JP5505925B2 - Network operation management method and network operation management apparatus - Google Patents

Description

  The present invention relates to operation management of a network in which a plurality of servers are connected. Specifically, one server device (physical server) is logically connected to a plurality of server devices installed in a data center. The present invention relates to a network operation management method and a network operation management apparatus for a network to which a server group in which a virtualization technology for effectively utilizing limited physical resources is introduced by operating as a plurality of server devices (logical servers) .

  The traffic volume in the Internet is expected to increase rapidly in the future due to the increase in the network population and the spread of various communication services such as moving image distribution. Fig. 17 is an estimate of traffic volume and power consumption shown in the "Green IT Initiative Conference (1st)" held by the Minister of Economy, Trade and Industry. Estimated to be 190 times. In addition, as the amount of traffic increases, IT devices (network devices) such as servers and routers that process the traffic tend to increase, and the amount of power consumption will increase fivefold in 2025 compared to 2006. Estimated.

  Reducing the power consumption of these servers and network devices is an important issue in supporting the increase in traffic volume. In particular, there is an urgent need to save power in a data center where a plurality of servers and network devices are centrally installed.

Normally, an air conditioner for cooling a plurality of servers (server groups) is installed in a data center, and power consumption for driving the air conditioner is also considered in order to save power in the entire data center. There is a need to.
As an example of the breakdown of data center power consumption, 42% is related to air conditioners (Chiller + CRAC (Computer room air conditioning)), 30% is related to servers and network equipment (IT Equipment), It is known that 23% are related to power system equipment (PDU (Power Distributed Unit), UPS (Uninterruptible PowerSupply)), and 5% are humidifiers to prevent generation of lighting and static electricity.

Assume a data center to which server virtualization technology is applied. Here, we will briefly touch on virtualization technology.
The virtualization techniques shown in Non-Patent Document 1 and Non-Patent Document 2 are techniques for making a physical computer resource appear as a plurality of logical resources. Application of virtualization technology makes it possible to operate two logical servers (actually, two or more logical servers can be executed on a single physical server) on a single physical server. . Further, the logical server divides and uses the hardware resources (CPU, memory, network, hard disk) of the physical server (the divided hardware resources are allocated to the logical server).
Patent Document 1, Non-Patent Document 3 and Non-Patent Document 4 apply virtualization technology to allocate physical resources (CPU, memory, network bandwidth, hard disk) to a virtualized server (logical server). There has been proposed a technique for further reducing power consumption by dynamically changing according to the resource usage status and stopping the power supply of devices that are no longer needed.

The operation outline of the prior art (Patent Literature 1, Non-Patent Literature 3 and Non-Patent Literature 4) will be described with reference to FIG.
In the prior art, the usage status of physical resources is regularly monitored, and when the usage status of resources is low (the example on the left in FIG. 15), the allocation of physical resources to logical servers is dynamically reconfigured. All logical servers are executed on the physical server, and the physical server that is no longer necessary is paused (right example in FIG. 15).
In the example of FIG. 15, the logical server is reassigned to the physical server only in the server rack in which five physical servers are installed. However, in actuality, any physical server installed in the data center is used. Can be assigned a logical server.

JP2007-310791A

VMware: http://www.vmware.com/ Xen: http://www.xen.org/ Keisuke Hatazaki, Yoshifumi Takamoto, "Power-saving Policy Operation Technology for Systems Using Server Virtualization," IEICE Technical Report Computer System Study Group CPSY2006-44 Vol.106, No.436, pp. 37- 42, Dec. 2006. VMware Distributed Power Management (DPM)

In the above-described conventional technologies (Patent Literature 1, Non-Patent Literature 3 and Non-Patent Literature 4), the power consumption of the server group has been reduced by reducing the number of operating physical servers.
That is, according to the prior art, as shown in FIG. 16, in the data center in which the server 1, the network device 2, and the air conditioner 5 are installed, the minimum number of servers is used each time according to the usage status of the server group. The server group's power consumption has been minimized by dynamically changing the configuration of the server group to process the service request.
On the other hand, when processing is concentrated on some servers, the power consumption of the server increases in proportion to the usage rate, and the heat exhausted by that server increases with the power consumption. The exhaust heat of the server becomes larger, and it is necessary to increase the output of the air conditioner 5 in order to cool them (the arrow indicated by the diagonal line in FIG. 16 is cold air), and the power consumption of the air conditioner 5 is included. There is a concern that the power consumption of the entire data center may increase.

  That is, as in the above-described prior art (Patent Document 1, Non-Patent Document 3, and Non-Patent Document 4), paying attention only to the usage status of physical resources, assigning physical resources to logical servers, and further eliminating physical The technology that pauses the server and reduces the power consumption of the server group may lead to an increase in the power consumption of the air conditioner (CRAC) in some cases, resulting in an increase in the power consumption of the entire data center DC. There is.

