OFS 34 processes the packet according to the description of the action.

  The OFS 34 holds a flow table in which a Match Field and an Instruction that indicates an action (processing content) of an IP packet that matches the Match Field condition are described.

  The operation of moving the VM 13 on the server 12 of the data center 10 to the server 22 of the data center 20 in the VM migration system of this embodiment will be described below.

  First, the processing of steps B1 to B5 similar to steps A1 to A5 of FIG. 7 is performed.

  Here, when the VM 13 before the movement is being executed, the VM 23 after the movement takes over the process being executed by the VM 13 and executes it. Moving a VM that is executing a process to another server is called VM live migration.

  At this time, it is assumed that the process handed over from the VM 13 to the VM 23 is a process of transmitting an IP packet toward the client side (for example, a process in which the PC 32 downloads data from the VM 13).

  In this case, the VM 23 not only notifies (broadcasts) the ARP packet, the RARP packet, the GARP (Gratuitous ARP) packet, etc. to the LAN in the data center 20 in step B6 after the activation, but also performs the processing taken over from the VM 13. Such an IP packet is also transmitted from the GW 21 to the client side via the VLAN 2. In the following, the IP packet that the VM 23 transmits immediately after startup is referred to as a Fast packet.

  Therefore, in step B7, the network switching device 35 performs L3 (Layer 3) learning based on the Fast packet and automatically switches the flow table.

  L3 learning is a function that autonomously changes the flow rules of uplink packets and downlink packets based on downlink packets. The L3 learning function can be implemented by programming the OFC 33 using OpenFlow.

  Here, the L3 learning operation of the network switching device 35 will be described in detail.

  2 and 3 are diagrams and flowcharts for explaining the operation of the OFC 33, respectively. FIG. 4 is a diagram for explaining the state before and after switching of the flow table of the OFS 34.

  3 and 4, SA (Source Address) represents a source IP address, and DA (Destination Address) represents a destination IP address.

  In FIG. 2, the physical IF (Interface) on the VLAN 1 and 2 side in the OFS 34 is represented as # 1, the physical IF (client terminal side IF) on the PC 32 side is represented as # 2, and the physical IF on the OFC 33 side is represented as # 3.

  In FIG. 4, the flow table held by the OFS 34 is assumed to be in the upper state of FIG. 4 before the VLAN switching from the VLAN 1 to the VLAN 2.

  Referring to FIG. 2 and FIG. 3, when the OFS 34 receives the Fast packet from the VLAN 2, the OFS 34 refers to the Match Field of the flow table in the upper state of FIG. Search for an entry.

  The above entry does not exist before VLAN switching from VLAN1 to VLAN2. Here, the OFS 34 is set to transfer a flow that does not exist in the Match Field to the OFC 33. Therefore, the OFS 34 transfers the Fast packet received from the VLAN 2 to the OFC 33 in Step C1.

  Next, the OFC 33 determines whether or not the VLAN-tag value of the Fast packet is 2 in Step C2. Here, the VLAN-tag value of the Fast packet is 2.

  If the VLAN-tag value of the Fast packet is 2 in step C2, then the OFC 33 extracts the SA value of the Fast packet as the VM IP address in step C3, and the extracted value is “A”. And

Next, in step C4, the OFC 33 refers to the Match Field of the flow table in the upper state of FIG. 4 and searches for an entry that meets the following conditions.
(1) Physical IF = # 2, DA = A
(2) SA = A, VLAN = 1
Here, the entry (1) relates to the upstream flow rule, and the condition of the IP packet is “received from physical IF # 2 and DA is A”.

  The entry (2) relates to the downstream flow rule, and the condition of the IP packet is “received from VLAN 1 and SA is A”.

  Next, in step C5, the OFC 33 determines whether both the entries (1) and (2) exist. Here, both entries (1) and (2) exist in the flow table in the upper state of FIG.

  If both the entries (1) and (2) exist in step C5, next, in step C6, the OFC 33 instructs the OFS 34 to update the flow table using the OpenFlow protocol. In response to this, the OFS 34 switches the flow table.

  Specifically, in the entry (1), the OFS 34 switches the processing content described in the instruction from “transmission to VLAN 1” to “transmission to VLAN 2”. In the entry of (2), the condition of the packet described in the Match Field is switched from “received from VLAN1 to SA is A”, and “received from VLAN2 to SA to A”. This completes the VLAN switching from VLAN1 to VLAN2.

  Thereafter, the OFC 33 returns a Fast packet to the OFS 34 in Step C7.

  After VLAN switching, the flow table is in the lower side of FIG. 4 and there is an entry (2) including “reception from VLAN 2” in the condition of the IP packet in this flow table.

  Therefore, the OFS 34 transmits the Fast packet returned from the OFC 33 from the physical IF # 2 according to the processing content described in the instruction of the entry (2).

  As described above, in the conventional method, switching of the network accompanying VM migration between different LANs is performed by the control device 50.

  On the other hand, in this embodiment, the network switching device 35 on the client side autonomously detects the movement of the VM by the L3 learning function and switches the flow table without using the control device 50, thereby switching the network. Switching.

