JP6259388B2 - Power control device, server virtualization system, and power control method - Google Patents

Description

  The present invention relates to a power control device, a server virtualization system, and a power control method.

In recent years, server virtualization systems that reduce the number of physical servers and improve resource usage efficiency by arranging and executing a plurality of virtual machines on one physical server have been studied. Furthermore, in the server virtualization system, operation cost reduction by power saving is also being studied.
Patent Document 1 describes that resources are allocated from a server device with a low resource usage rate, and that a physical server that is not used by a virtual machine is powered off.
Patent Document 2 describes that the aggregation efficiency of virtual machines is improved by suppressing an increase in physical servers as much as possible.

JP 2010-61278 A JP 2013-239095 A

In the operation policy to reduce power consumption as much as possible, excessive power control is performed on the server device, such as repeated power-on and power-off frequently, and server device hardware (power supply components, etc.) is consumed. Will also be intense. In other words, balanced power-saving control is required in actual operations to operate the server virtualization system stably without reducing the power consumption and without overloading the hardware of the server device. ing.
However, in the conventional technology, although a system optimized for individual evaluation indexes such as “perform power saving” is proposed, balanced power supply control is not proposed.

  Therefore, the main object of the present invention is to implement power control that stabilizes and operates the server virtualization system while reducing power consumption.

In order to solve the above problems, the power control device of the present invention includes an active state in which a virtual machine is placed on a physical server that is powered on, and the virtual machine is not placed in the physical server. A storage unit that stores a standby state in which at least some of the components are not energized as separate power states, and stores the power state of each physical server;
In accordance with the load amount of the virtual machine operating on the physical server, the resource amount of the active state is calculated with respect to the resource amount of the entire physical server, and the calculated state of the active state is calculated from the resource amount of the entire physical server. the resource amount of the preliminary state, except for the resource amount calculated in accordance with the resource amount of each power supply state of the calculation, Bei give a, a power state control part for updating the power state of each of the physical servers,
The power supply state control unit is
Sending a control signal to turn on the power to the physical server to be updated to the active state,
By energizing a part of the components to the physical server corresponding to the resource amount that is a predetermined ratio in the resource amount in the active state among the physical servers to be updated to the spare state. Send a control signal to shift to the standby state,
A control signal is transmitted to the physical server that is not shifted to the standby standby state among the physical servers that are updated to the standby state, by switching off the power supply. It is characterized by.

As a result, it is possible to appropriately reduce the power consumption of the physical server in the standby state, and to stabilize and operate the server virtualization system by reducing the number of times the power supply state is changed through simple power management of the active state or the standby state. .
By further classifying the standby state into a standby standby state that can quickly return to the active state, or a standby off state in which almost no power consumption occurs, the amount of power consumption can be reduced and the server virtualization system can be stabilized. Can be improved.

The power control apparatus according to the present invention includes an active state in which a virtual machine is arranged on a physical server that is powered on, and at least a part of the components of the physical server in which the virtual machine is not arranged. A storage means for storing a power state of each of the physical servers, with a standby state that is not set as a separate power state;
In accordance with the load amount of the virtual machine operating on the physical server, the resource amount of the active state is calculated with respect to the resource amount of the entire physical server, and the calculated state of the active state is calculated from the resource amount of the entire physical server. A power state control unit that calculates the resource amount of the spare state excluding the resource amount, and updates the power state of each physical server according to the calculated resource amount of each power state,
The power supply state control unit
When there are a plurality of working detailed states that are further subdivided from the working state and have different power control contents as the power states of the physical servers, the working detailed state is distributed according to a predetermined first distribution ratio. Decide
When a plurality of spare detailed states having different power control contents, which are further subdivided from the spare state, exist as the power states of the respective physical servers, the spare detailed state is distributed according to a predetermined second distribution ratio. Decide
A control signal for updating to the allocated active detailed state and a control signal for updating to the allocated spare detailed state are transmitted to each of the physical servers, respectively. To do.

As a result, it is possible to appropriately reduce the power consumption of the physical server in the standby state, and to stabilize and operate the server virtualization system by reducing the number of times the power supply state is changed through simple power management of the active state or the standby state. .
And even if there are a plurality of subdivided states corresponding to the implementation of various power saving modes, an appropriate implementation can be selected for the convenience of the vendor.

In the present invention, the power supply state control unit
Sending a control signal to turn on the power to the physical server to be updated to the active state,
A control signal is transmitted to the physical server to be updated to the standby state by causing a part of the components to be energized to shift to the standby state.

  As a result, the physical server that has been updated to the standby state can be quickly returned to the active state when a part of the physical server is energized, so that the time required for relocation of the virtual machine can be reduced.

In the present invention, the power supply state control unit
When calculating the amount of resources in the active state according to the load amount of the virtual machine, the amount of resources in the active state is increased by a predetermined rate from the amount of resource in the active state that can process the load amount of the virtual machine. Calculate
Sending a control signal to turn on the power to the physical server to be updated to the active state,
A control signal is transmitted to the physical server to be updated to the spare state to turn it to the spare off state by turning off the power.