  The present invention has been proposed in view of the above circumstances, and in order to reduce power consumption in the data center, when dynamically allocating physical resources to logical servers, in addition to server power consumption, An object of the present invention is to provide a network operation management method and a network operation management apparatus capable of reducing the power consumption of the entire data center by taking into consideration the power consumption of the air conditioner.

  In order to achieve the above object, the present invention provides a data center configuration (to a logical server of a physical resource) in a data center mainly comprising a server to which a virtualization technology is applied and an air conditioner for cooling the server. Management server to manage the allocation of air conditioners and the setting of air conditioners. And this management server considers the power consumption by the air conditioner according to the usage condition of the physical resource in order to minimize the total power consumption by the server group and the air conditioner among the power consumption of the data center. Change the dynamic allocation of the physical resource to the logical server.

In other words, the invention of claim 1 is provided with a plurality of physical servers to which a virtualization technology is applied in which a plurality of physical servers are installed in a data center and one physical server is logically operated as a plurality of logical servers. Each physical server is connected via a network and dynamically changes the allocation of physical resources to logical servers in accordance with the resource usage status of each physical server. In the network operation management method for cooling
Can be calculated from the physical server power consumption model that can be calculated from the hardware specifications such as the CPU product model, the air conditioner power consumption model that can be calculated from the product model of the air conditioner, and the physical location of the physical server and air conditioner The air flow model in the data center is managed in advance,
When changing the allocation of a physical resource to a logical server, the power consumption of each physical server after the allocation change and the total power consumption of all physical servers in the data center are determined from the power consumption model of each physical server. And the physical server power consumption model, the air conditioner power consumption model, the air flow model in the data center (the overall air flow in the data center, and the overall air flow). As for the ratio α of the air that flows into the intake side of the physical server without the air exhausted from each physical server circulating to the air conditioner, the exhaust surface and the air supply surface in different time zones, operating conditions in each physical server of temperature by detecting respectively, said sets of different temperatures in each physical server calculated by solving two or more collected created equations, the air From considering if α model), by estimate the power consumption of the air conditioner for cooling each physical server after allocation change, estimates the power consumption of the entire data center,
Depending on how the physical resources are used, calculate multiple combinations of the physical resources to be allocated to the logical servers, and assign the physical servers to which physical resources in the data center. Calculate the power consumption of the entire data center after the allocation change (the sum of the power consumption of the physical servers and the power consumption of the air conditioner) according to the above procedure, and the power consumption of the entire data center is the smallest. Select a combination
Based on the selection result, the allocation of physical resources to logical servers is dynamically changed.

According to a second aspect of the present invention, a plurality of physical servers to which a virtualization technology for installing a plurality of physical servers in a data center and operating a single physical server as a plurality of logical servers is applied via a network. A network that is connected and dynamically changes the allocation of physical resources to logical servers according to the resource usage status of each physical server, while cooling the physical servers with an air conditioner installed in a data center. The operation management apparatus is characterized by including the following configuration.

A physical server information management database that manages hardware specifications such as the product model of the CPU of the physical server and the resource amount when allocated to a logical server.
From the power consumption model of the physical server that can be calculated from hardware specifications such as the CPU product model, the power consumption model of the air conditioner that can be calculated from the product model of the air conditioner, and the physical location of the physical server and air conditioner A model of the total air flow in the data center that can be calculated, and the air exhausted from each physical server with respect to the total air flow without circulating to the air conditioner By detecting the temperature of the exhaust surface and the air supply surface of each physical server at different time zones, the temperature ratio α flowing into each physical server, two or more sets of different temperatures at each physical server are collected. Power consumption model management database that is calculated and managed by solving the created equations .
Provide a management interface for registering the power consumption model of the physical server or the physical server, the power consumption model of the air conditioner, and the model of the air flow in the data center in the physical server information management database and the power consumption model management database Department.
A resource usage status acquisition unit that periodically acquires the resource usage status of the physical server and stores it in the physical server information management database.
It consists of a physical server power consumption calculation unit for calculating the total power consumption of each physical server and an air conditioner power consumption calculation unit for calculating the power consumption of the air conditioner. Data center power consumption estimation section.
According to the usage status of each physical server, a plurality of combinations are calculated for the allocation of physical resources to logical servers, and the power consumption in each set is estimated by the data center power consumption calculation unit, and the power consumption is the smallest A physical server arrangement calculation unit for selecting.
A physical server configuration management unit that changes allocation of physical resources to logical servers in a data center based on an output result of the physical server allocation calculation unit, and stores a physical server relocation result in the physical server information management database.
A central control unit that controls each unit.