Therefore, the following effects can be obtained.
(A) Since the control device 50 becomes unnecessary, it is avoided that the entire network cannot be connected due to a failure of the control device 50. (B) The operator details the node, network, etc. It is not necessary to set data. (2) Second Embodiment FIG. 5 is a diagram illustrating the configuration and operation of a VM migration system according to this embodiment.

  Referring to FIG. 5, the VM migration system of this embodiment has the same configuration as the first embodiment, but the operation is different.

  The operation of moving the VM 13 on the server 12 of the data center 10 to the server 22 of the data center 20 in the VM migration system of this embodiment will be described below.

  First, steps D1 to D5 similar to steps B1 to B5 in FIG. 1 are performed.

  Here, when the VM 13 before movement is not executing processing, the VM 23 after movement has no process to be taken over from the VM 13 and is in a no-load state.

  In this case, the VM 23 only notifies (broadcasts) the ARP packet, the RARP packet, the GARP packet, and the like on the LAN in the data center 20 in step D6 after activation. Therefore, the Fast packet is not transmitted from the VM 23 toward the client side.

  Therefore, when the GW 21 receives the GARP packet from the VM 23 in step D6, the GW 21 generates an IP packet (dummy) for L3 learning based on the GARP packet in step D7, and sends it to the client side via the VLAN 2. Send. GARP is a protocol for detecting duplication of IP addresses and updating ARP tables of other servers.

  Therefore, in step D8, the network switching device 35 performs L3 learning based on the L3 learning IP packet (dummy) and automatically switches the flow table. The process in step D8 is the same as the process in step B7 in FIG.

  Here, the operation of the GW 21 will be described in detail.

  FIG. 6 is a diagram for explaining the operation of the GW 21.

  The L3 learning IP packet (dummy) may be either a multicast IP packet or a unicast IP packet. FIG. 6 shows an example of the latter case.

  Referring to FIG. 6, when the GW 21 receives an ARP packet from the VM 23 in step E1, the GW 21 checks in step E2 whether the source IP address and the destination IP address of the received ARP packet are the same. If they are the same, it is determined that the received ARP packet is a GARP packet.

  Next, in step E3, the GW 21 generates an IP packet (dummy) for L3 learning based on the GARP packet.

  At this time, the GW 21 sets the source IP address of the GARP packet as the source IP address of the IP packet (dummy) for L3 learning, and sets the destination IP address of the network node (not shown) on the client side. Set the IP address in the config.

  Thereafter, the GW 21 transmits an L3 learning IP packet (dummy) to the client side via the VLAN 2 in step E4.

  When a multicast IP packet is generated as the L3 learning IP packet (dummy), the setting of the destination IP address of the L3 learning IP packet (dummy) in FIG. 6 may be changed to multicast. good.

  The effect of this embodiment is the same as that of the first embodiment.

10 Data center 11 GW
12 servers 20 data centers 21 GW
22 Server 32 PC
33 OFC
34 OFS
35 Network switching device 40 VCenter server

Claims (2)

A first GW and a first server provided on the first LAN of the first data center;
A second GW and a second server provided on the second LAN of the second data center;
A network switching device provided on a third LAN on the client side, connected to the first GW via the first VPN, and connected to the second GW via the second VPN; Have
When the virtual machine on the first server moves to the second server during execution of processing for transmitting an IP packet to the client side on the first server, the virtual machine on the second server Take over the above process after starting with,
The second GW transmits an IP packet transmitted to the client side in the process after the virtual machine has taken over the process on the second server via the second VPN. and sent to,
The network switching device, the IP packet received via the second VPN, performs L3 learning based on the IP packet, showing a pre-Symbol the second VPN in the extracted flow information from the I P packet A second ID is included, and a search is made from the flow table for an entry that matches the entry in which the second ID of the flow information is replaced with the first ID indicating the first VPN. If it exists, it is determined that the virtual machine has moved from the first server to the second server, and the flow table describing the condition of the IP packet and the processing content of the IP packet is switched. Virtual machine migration system. A first GW and a first server on the first LAN of the first data center;
A second GW and a second server are provided on the second LAN of the second data center;
On the third LAN on the client side, there is provided a network switching device that is connected to the first GW via the first VPN and is connected to the second GW via the second VPN,
When the virtual machine on the first server moves to the second server during execution of processing for transmitting an IP packet to the client side on the first server, the virtual machine on the second server Take over the above process after starting with,
The second GW transmits an IP packet transmitted to the client side in the processing after the virtual machine has taken over the processing on the second server via the second VPN. and sent to,
It said network switching device, the IP packet received via the second VPN, performs L3 learning based on the IP packet, indicating the second VPN in the extracted flow information from the previous SL I P packet A second ID is included, and a search is made from the flow table for an entry that matches the entry in which the second ID of the flow information is replaced with the first ID indicating the first VPN. If it exists, it is determined that the virtual machine has moved from the first server to the second server, and the flow table describing the condition of the IP packet and the processing content of the IP packet is switched. To migrate virtual machines.