  As a result, there is a margin in the amount of resources in the active state secured by a predetermined ratio, so that a certain amount of load increase does not cause a shortage of physical server resources and operates stably. Furthermore, the amount of reduction in power consumption can be increased by utilizing a standby OFF state in which power consumption hardly occurs.

In the present invention, the power supply state control unit
When calculating the amount of resources in the active state according to the load amount of the virtual machine, the amount of resources in the active state is increased by a predetermined rate from the amount of resource in the active state that can process the load amount of the virtual machine. Calculate
Sending a control signal to turn on the power to the physical server to be updated to the active state,
A control signal is transmitted to the physical server to be updated to the standby state by causing a part of the components to be energized to shift to the standby state.

  As a result, the amount of resources in the active state secured by a predetermined ratio is large, and even a physical server in the standby state can quickly return to the active state because some of its components are energized. The time required for elimination when a high load occurs can be shortened.

  According to the present invention, it is possible to perform power supply control that stabilizes and operates a server virtualization system while reducing power consumption.

It is a block diagram of the server virtualization system regarding one Embodiment of this invention. FIG. 2A is a diagram illustrating a transition relationship between power states of physical servers. FIG. 2B is a diagram showing details of each power supply state of the physical server. FIG. 2C is a configuration diagram of the power supply state management unit. FIG. 3A is a schematic diagram of four systems in which the power supply state control unit performs state management. FIG. 3B is a comparison table for the four systems shown in FIG. It is a flowchart which shows the management process of the power supply state which the power supply control apparatus regarding one Embodiment of this invention performs. FIG. 5A shows the initial state of method 1. FIG. 5B shows a state in which the load of method 1 is reduced. FIG. 5C shows a state in which the load of method 1 is increased. FIG. 5D shows a state in which the load increase in method 1 is dealt with. FIG. 6A shows an initial state of method 2. FIG. 6B shows a state in which the load of method 2 is reduced. FIG. 6C shows a state where the load of method 2 is increased. FIG. 6D shows a state in which the load increase in method 2 is dealt with. FIG. 7A shows the initial state of method 3. FIG. 7B shows a state in which the load of method 3 is reduced. FIG. 7C shows a state in which the load of method 3 is increased. FIG. 7D shows a state in which the load increase in method 3 is dealt with. FIG. 8A shows the initial state of method 4. FIG. 8B shows a state in which the load of method 4 is reduced. FIG. 8C shows a state where the load of method 4 is increased. FIG. 8D shows a state in which the load increase in method 4 is dealt with.

  Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a configuration diagram of a server virtualization system.
The server virtualization system is configured by connecting a data center 1 in which one or more physical servers 10 are arranged and a power supply control device 4 via a network. Each device (physical server 10, power supply control device 4) of these server virtualization systems is configured as a computer having a CPU (Central Processing Unit), a memory, a hard disk (storage means), and a network interface. However, each processing unit is operated by executing the program read on the memory.

Each physical server 10 in the data center 1 sets its own power state in accordance with a power control signal instructed from the power control device 4.
The power supply control device 4 arranges and executes a virtual machine (VM) 20 on each physical server 10 and determines the state of power supply control according to the load of the virtual machine 20.
Note that the number and processing capacity of the physical servers 10 are considered in advance at the design stage of the data center 1 so that the total load of the virtual machines 20 operating on the physical server 10 does not exceed the total processing capacity of the physical servers 10. ing.

Here, the physical server 10 in the data center 1 is classified into any one of the active on-server group 31, the standby standby server group 32, and the standby off-server group 33 according to its own power state.
First, the “working” state is a state in which the virtual machine 20 is disposed, and the “standby” state is a state in which the virtual machine 20 is not disposed.
The active on-server group 31 is a physical server 10 in an “active on state” in which the virtual machine 20 is placed and operating and in which the power is on (all the components of the physical server 10 are energized). Is a set of
The standby server group 32 is a set of physical servers 10 in a “standby standby state” in a standby state in which the virtual machine 20 is not placed and in which some of the components of the physical server 10 are energized.
The spare off-server group 33 is a set of physical servers 10 in a spare state in which the virtual machine 20 is not placed and the power is off (all the components of the physical server 10 are not energized). is there.

Further, each physical server 10 includes a resource information collection unit 11 that collects information indicating a resource usage state such as a load amount of its own device.
Examples of the load to be collected by the resource information collection unit 11 include resource usage rates of the physical server 10 such as a CPU usage rate and a memory usage rate.

The power supply control device 4 includes a VM placement destination determination unit 41, a VM relocation execution unit 42, a load amount reception unit 51, a power supply state management unit 52, and a power supply state control unit 53.
The load amount receiving unit 51 receives the resource usage status such as the load amount of the physical server 10 from the resource information collecting unit 11 of each physical server 10, and writes the reception result to the power state management unit 52.
The power state management unit 52 manages the load amount from the load amount receiving unit 51 and the resource amount (physical server specification value) mounted on each physical server 10 in the data center 1. The physical server specification value is, for example, the number of CPU cores mounted on the physical server 10, the clock frequency (GHz), or the like.
The power supply state control unit 53 determines the power supply state of each physical server 10 based on the information (load value, specification value) of the power supply state management unit 52. Then, the power state control unit 53 transmits a power control signal to the physical server 10 that changes the power state.