3. The network operation management apparatus according to claim 2, wherein the power consumption model management database is air that flows into an intake side of a plurality of physical servers including the physical server with respect to the plurality of physical servers. The ratio of air is managed by providing a table in which the ratio of air is registered in advance.

According to the present invention, the power consumption consumed by the physical server and the air conditioner for cooling them when performing the process of dynamically changing the allocation of the physical resource to the logical server according to the usage status of the physical server. Considering both power consumption, the power consumption of the entire data center is the smallest among the combinations of allocation of multiple physical resources to the logical server so that the power consumption of the entire data center is minimized. And the allocation of physical resources to logical servers is changed, and the entire air flow in the data center and the air exhausted from each physical server with respect to the entire air flow are air-conditioned. the proportion of air flowing to the intake side of the physical server without circulating the machine alpha, each physical server, different times, the operating status exhaust surface and the sheet By detecting the temperature of the surface, respectively, wherein by considering the proportion α of the air calculated by solving different temperature set two or more collected created equations in each physical server, the usage of the physical servers Compared with the prior art in which only the attention is paid and the assignment is changed, the power saving of the data center can be realized.

It is a block diagram which shows an example of the network operation management apparatus (management server) of this invention. It is a table figure which shows the example of the information which the physical server information management database of a network operation management apparatus manages. It is a table figure which shows the example of the information which the power consumption model management database of a network operation management apparatus manages. It is a model figure which shows the assumed environment and operation | movement outline | summary in the data center in which a network operation management apparatus is installed. It is a model figure which shows the operation | movement outline | summary of the physical server arrangement | positioning calculation part with respect to one server rack in the data center in which a network operation management apparatus is installed, a data center power consumption calculation part, and a physical server structure management part. It is a model figure which shows the relationship between an air conditioner, a server, and network equipment. It is a model figure for demonstrating the flow of the air in a data center. It is a model figure which shows the example of arrangement | positioning of the server rack in a data center. It is a model figure which shows the example of the exhaust efficiency (how much the exhaust heat from a physical server reaches to an air conditioner) corresponding to FIG. It is a model figure which shows the example of the air supply efficiency (how much the exhaust_gas | exhaustion from another physical server mixes with the cool air from an air conditioner) corresponding to FIG. It is a model figure which shows the flow of the air paying attention to one server in a data center. It is a graph which shows the power efficiency model of a certain air conditioner. It is a graph which shows the model of the power consumption of a server. It is a model figure which shows the assumed environment in the data center in which a network operation management apparatus is installed. It is a model figure which shows the operation | movement outline | summary in the technique which changes allocation to the logical server of the physical resource according to the utilization rate of the conventional physical server. It is a model figure which shows the relationship of each power consumption of an air conditioner, a server, and a network apparatus in a prior art. It is the graph which showed the estimation of traffic amount and power consumption.

Hereinafter, an example of an embodiment of a network operation management apparatus of the present invention will be described with reference to FIGS.
The network operation management apparatus of the present invention is sometimes used in a data center in which a plurality of servers and network devices and air conditioners for stable operation thereof are installed in order to minimize power consumption of the data center. According to the usage amount of the data center of the user changing every moment, the processing of dynamically changing the configuration of the data center is performed in consideration of both the power consumption of the server and the power consumption of the air conditioner.

  As shown in FIG. 4, the network operation management apparatus is configured so that the physical server 1 to which the virtualization technology is applied (logically operates as a plurality of logical servers 1 ′) and the physical server 1 for stable operation. This is realized by newly installing a management server 10 for managing the data center configuration (assignment of physical resources to logical servers) in the data center DC including the air conditioner 5 that cools the server 1 as a main component. . Each physical server 1 is provided with a temperature sensor 8 for measuring the temperature of the air exhausted from the physical server 1 and a temperature sensor 9 for measuring the temperature of the air supplied by each physical server 1.

For each physical server 1, an operation guarantee temperature for stable operation of the physical server 1 is set by a manufacturing company or the like that manufactures the physical server. In the data center DC, the temperature of the air output from the air conditioner 5 is determined so that the temperature of the air supplied by each physical server 1 is equal to or lower than the operation guarantee temperature. The air from the air conditioner 5 that has reached the air supply side of the physical server 1 has a high temperature in each physical server 1 because the power consumption of each physical server 1 becomes thermal energy, and is exhausted from the exhaust side. The exhausted air is circulated to the air conditioner 5, cooled by the air conditioner 5, and reaches each physical server 1 again. However, the air heated by the power consumption exhausted from each physical server 1 is circulated. A part (ratio α to the total air flow) flows into the air supply side of some physical servers 1 (the upper three physical servers 1 in the example of FIG. 4) without circulating to the air conditioner 5. The ratio α of the air exhausted from each physical server 1 to the air intake side of the physical server without being circulated to the air conditioner 5 is the size of the data center DC, the number of physical servers 1 and the air conditioners 5 It depends on the arrangement state.
When using the network operation management apparatus of the present invention, the operator registers in the management server 10 the air flow ratio α for all combinations of physical servers 1 in the data center DC. In addition, the physical server power consumption model that can be calculated from the hardware specifications such as the CPU product model and the air conditioner power consumption model that can be calculated from the product model of the air conditioner are added to the air flow rate α, and the management server 10 to register.

The air flow ratio (the ratio including the exhaust of other nodes or the node itself) α may be registered in the management server 10 as obtained by simulation by the operator at the time of designing the data center DC. The management server 10 may calculate based on environmental information obtained using the temperature sensors 8 and 9 provided for each server in the center DC.
For example, in the operation status of each node, the temperature of the exhaust surface and the supply surface is detected from the temperature sensor 8 and the temperature sensor 9, and the temperature sensor 8 and the temperature sensor 9 (value of each sensor) at the time when the operation status of each node is different. Shows different temperatures because the operating conditions are different), and create α and find α by solving.

  For the sake of simplicity, a data center DC composed of only two nodes will be described as an example. The temperature of the air supplied by the air conditioner is Tsup, the temperature of the temperature sensor 8 of the node 1 is T1_out, the temperature of the temperature sensor 8 of the node 2 is T2_out, and similarly the temperature of the temperature sensor 9 of the node 1 is T1_in and the temperature of the node 2 Is T2_in. Furthermore, when the ratio α of the exhaust from the node j included in the supply of the node i is denoted as αji, T1_in = (1− (α11 + α21)) × Tsup + α11 × T1_out + α21 × T2_out. By collecting two or more sets of different temperatures, the above equation can be solved to obtain α.

The management server 10 periodically uses the resource usage status of the logical server 1 ′ (how much physical resources such as CPU, memory, network, and disk allocated to each logical server are used) or from the operator. Collect on request.
Then, the management server 10 assigns which logical server 1 'to which physical server 1 so that the sum of the power consumption by the physical server 1 and the power consumption by the air conditioner (CRAC) 5 is minimized, The number of physical resources to be allocated to the server 1 ′ is calculated, and the allocation of the physical resources to the logical server in the data center DC is dynamically changed.

The management server 10 repeatedly executes these processes, so that the power consumption of the data center DC is minimized, so that the management server 10 optimizes the power consumption every time according to the service request amount from the user that changes from moment to moment. The configuration in the data center DC (which minimizes the consumption amount) is calculated, and the configuration is further changed.
In other words, the management server 10 allocates physical resources to the logical server according to the usage status of the logical server in order to minimize the power consumption of the server group and the air conditioner among the power consumption of the data center. It is configured to perform a process of dynamically changing.

A detailed configuration of the management server 10 realizing the network operation management apparatus will be described with reference to FIG.
The network operation management apparatus (management server) 10 includes a physical server information management database 11 that manages various information related to each physical server 1, and a power consumption model management database 12 that manages power consumption models of the physical server 1 and the air conditioner 5. A management interface providing unit 13 for an operator to input information, a physical server placement calculating unit 14 for obtaining a configuration candidate of a data center DC that minimizes power consumption, and a change in allocation of physical resources to logical servers (Physical server relocation) and a physical server configuration management unit 15 that manages the result of the relocation, a data center power consumption calculation unit 16 that calculates power consumption from the configuration candidates of the data center DC, and a resource usage status acquisition unit 17 The central control unit 18 is configured to control each unit.

  As shown in FIG. 2, the physical server information management database 11 includes the hardware specifications of a plurality of physical servers 1 and the amount of resources allocated when each physical server is allocated to a plurality of logical servers (usage rate of each logical server). Each information related to the physical server is managed. The hardware specifications of the physical server are information such as the product type of the CPU, for example. These pieces of information are input and stored by the operator from the management interface providing unit 13.

As shown in FIG. 3, the power consumption model management database 12 registers and manages a power consumption model of the physical server 1, a power consumption model of the air conditioner 5, and a model of air flow (air flow ratio α). ing. The power consumption model of the physical server 1 is a value of the power consumption with respect to the usage amount of the physical resource in each physical server, and is registered as a mathematical formula in each server.
The power consumption model of the physical server 1 can be, for example, the CPU usage rate (physical resource usage) that can be uniquely determined from the product type of the CPU and the power consumption at that time. The power consumption model of the air conditioner 5 is the power consumption with respect to the cooling capacity of the air conditioner.
The power consumption model of the air conditioner 5 can be uniquely determined from the product type of the air conditioner, for example. The model of the air flow (air flow rate α) is a ratio of the exhaust heat of each server flowing into the supply side of other servers, and is managed by a table for each server. The air flow model can be obtained by simulation by the operator at the time of designing the data center DC, for example. A specific example of the managed table will be described later.