  The VM placement destination determination unit 41 relocation destination of the virtual machine 20 that needs to be relocated in response to a change in the load amount received by the load amount reception unit 51 or a change in the power supply state determined by the power supply state control unit 53. The physical server 10 is determined. For example, when the load on the physical server 10 deviates from a predetermined range (load lower limit threshold to load upper limit threshold), the VM placement destination determination unit 41 is deployed on the removed physical server 10. The virtual machine 20 is determined so as to be rearranged on another physical server 10.

  The VM relocation execution unit 42 transmits a relocation signal to the relocation source and relocation destination physical servers 10 in accordance with the relocation contents of the virtual machine 20 determined by the VM placement destination determination unit 41. Thus, the rearrangement of the virtual machine 20 is executed. The VM rearrangement execution unit 42 reflects the latest arrangement state of the virtual machine 20 on the power supply state management unit 52.

  FIG. 2A is a diagram illustrating a transition relationship between the power states of the physical server 10. It is possible to switch directly between power on and power off without waiting. It is possible to switch directly between power on and standby without powering off. On the other hand, it is necessary to switch between standby and power-off after power-on.

FIG. 2B is a diagram showing details of each power supply state of the physical server.
Each power supply state is described by its state name, hardware implementation (ACPI State), VM operation, state-time consumption, and consumption at the time of shifting to working-on.

  The active on state is implemented as “G0 (S0) -Working”, for example, and the virtual machine 20 in the active on state can operate. For this reason, the amount of electricity consumed during the working on state is larger than in other states. In ACPI State, the notation of Gn (n = numerical value) indicates a major classification, and the notation of Sn (n = numerical value) indicates a minor classification.

  The standby state is implemented, for example, as “G1 (S1-S4) -Sleeping or G2 (S5) -Soft-off”, and the operation of the virtual machine 20 in the standby state is not possible. In the standby standby state, the resources of the physical server 10 are partially powered on, so the power consumption in that state is smaller than the active on state, but the standby standby state shifts to the active on state. Some costs (electricity consumption and transition period) are required.

  The spare off state is implemented as “G3-Mechanical-off”, for example, and the operation of the virtual machine 20 in the spare off state is impossible. In the standby-off state, the power of the physical server 10 is off, so that the amount of electricity consumed in that state is almost zero, but the cost (electricity) for shifting from the standby-off state to the working-on state is sufficient. (Consumption and transition period) are greatly generated by restarting.

As described above, the power supply control device 4 roughly classifies the power supply state of the physical server 10 into three logical types (active on state, standby standby state, and standby off state).
On the other hand, the physical power supply state (specific hardware implementation such as ACPI State) of each physical server 10 is not limited to the total number of power supply states, but may be an OS (Operating System) vendor, a hardware vendor, or the like. Depending on the circumstances, there are various ways of calling (normal operation mode, energy saving mode, suspend mode, standby mode, etc.).

Therefore, as illustrated in FIG. 2B, the power supply control device 4 has a correspondence table between a logical power supply state (working on state, etc.) and a physical power supply state (G0 (S0) -Working, etc.). Is entered in advance by an administrator or the like. This correspondence table comprehensively defines the correspondence of each physical power supply state so that each physical power supply state corresponds to any one logical power supply state.
When the power supply state control unit 53 creates a control signal for setting the power supply state to the “preliminary off state” for a certain physical server 10, the power supply state control unit 53 refers to the power supply state correspondence table described above, What is necessary is just to transmit to the physical server 10, after converting (translating) into the control signal of setting a state to "G3-Mechanical-off".

Furthermore, in the correspondence table, one logical power supply state (for example, standby standby state) is a plurality of physical power supply states (for example, power supplies such as G1 (S1-S4) -Sleeping, G2 (S5) -Soft-off) In the case of corresponding to (control content), the power supply control device 4 also registers the ratio of the scale (predetermined distribution ratio) between a plurality of physical power supply states of the correspondence destination in advance in the correspondence table. And the power supply state control part 53 may determine the number of each physical power supply state according to the ratio of this scale.
As a result, the power supply state control unit 53 sets the predetermined first distribution ratio when a plurality of physical power supply states (working detailed states) obtained by further subdividing the active state exist as the power supply states of the physical servers 10. Accordingly, the distribution of the physical servers 10 in the active detailed state is determined.
Similarly, the power supply state control unit 53 sets the second distribution ratio to a predetermined second distribution ratio when a plurality of physical power supply states (preliminary detailed states) obtained by further subdividing the standby state exist as the power supply states of the physical servers 10. Accordingly, the distribution of the physical servers 10 in the preliminary detailed state is determined.
Then, the power supply state control unit 53 transmits a control signal for updating to each physical server 10 to the determined working detailed state or standby detailed state.
For example, there are four physical servers 10 to be updated to the standby standby state, and the standby detailed state “G1 (S1-S4) -Sleeping” in the standby standby state: the standby detailed state “G2 ( It is assumed that the second distribution ratio indicating that “S5) -Soft-off” = 1: 3 is set in advance.
At that time, according to the second distribution ratio, the power supply state control unit 53 sets one physical server 10 to be updated to “G1 (S1-S4) -Sleeping” as “G2 (S5) -Soft-off”. The number of physical servers 10 to be updated is determined as three.

FIG. 2C is a configuration diagram of the power supply state management unit 52.
The power state management unit 52 manages the power state of each physical server 10, the virtual machine 20 arranged, and the specifications of the physical server 10 in association with each other.
In the virtual machine 20 column in the power state management unit 52, in addition to the specific information (VM1) of the virtual machine 20, the load (load = 10) associated with the operation of the virtual machine 20 is also received from the load amount receiving unit 51. It is set based on the load. This load is, for example, the usage amount and usage rate of the components (CPU, memory, disk I / O, network I / O, etc.) of the physical server 10. Alternatively, the throughput value (communication amount or transaction amount per unit time) of the physical server 10 may be used as a load, for example, load = 10 [tps (Transactions Per Second)].
Note that the load value (rough unit) for each physical server 10 may be stored in the power state management unit 52 instead of the load value (fine unit) for each virtual machine 20. Thereby, measurement of a load value becomes easy.

In the specification column in the power supply state management unit 52, the processing capacity value for the load calculated from the physical server specification value is described. For example, when the CPU load of the physical server 10 is used, the processing capability value described in the specification column = CPU frequency × the number of cores.
For example, the physical server 10 of PM1 has two virtual machines 20 (load = 10 + 8 = 18) named VM1 and VM2, and has little room for its processing capacity (load 20 minutes). .
On the other hand, the physical server 10 of PM2 is provided with one virtual machine 20 (load = 5) called VM3, which has a margin for its processing capacity (load 20 minutes).

FIG. 2C illustrates a simple case where the processing capabilities of the six physical servers 10 are the same. In this case, calculating the processing capacity of the physical server 10 necessary for the active on state and calculating the number of physical servers 10 to be in the active on state are the same.
On the other hand, when there is a variation in processing capability among the physical servers 10 in the data center 1 (heterogeneous environment), the power supply state control unit 53 calculates the processing capability of the physical server 10 required for the active on state. The physical server 10 that provides the processing capability may be specified by referring to the specification column of the power supply state management unit 52.
In the following description, for the sake of clarity, a case where the processing capabilities of the physical servers 10 are the same is illustrated.

FIG. 3A is a schematic diagram of four methods in which the power supply state control unit 53 performs state management.
The power supply state control unit 53 employs any one of the four methods, Method 1 to Method 4, as power management for the same group (data center 1). The difference between these methods is the difference in the number (distribution) of physical servers 10 in each state (active on state, standby standby state, standby off state).
As shown below, in method 1, the logical power state of the physical server 10 is defined as three states, and in methods 2-4, the logical power state is defined as two states.

  In methods 1 and 2, the number of servers that are currently active is the minimum number of servers required for system load processing (normal size). However, in methods 3 and 4, there is a certain amount of margin than the normal size. It is assumed that the number of servers is large (for example, 1.5 times). The “scale” is the size of the resource amount of the entire physical server 10 and is described using the number of units as an example.

The number of standby standby units is set to a predetermined ratio (for example, half the small scale) in the method 1 in the method 1, and in the methods 2 and 4, the current on is determined from the number of all physical servers 10 in the data center 1. The remaining number of units minus the number of states. On the other hand, in the system 3, since the standby standby state is not used in the first place ("**" in the figure), the number is always zero.
In method 1, the number of standby standby units is appropriate for the number of active on states. Therefore, when the number of physical servers 10 in the active on state is insufficient, the prepared standby standby state physical units are available. The server 10 can be operated immediately. On the other hand, when the number of active on-states increases due to an increase in load, and half of the standby standby state cannot be secured, notification (screen display, alarm sound, etc.) to that effect (insufficient reserve) Also good.

In the system 1, the number of spare off states is the remaining number obtained by subtracting the number of active on states and the number of standby standby states from the number of all physical servers 10 in the data center 1. In method 3, the remaining number is obtained by subtracting the number of active servers from the number of all physical servers 10 in the data center 1. On the other hand, in the systems 2 and 4, since the standby standby state is not used in the first place ("**" in the figure), the number is always zero.
In method 1, since the power of the physical server 10 that does not correspond to the active on state or the standby standby state is turned off, the total power consumption of the entire data center 1 can be suppressed.
In method 3, since the power of the physical server 10 that does not correspond to the active on state is turned off, the total power consumption of the entire data center 1 can be suppressed.