  The contents managed in the power consumption model database 12 are registered by manually inputting in advance by the operator via the management interface unit 13 that the operator has obtained through simulation at the time of designing the data center DC. Further, if necessary, the network operation management device (management server) 10 may calculate based on information collected from the sensors by using the temperature sensors 8, 9 and the like arranged in the data center DC. Good.

The management interface providing unit 13 has a function as a user interface for an operator to register a physical server in the data center DC, its configuration, each power consumption model, and the like. For example, the operator connects to the management server 10 using a browser or the like installed on his / her PC and registers information related to each physical server 1. The registered information is stored in the physical server information management database 11.
The air flow in the data center DC registered in the power consumption model management database 12 is calculated and registered based on the information collected from the sensors by the management server 10 using the temperature sensors 8 and 9 and the like. Alternatively, the air flow obtained by simulation by the operator at the time of designing the data center DC may be manually registered in the power consumption model management database 12 via the direct management interface providing unit 13.
Information that the operator manually registers to estimate the power consumption of the data center DC includes the power consumption model of each physical server 1, the power consumption model of the air conditioner 5, and the air flow model in the data center DC. Conceivable.

The physical server arrangement calculation unit 14 obtains the configuration of the data center DC that minimizes the power consumption of the data center DC (how to allocate physical resources to the logical server). It is executed by an instruction from every operator or a periodic call from the central control unit 18.
As shown in FIG. 5, for example, as shown in FIG. 5, the usage status of each logical server (how much physical resources allocated to the logical server are used, Data resource power consumption calculation is performed by obtaining multiple combinations of allocation of physical resources to logical servers according to how much resources are being used) The calculation is performed for each combination by the unit 16, and as a result, the combination of the lowest power consumption is selected.

  The physical server configuration management unit 15 changes the configuration of the data center DC based on the output result of the physical server arrangement calculation unit 14. In the example of FIG. 5, regarding the allocation of physical resources to logical servers, the physical server placement calculation unit 14 calculates the power consumption of the data center DC for each of the four types of configurations, and as a result, the power consumption of the data center DC. The rightmost configuration (power consumption: 1465.7 W) in FIG. The physical server configuration management unit 15 changes the allocation of physical resources to logical servers according to the result of the physical server arrangement calculation unit 14. Further, in the physical server information management database 11, the physical server configuration management unit 15 stores the result of the assignment change in the physical server information management database 11 in order to manage the logical server to be executed for each physical server.

  The data center power consumption calculation unit 16 is called from the physical server relocation calculation unit and provides a function of calculating the power consumption of the data center DC at that time from the configuration candidates of the data center DC. The data center power consumption trial calculation unit 16 includes a physical server power consumption trial calculation unit 16a and an air conditioner power consumption trial calculation unit 16b, and the power consumption amount of the data center DC is obtained from the sum of the power consumptions calculated by these units. For trial calculation of each part, a model of air flow in the physical server, the air conditioner, and further the data center DC is acquired from the power consumption model management database 12.

  The resource usage status acquisition unit 17 periodically acquires the usage status of the resource of the logical server by an instruction from the operator or a periodic call from the central control unit. As described above, the usage status of the resources of each logical server changes from moment to moment according to the service request amount from the user. The usage status can also be obtained by using a management interface provided by the virtualization software. The acquired information on the usage status is stored in the physical server information management database 11, and the usage status of the physical resource by each logical server that changes every moment is reflected in the physical server information management database 11.

  The central control unit 18 controls each unit to realize the function of the power saving operation system of the data center DC.

  In the network operation management apparatus described above, the data center power consumption trial calculation unit 16 calculates the power consumption of the entire data center DC. Therefore, it is necessary to input the air flow from the air conditioner 5 to each server in the data center as data in the power consumption model management database 12. Hereinafter, how to grasp the flow of air from the air conditioner 5 to each server will be described with reference to FIGS.

  In FIG. 6, a service request from a user reaches the data center DC via a network such as the Internet. A network device 2 such as a router or a switch exists inside the data center DC, and transfers a service request from a user to an appropriate server 1. The server 1 that has received the service request returns a service response to the user. The amount of service requests from users changes from moment to moment, for example, during weekdays during daytime and during holiday nights.

  Further, the processing load of the network device group and the server group also changes every moment according to the service request amount. Each server 1 is operated by electric power. In addition, the power consumed by each server 1 is released into the data center DC as heat. Each server 1 has a temperature (operation guarantee temperature) for the server 1 to operate normally as a specification for each server 1. In order to operate the server 1 normally, the server 1 operates in the data center DC. It is necessary to supply air below the guaranteed temperature to the server 1. Therefore, the exhaust heat from each server is taken into the air conditioner (CRAC) 5 installed in the data center, the air containing the exhaust heat of each server taken in by the air conditioner (CRAC) 5 is cooled, and again Supplying to the server 1 is performed.

  In this case, as shown in FIG. 6, 100% of the exhaust heat from the server 1 is not taken into the air conditioner (CRAC) 5, and a part thereof is supplied to the server 1 (air conditioner (CRAC)). 5) and is taken into the server 1. As described above, since it is necessary to supply air below the temperature (operation guarantee temperature) at which each server 1 operates normally, exhaust heat from the server 1 is mixed in a large amount with the air supplied to the server. In this case, it is necessary to increase the output of the air conditioner (CRAC) 5 (the power consumption is also increased). In addition, the Chiller is a device that generates a heat source for cooling installed outside the data center DC, which is an outdoor unit compared to an ordinary air conditioner in the home. A heat source (for example, cooling water) for cooling the air conditioner (CRAC) 5 is provided.

Various methods for cooling the data center DC have been proposed by various vendors. As a structure generally employed in many data centers DC, for example, as shown in FIG. 7, a server in which air from an air conditioner (CRAC) 5 is stored in a server rack 3 from a floor 4 raised to the bottom. The exhaust heat is collected from the ceiling surface.
In addition, the ceiling cooling solution system that supplies air from the air conditioner (CRAC) 5 to the server from the ceiling of the data center and further exhausts from the ceiling, and the air conditioner (CRAC) 5 is arranged for each server rack 3 and the server A method of supplying and exhausting air in a rack has been proposed. In any system, exhaust heat is generated due to power consumption of the server, the air conditioner (CRAC) 5 takes in the exhaust heat, cools it, and supplies it to the server again. Mixing with the cooled air from CRAC) 5 (air flow in FIG. 6) remains the same.

  In a general data center DC, as shown in FIG. 7, a plurality of vertically long server racks 3 are installed, and a plurality of servers are stored in each server rack 3. The air cooled by the air conditioner (CRAC) 5 passes under the floor and is supplied from the air supply surface of the server rack 3 (in many cases, the front surface of the rack). Further, the heat generated by the power consumption of the server is exhausted from the exhaust surface of the server rack 3 (in many cases, the rear surface of the rack). The exhaust passes through the vicinity of the ceiling, is taken into the air conditioner (CRAC) 5, and the taken-in air is cooled by the air conditioner (CRAC) 5 and supplied again to the server rack 3 side.

Each server rack 3 prevents the exhaust heat of the server from being mixed with the air for supply cooled by the air conditioner 5, so that the supply surface of the server rack 3 is generally the supply air of the other server racks 3. The exhaust surface is installed so as to face the surface, and the exhaust surface of the other server rack 3 is faced.
A part of the exhaust heat from each server stored in the server rack 3 is mixed with the cooled air supplied from the air conditioner (CRAC) 5, and the mixing method is the server rack 3 or the cooled air. The position where the heat is supplied and the path through which the exhaust heat is taken into the air conditioner (CRAC) 5 become non-uniform for each data center DC. Therefore, the air flow in the data center DC differs for each data center DC.

As in the structure of the data center DC shown in FIG. 7, the flow of air supply and exhaust in the data center DC will be described by taking as an example the case where cold air is supplied from the floor and exhaust is taken in from the ceiling. An example will be described with reference to FIGS.
In the data center DC, as shown in FIG. 8, it is assumed that five servers (servers A to E) are installed in each of ten server racks. As described in FIG. 6, a part of the exhaust heat from the server is mixed with the air supply of itself or another server, the ratio is α, and the exhaust heat can be expressed as “α × Tout”. . The remainder is taken into the air conditioner (CRAC) 5 and the heat taken into the air conditioner is TACin, which can be expressed as “TACin = Tout− (α × Tout)”.
FIG. 9 shows an example of how much the exhaust of each server is taken into the air conditioner (CRAC) 5 (= (Tout− (α × Tout)) / Tout). 1.0 on the vertical axis indicates that all of the exhaust is taken into (CRAC) 5.

  FIG. 10 shows an example of the ratio (1-α) of how much other or own exhaust heat is mixed with each server air supply. In this example, the supply air of the servers (servers A, B, C) installed at physically low positions in the server rack is hardly mixed with exhaust heat from other or itself, and is physically high. It can be seen that the air supplied to the servers (servers D and E) installed in the station is mixed with a lot of other or own exhaust.

The exhaust efficiency of FIG. 9 and the air supply efficiency of FIG. 10 have different distributions depending on the assumed environment of the data center, but generally the air flow can be expressed as shown in FIG. In FIG. 11, the ratio of the exhaust from the server (Si) (i = 1 to 3) mixed with the supply air of the adjacent server (Sj) (j = 1 to 3) is expressed as αij.
The power consumption of the air conditioner (CRAC) 5 is obtained from the temperature (Tsup) of the supplied cool air and the amount of heat to be cooled. FIG. 12 shows the temperature (Tsup) of air supplied by an air conditioner (CRAC) 5 and its power efficiency (COP, Coefficient of Performance) obtained experimentally.