The distribution of each power state shown in FIG. 3A is merely an example, and a multiplier for determining the scale (for example, a small scale is set to “½” of the active on state) The administrator may appropriately set a multiplier (for example, “2/3”).
In FIG. 3A, the number of servers is used as a unit for the distribution of each power state, but instead, the server specification total value may be used as a unit. Then, after calculating the server spec total value (for example, only “spec = 60” is required as the active on state), the power supply state control unit 53 determines the physical server 10 to satisfy the server spec total value. (For example, four physical servers 10 with specifications = 10, 10, 20, 20).
Furthermore, the power supply state control unit 53 has a plurality of power supply state switching targets based on a selection algorithm of power supply switching targets such that a physical server 10 with a small past power supply switching history is easily selected as a current switching target. The physical server 10 to be actually switched from the physical server 10 is determined.

  FIG. 3B is a comparison diagram of the four status management methods. The detailed description of FIG. 3B will be given after the detailed description of the four systems (FIGS. 4 to 8).

FIG. 4 is a flowchart showing power state management processing executed by the power supply control device 4. In the power supply control device 4, before the start of this flowchart, the administrator sets any one of the four methods shown in FIG. 3A as the method to be used this time.
In S11, the power supply state control unit 53 initializes the number of server groups in each power supply state according to the number distribution indicated by the method used this time. This initial setting is reflected from the power supply state control unit 53 to each physical server 10.
In S12, the VM relocation execution unit 42 initially arranges the virtual machine 20 in the active on-server group 31 set in S11. For example, a method of “allocating resources from a server device having a low resource usage rate” described in Patent Document 1 may be used as a method for determining the initial placement destination of each virtual machine 20. When each of the arranged virtual machines 20 operates on each physical server 10, the load amount is measured by the resource information collection unit 11 and stored in the power state management unit 52 via the load amount reception unit 51.

In S <b> 21, the power supply state control unit 53 determines whether or not the total load processed by the current on-server group 31 has decreased with respect to the number of the current on-server groups 31 already arranged. In other words, it is determined whether or not a surplus resource has occurred in the physical server 10 of the active on-server group 31.
This determination formula indicates that the surplus resource is determined if the number of the currently active on state physical servers 10 exceeds the number of the appropriate size as “current on” corresponding to the load of the current VM group shown in FIG. It may be determined that it has occurred, or it may be determined that surplus resources have occurred if the number of appropriate scales exceeds B (B> 1). If Yes in S21, the process proceeds to S22, and if No, the process proceeds to S31.

  In S22, the power supply state control unit 53 uses the current VM group load that has been reduced in S21 to the appropriate scale indicated by the “working on” column of each method shown in FIG. The number of on-server groups 31 is reduced. Then, when the number of working-on states is updated, the power supply state control unit 53 recalculates other power supply states (the standby standby state and the standby off state) in conjunction with each other.

In S23, the power supply state control unit 53 sets the active on-server group 31 remaining after being reduced in S22 as a VM relocation destination. For example, when the active on state is reduced from four to three, the virtual machine 20 arranged on the reduced one physical server 10 is placed on any of the remaining three physical servers 10. By rearranging, one physical server 10 is put into a spare state.
As a method for determining the relocation destination of the virtual machine 20, for example, as described in the document “Japanese Patent Application Laid-Open No. 2012-252602”, “a server whose resource usage rate exceeds the upper limit threshold value” In some cases, a method of “migrating a logical server to another server and migrating a logical server to another server when there is a server that is below the lower threshold” can be used.

  In S24, the power supply state control unit 53 updates the power supply state of each server group based on the latest number distribution determined in S22. For example, after all the virtual machines 20 on the one machine reduced in S23 have moved to another physical server 10, the current on state of the one virtual machine 20 is changed to the standby standby state.

In S <b> 31, the power supply state control unit 53 determines whether the total load processed by the current on-server group 31 has increased with respect to the number of the current on-server groups 31 already arranged. In other words, it is determined whether or not a resource shortage has occurred in the physical server 10 of the active on-server group 31.
The determination formula is as follows. When the number of the currently active on state physical servers 10 falls below the number of the appropriate size as “active on” corresponding to the load of the current VM group shown in FIG. It may be determined that it has occurred, or it may be determined that insufficient resources have occurred if the number of units of the appropriate scale is less than A (A> 1). If Yes in S31, the process proceeds to S32. If No, the process returns to S21.

  In S32, the power supply state control unit 53 uses the current VM group load increased in S31 to the current scale up to the appropriate scale indicated by the “working on” column of each method shown in FIG. The number of on-server groups 31 is increased. Then, when the number of working-on states is updated, the power supply state control unit 53 recalculates other power supply states (the standby standby state and the standby off state) in conjunction with each other.

  In S33, the power supply state control unit 53 updates the power supply state of each server group based on the latest distribution of the number of units determined in S32. For example, the physical server 10 increased in S32 is changed from the standby state to the active on state.

In S34, the power supply state control unit 53 sets the active on-server group 31 increased in S32 (newly in the active on state in S33) as the VM relocation destination. For example, when the active on state is increased from three to four, the virtual machines 20 on any of the three existing physical servers 10 are rearranged on the increased one physical server 10.
As a method for determining the relocation destination of the virtual machine 20, for example, the method described in the document “Japanese Patent Laid-Open No. 2012-252602” can be used as in S23.

  Hereinafter, the details of each process shown in FIG. 4 will be described with reference to FIGS. The method 1 (FIG. 5), the method 2 (FIG. 6), the method 3 (FIG. 7), and the method 4 (FIG. 8) will be described in this order.

FIG. 5 is a specific example of method 1.
FIG. 5A shows the power state of each server group when initially placed in S12. In an example in which a total of eight physical servers 10 exist in the data center 1, in method 1, the active on state is normal scale (4 units), the standby standby state is half (2 units), and the standby off state is the rest The number of units (2 units) (see the “method 1” line in FIG. 3A).
A description like reference numeral 101 described as “VM group” in FIGS. 5 to 8 indicates the total load of the virtual machine 20. The area denoted by reference numeral 101 indicates the magnitude of the load, and indicates on which physical server 10 the position of reference numeral 101 is arranged. For example, since reference numeral 101 is described on four physical servers 10, it can be seen that each of them operates on four active on-server groups 31.

FIG. 5B shows the power supply state of each server group when the VMs are rearranged as the load decreases in S23. The appropriate scale of the working on state is reduced from four to three. Then, the standby standby state remains half (1.5≈2 units) of the active on state, and the standby off state is the rest (8-3-2 = 3 units). Thereby, power saving can be realized by increasing the number of physical servers 10 in the standby-off state.
FIG. 5C shows the power supply state of each server group when an increase in load is detected in S31. The three working-on states are insufficient to handle the increased load.
FIG. 5D shows the power supply state of each server group when the VMs are rearranged as the load increases in S34. As the appropriate scale of the active on state has changed from three to five, the virtual machines 20 arranged on the three physical servers 10 have been relocated (load distribution) to the five active on states. . Furthermore, since the normal scale of the active on state has become five, the standby standby state is half that (2.5≈3), and the standby off state is the rest (8-5-3 = 0). Become.

FIG. 6 is a specific example of method 2.
FIG. 6A shows the power state of each server group when initially placed in S12. In the example in which a total of eight physical servers 10 exist in the data center 1, in the method 2, the active on state is the normal scale (4 units), and the standby state is the remaining number (4 units) (FIG. 3 ( (Refer to “Method 2” line in a)).
FIG. 6B shows the power supply state of each server group when the VMs are rearranged as the load decreases in S23. Since the appropriate scale of the active on state has decreased from four to three, power saving can be realized by shifting one of the units to the standby state.
FIG. 6C shows the power state of each server group when an increase in load is detected in S31. The three working-on states are insufficient to handle the increased load.
FIG. 6D shows the power supply state of each server group when the VMs are rearranged as the load increases in S34. As the appropriate scale of the active on state has changed from three to five, the virtual machines 20 arranged on the three physical servers 10 have been relocated (load distribution) to the five active on states. . Furthermore, since the normal scale of the active on state is five, the standby state is the remaining (8-5 = 3).

FIG. 7 is a specific example of method 3.
FIG. 7A shows the power supply state of each server group when initially placed in S12. In an example in which a total of eight physical servers 10 exist in the data center 1, in method 3, the active on state is a large scale (6 units) with sufficient processing capacity, and the standby off state is the remaining number (2 units). (See the “method 3” line in FIG. 3A).
FIG. 7B shows the power supply state of each server group when the VMs are rearranged as the load decreases in S23. Since the appropriate scale in the active on state has decreased from six to five, power saving can be realized by shifting the one unit to the standby off state.
FIG. 7C shows the power supply state of each server group when an increase in load is detected in S31. In FIG. 7A, the processing capacity in the active on state is given a margin, so that the load of the load increases to some extent, and the physical server 10 operates stably without running out of resources.
FIG. 7D shows the power supply state of each server group when the VMs are rearranged as the load increases in S34. As the appropriate scale of the active on state has changed from five to eight, the virtual machines 20 arranged on the five physical servers 10 have been rearranged (load distribution) to eight active on states. . Furthermore, since the normal scale of the active on state is eight, the spare off state is the remaining (8-8 = 0).

FIG. 8 is a specific example of method 4. FIG. 8 is obtained by replacing the “preliminary off state” in FIG. 7 with the “preliminary standby state” in FIG. As in method 3, method 4 also has a margin for the processing capacity in the active on state, so that it can operate stably without a shortage of resources of physical server 10 if the load increases to some extent.
In the above, it demonstrated according to the system with reference to FIGS.

Returning to FIG. 3B, a comparative diagram of the four status management methods will be described. The comparison of the four methods is a relative evaluation between the four methods, and compares whether a certain method is better or worse than the other methods. FIG. 3B shows the following four indicators and comprehensive evaluation (◎, ○, Δ in order of goodness) based on those indicators.
(1) Reduced power consumption (the higher the better)
(2) Number of power supply state changes (the smaller the better, the less ○ in the figure than in △)
(3) Number of occurrences of high server load (the smaller the better, the less ○ in the figure than the triangle)
(4) Time required for elimination when high load occurs (shorter is better)

(1) The power consumption reduction amount is an index indicating how much power consumption can be reduced compared to the operation in which all the physical servers 10 are always in the active-on state. As shown in FIG. 2B, the reduction amount increases in the order of the active on state, the standby standby state, and the standby off state (power consumption can be reduced).
Therefore, the reduction amount of method 3> the reduction amount of method 4 and the reduction amount of method 1> the reduction amount of method 2. Note that when the number of active on-states is sufficiently larger than the total number of physical servers 10, the reduction amount of method 2> the reduction amount of method 3.