From FIG. 12, it can be seen that the higher the temperature of the supplied air, the better the power efficiency. For example, if you want to cool 10 kW of heat and set it to 15 ° C., COP = 2 = 0.068 * 15 * 15 + 0.0008 * 15 + 0.458, so the power consumption of the air conditioner (CRAC) 5 is 5 kW = 10 kW / 2 It becomes.
Further, when it is desired to cool the heat of 10 kw to 20 ° C., since COP = 3.338, the power consumption of the air conditioner (CRAC) 5 is 2.996 kw (the power consumption is reduced by 40%). On the other hand, the temperature (Tin) of the air finally reaching each server needs to be equal to or lower than the temperature (operation guarantee temperature) at which each server operates normally.
The power consumption of Ciller is closely related to the operation of the air conditioner (CRAC) 5, as shown in FIG. 6, and can be considered to be simply proportional to the power consumption of the air conditioner (CRAC) 5. it can.

It is known that the power consumption of a server has a characteristic that changes according to the usage rate of a physical resource such as a CPU of the server, as indicated by a solid line or a dotted line in FIG. Whether the characteristic is a solid line or a dotted line in FIG. 13 is determined by the product type of the CPU. In both the solid line and dotted line characteristics, it can be assumed that “the power consumption increases in proportion to the increase in the usage rate”. The solid line in FIG. 13 is expressed by the following equation (1). Such a power consumption model is generally applicable to all servers.
P = Pidle + (Pmax−Pidle) * n (Formula 1)
Pidle: Power consumption when server CPU usage is 0%
Pmax: Power consumption when the CPU usage rate of the server is 100% n: Server usage rate (0 to 1)

A specific example of the power consumption of the data center DC will be described with reference to FIG.
In FIG. 14, a data center DC in which five servers are installed is assumed. The air flow α in the data center DC is assumed to be a table between servers as shown in Table 1. The model of the air flow α in the data center DC shown in Table 1 is stored in the power consumption model management database 12, and these pieces of information are used to calculate the power consumption of the data center DC by the data center power consumption calculation unit 16 described above. Used when making trial calculations.
In Table 1, of the exhaust heat from server A, 5% is for itself, 2.5% is for server B, 10% is for server C, 25% is for server D, and 25% is for server E. The mixture is mixed and the rest is taken into the air conditioner (CRAC) 5.

  In addition, the power consumption of the server is calculated by the above-described equation 1, Pidle = 100 W (power consumption when the server usage rate is 0%), Pmax = 300 W (power when the server usage rate is 100%) (Consumption amount) (the actual high-density server consumes a larger amount of power, but here is a value for illustration only).

Assuming that the supply air temperature is 15 ° C., the exhaust temperature of each server is assumed to be 30 ° C. when 300 W (server usage rate 100%) is consumed, and 20 ° C. when 100 W (Pidle) is consumed.
Exhaust temperature = Supply air temperature + K * Server power consumption (Note) K is a coefficient that can be obtained from the density of air, the flow rate of air flowing through the server, and the specific heat of air. Here, 0.05 (constant) is used for simplicity.
Furthermore, the power consumption of the air conditioner (CRAC) uses the above-described power consumption model in FIG. The amount of heat to be cooled is the sum of the power consumption of the server for simplicity.
The temperature for the stable operation of the server is 25 ° C., and the air conditioner changes the output in 0.5 ° C. increments so that the supply air temperature of the running server is 25 ° C. or less.

When the above values are used and the usage rate of each server is the left example in FIG. 15, the power consumption of the entire data center DC (the sum of the power consumption of the server and the air conditioner (CRAC)) is
Air conditioner (CRAC) set temperature: 16.5 degrees Sum of server power consumption: 1090W
Sum of power consumption of air conditioner (CRAC): 469.3W
Power consumption of the entire data center DC: 1559.3W
It becomes.

Further, when the usage amount of each server is the right example in FIG. 15, the power consumption of the entire data center DC (the sum of the power consumption of the server and the air conditioner (CRAC)) is
Air conditioner (CRAC) set temperature: 16 degrees Sum of server power consumption: 1080W
Sum of power consumption of air conditioner (CRAC): 488.3W
Power consumption of the entire data center DC: 1568.3 W
It becomes.