(2) The number of power supply state changes is an index indicating the number of times of change (switching) between power supply states (active on state, standby standby state, standby off state). If the number of times of change increases, it may cause a failure due to overuse of the power supply unit of the physical server 10.
In each system of the present embodiment, since the transition is made between three power supply states (system 1) or two (systems 2 to 4), the number of power supply state changes can be reduced. Since the number of power supply states is smaller in methods 2 to 4 than method 1, the number of changes can be further reduced.

(3) The number of occurrences of server high load is the number of times a high load is generated for the physical server 10 in the active on state, for example, the number of times determined as Yes in S31 of FIG. That is, the high load is a state in which there is no room for adding the virtual machine 20 to the own device due to an increase in the resource usage rate on the physical server 10 in the active on state. The number of occurrences of server high load can be said to be the number of times the virtual machine 20 is rearranged in order to distribute the generated high load.
As described with reference to FIGS. 7C and 8C, since the methods 3 and 4 have a margin for the processing capacity in the active on state, the number of occurrences can be reduced compared to the methods 1 and 2. it can.

(4) The time required for elimination when a high load occurs indicates how long a high load state can be eliminated when a high load occurs in the physical server 10.
In methods 1 and 2, the time required to change the standby state to the active-on state (10 seconds) + the relocation time of the virtual machine 20 (15 seconds) = total 25 seconds is required.
In Method 3, a total of 5 minutes is required for the restart time for changing the standby-off state to the active-on state and the relocation time of the virtual machine 20.
In Method 4, it takes time (15 seconds) to relocate the virtual machine 20 to the physical server 10 in the active on state.

Furthermore, the power supply state control unit 53 may calculate one comprehensive evaluation value obtained by integrating the indices (1) to (4). That is, the power supply state control unit 53 obtains a comprehensive evaluation value by weighting and adding the weighting value of each index specified in advance to the evaluation value of each index. The weight value of each of these indices indicates the power control policy of the administrator.
For example, in order to realize a policy of “I do not want to cause server failure as much as possible because the power consumption may be increased to some extent”, (2) the weighting value of the power supply state change number is set to be higher than the weighting values of other indexes. A large value may be used.
In addition, in order to realize the policy of “requiring high performance quality for the application”, (3) a weighting value for the number of occurrences of server high load and (4) a weighting value for the time required for resolution when a high load occurs are The value may be larger than the weighting value of the index.

Further, the power supply state control unit 53 tries the power supply state management process shown in FIG. 4 by each method, and observes the result of the trial for each index of (1) to (4). Obtain an evaluation value. Then, the power supply state control unit 53 recommends (screen display) to the administrator, as a power supply control method suitable for the administrator's policy, a method having a high overall evaluation value among the tried methods 1 to 4. It may be adopted as a method used for future power control.
In this way, in the physical server power control, two states, the working state and the standby state, are defined, and a plurality of methods are proposed as the power state of each state (for example, a total of four variations in method 1 to method 4). From this method, a method with a high overall evaluation value reflecting the administrator's policy is adopted.
As a result, depending on how much importance the maintenance person places on each aspect of power consumption reduction, server fault tolerance, server overload frequency, and time required for server overload occurrence, an appropriate power supply A control method can be adopted.

The power control device 4 of the present embodiment described above uses the number of distributions between the power states (active on state, standby standby state, standby off state) of each physical server 10 in the data center 1 as the load of the virtual machine 20. Manage accordingly. That is, as shown in FIG. 3A, each method of power control includes an active on state in which an appropriate scale is determined according to the load of the virtual machine 20, a standby state that varies according to the scale of other states, and It is defined to transition in the preliminary off state.
As a result, as shown in FIG. 3B, power consumption is reduced, the number of power supply state changes is reduced, the number of server high load occurrences is reduced, and the time required for resolution when the high load occurs is reduced. Can be Therefore, the server virtualization system can be operated stably while reducing power consumption.

DESCRIPTION OF SYMBOLS 1 Data center 4 Power supply control apparatus 10 Physical server 11 Resource information collection part 20 Virtual machine 31 Current on-server group 32 Backup standby server group 33 Backup off-server group 41 VM placement destination determination part 42 VM rearrangement execution part 51 Load amount receiving part 52 Power State Management Unit 53 Power State Control Unit

Claims (8)

The active state in which a virtual machine is placed on a physical server that is powered on and the spare state in which the virtual machine is not placed and at least some of the components of the physical server are not energized are separated Storage means for storing the power state of each of the physical servers,
In accordance with the load amount of the virtual machine operating on the physical server, the resource amount of the active state is calculated with respect to the resource amount of the entire physical server, and the calculated state of the active state is calculated from the resource amount of the entire physical server. the resource amount of the preliminary state, except for the resource amount calculated in accordance with the resource amount of each power supply state of the calculation, Bei give a, a power state control part for updating the power state of each of the physical servers,
The power supply state control unit
Sending a control signal to turn on the power to the physical server to be updated to the active state,
By energizing a part of the components to the physical server corresponding to the resource amount that is a predetermined ratio in the resource amount in the active state among the physical servers to be updated to the spare state. Send a control signal to shift to the standby state,
A control signal is transmitted to the physical server that is not shifted to the standby standby state among the physical servers that are updated to the standby state, by switching off the power supply. A power supply control device. The active state in which a virtual machine is placed on a physical server that is powered on and the spare state in which the virtual machine is not placed and at least some of the components of the physical server are not energized are separated Storage means for storing the power state of each of the physical servers,
According to the load amount of the virtual machine operating on the physical server, the resource amount of the active state is calculated with respect to the resource amount of the entire physical server, and the calculated state of the active state is calculated from the resource amount of the entire physical server the resource amount of the preliminary state, except for the resource amount calculated in accordance with the resource amount of each power supply state of the calculation, Bei give a, a power state control part for updating the power state of each of the physical servers,
The power supply state control unit
When there are a plurality of working detailed states that are further subdivided from the working state and have different power control contents as the power states of the physical servers, the working detailed state is distributed according to a predetermined first distribution ratio. Decide
When a plurality of spare detailed states having different power control contents, which are further subdivided from the spare state, exist as the power states of the respective physical servers, the spare detailed state is distributed according to a predetermined second distribution ratio. Decide
A control signal for updating to the allocated active detailed state and a control signal for updating to the allocated spare detailed state are transmitted to each of the physical servers, respectively. Power control device. The power supply state control unit
Sending a control signal to turn on the power to the physical server to be updated to the active state,
A control signal is transmitted to the physical server to be updated to the standby state by energizing some of its components to shift to the standby state.
The power supply control device according to claim 2 . The power supply state control unit
When calculating the amount of resources in the active state according to the load amount of the virtual machine, the amount of resources in the active state is increased by a predetermined rate from the amount of resource in the active state that can process the load amount of the virtual machine. Calculate
Sending a control signal to turn on the power to the physical server to be updated to the active state,
A control signal is transmitted to the physical server to be updated to the spare state to turn it into the spare off state by turning off the power.
The power supply control device according to claim 2 . The power supply state control unit
When calculating the amount of resources in the active state according to the load amount of the virtual machine, the amount of resources in the active state is increased by a predetermined rate from the amount of resource in the active state that can process the load amount of the virtual machine. Calculate
Sending a control signal to turn on the power to the physical server to be updated to the active state,
A control signal is transmitted to the physical server to be updated to the standby state by energizing some of its components to shift to the standby state.
The power supply control device according to claim 2 . 6. A server virtualization system comprising: the power control apparatus according to claim 1 ; and each physical server connected via a network. The power supply control device includes storage means and a power supply state control unit,
The storage means has a working state in which a virtual machine is placed on a physical server that is powered on, and the virtual machine is not placed, and at least some of the components of the physical server are energized. A separate standby power state and a stored power state of each physical server,
The power supply state control unit calculates a resource amount in the active state with respect to a resource amount of the entire physical server according to a load amount of the virtual machine operating on the physical server, and calculates from the resource amount of the entire physical server When calculating the resource amount of the spare state excluding the calculated resource amount of the active state, and updating the power state of each physical server according to the calculated resource amount of each power state ,
Sending a control signal to turn on the power to the physical server to be updated to the active state,
By energizing a part of the components to the physical server corresponding to the resource amount that is a predetermined ratio in the resource amount in the active state among the physical servers to be updated to the spare state. Send a control signal to shift to the standby state,
A control signal is transmitted to the physical server that is not shifted to the standby standby state among the physical servers that are updated to the standby state, by switching off the power supply. A power control method. The power supply control device includes storage means and a power supply state control unit,
The storage means has a working state in which a virtual machine is placed on a physical server that is powered on, and the virtual machine is not placed, and at least some of the components of the physical server are energized. A separate standby power state and a stored power state of each physical server,
The power supply state control unit calculates a resource amount in the active state with respect to a resource amount of the entire physical server according to a load amount of the virtual machine operating on the physical server, and calculates from the resource amount of the entire physical server When calculating the resource amount of the spare state excluding the calculated resource amount of the active state, and updating the power state of each physical server according to the calculated resource amount of each power state ,
When there are a plurality of working detailed states that are further subdivided from the working state and have different power control contents as the power states of the physical servers, the working detailed state is distributed according to a predetermined first distribution ratio. Decide
When a plurality of spare detailed states having different power control contents, which are further subdivided from the spare state, exist as the power states of the respective physical servers, the spare detailed state is distributed according to a predetermined second distribution ratio. Decide
A control signal for updating to the allocated active detailed state and a control signal for updating to the allocated spare detailed state are transmitted to each of the physical servers, respectively. Yes Power control method.

Images