According to the above-mentioned two examples, although the example from the left in FIG. 15 is changed to the right example, the total power consumption of the server is reduced from 1090 W to 1080 W by putting one server in a dormant state. However, the power consumption of the air conditioner (CRAC) has increased, and the power consumption of the entire data center DC has increased from 1559.3 W to 1568.3 W.
That is, paying attention only to the usage rate of the server and reducing the power consumption of the server group may increase the power consumption of the entire data center DC.

  On the other hand, according to the embodiment of the present invention described above, the data center power consumption trial calculation unit 16 is provided, and the data is calculated by the sum of the power consumption calculated by the physical server power consumption trial calculation unit 16a and the air conditioner power consumption trial calculation unit 16b. Since the power consumption of the center DC is obtained, the physical server arrangement calculation unit 14 selects the logical server configuration (combination of allocation of physical resources to the logical server) that minimizes the power consumption of the entire data center DC. can do.

  DESCRIPTION OF SYMBOLS 1 ... Physical server (server) 1 '... Logical server 2 ... Network equipment 3 ... Server rack 4 ... Floor surface 5 ... Air conditioner 6 ... Outdoor unit 8, 9 ... Temperature sensor 10 ... Management server ( Network operation management apparatus), 11 ... physical server information management database, 12 ... power consumption model management database, 13 ... management interface providing unit, 14 ... physical server arrangement calculation unit, 15 ... physical server configuration management unit, 16 ... data center power Consumption estimation part, 17 ... Resource utilization status acquisition part, 18 ... Central control part,

Claims (3)

A plurality of physical servers to which a virtualization technology is applied in which a plurality of physical servers are installed in a data center and one physical server is logically operated as a plurality of logical servers are connected via a network. In a network operation management method for dynamically changing the allocation of physical server resources to logical servers according to the usage status of physical server resources, while cooling each physical server with an air conditioner installed in a data center ,
Managing the hardware specifications of the physical server and the amount of resources allocated to the logical server;
The power consumption related to the physical server and the air conditioner is managed in advance, and the entire air flow in the data center and the air exhausted from each physical server with respect to the entire air flow are air-conditioned. By detecting the temperature of the exhaust surface and the air supply surface in different time zones and operating states at each physical server with respect to the ratio α of air flowing into the intake side of the physical server without circulating to the machine, the physical server It is calculated by collecting two or more sets of different temperatures in and solving the created equation, and managing in advance or dynamically,
While calculating the total amount of power consumption of each physical server and calculating the power consumption of the air conditioner for cooling each physical server, the power consumption of the entire data center is calculated,
According to the usage status of each physical server, a plurality of combinations are calculated for the allocation of the physical server resources to the logical servers, and the power consumption in each set is estimated by the data center power consumption calculation unit. Choose a small one,
A network operation management method characterized by dynamically changing allocation of physical server resources to logical servers in a data center based on the output result. A plurality of physical servers to which a virtualization technology is applied in which a plurality of physical servers are installed in a data center and one physical server is logically operated as a plurality of logical servers are connected via a network, A network operation management device that dynamically changes the allocation of physical server resources to logical servers according to the usage status of each server resource, and cools each physical server with an air conditioner installed in a data center. Because
A physical server information management database for managing the hardware specifications of the physical server and the resource amount allocated to the logical server;
The air consumption from the physical server and the air conditioner, the entire air flow in the data center, and the air exhausted from each physical server with respect to the entire air flow circulate to the air conditioner. As for the ratio α of the air flowing into the intake side of the physical server, the temperature of the exhaust surface and the supply surface in different time zones and operating states is detected in each physical server, so that a set of different temperatures in each physical server is detected. A power consumption model management database that calculates and manages two or more by collecting and creating an equation ,
A management interface providing unit for registering information on the physical server and the air conditioner in the physical server information management database and the power consumption model management database;
A resource usage status acquisition unit that periodically acquires the resource usage status of the physical server and stores the resource usage status in the physical server information management database;
The physical server power consumption calculation unit for calculating the total power consumption of each physical server and the air conditioner power consumption calculation unit for calculating the power consumption of the air conditioner for cooling each physical server. A data center power consumption trial calculation unit that calculates the power consumption of the entire data center by
According to the usage status of each physical server, a plurality of combinations are calculated for allocation of physical server resources to logical servers, and the power consumption of the entire data center in each set is estimated by the data center power consumption calculation unit, A physical server placement calculation unit that selects the one with the lowest power consumption;
Based on the output result of the physical server allocation calculation unit, the allocation of the physical server resource to the logical server in the data center is changed, and the physical server relocation (physical server resource allocation to the logical server) result is A physical server configuration management unit stored in the physical server information management database;
A network operation management apparatus comprising: a central control unit that controls each unit.   The power consumption model management database includes, for each of the plurality of physical servers, a table in which the ratio of air flowing into the intake side of each of the plurality of physical servers including the physical server is registered in advance. The network operation management apparatus according to claim 2, which manages the ratio α.