JP2012053899A - Operation management device and information processing system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce power consumption of servers performing parallel processing.SOLUTION: An operation management device 1010 comprises: a load history holding unit 1112 for holding a history of the total access number for an information processing device composed of a plurality of servers connected via a network; a load estimation unit 1134 for estimating the total access number occurring in the future by referring to the load history holding unit 1112; a state setting unit 1140 for setting at least one of the servers in a power saving state when the total access number estimated by the load estimation unit 1134 is less than an access threshold value; and an index calculation unit 100 for calculating indexes indicating operation conditions of the plurality of servers. The state setting unit 1140 selects servers to set in the power saving state based on the indexes calculated by the index calculation unit 100.

Description

  The present invention relates to an operation management apparatus and an information processing system.

  The problem of global warming has been heard recently. One measure against global warming is to reduce demand for electricity. Electricity is generally supplied by hydropower, solar power, wind power, geothermal power, and other natural power generation, nuclear power generation using nuclear power, and thermal power generation that uses gas, oil, coal, etc. as fuel to generate power. ing. Of these, thermal power generation is a major factor in promoting global warming. Decreasing power demand can reduce the amount of thermal power generation, so global warming can be suppressed accordingly.

  In the fields of IT (Information Technology) and ICT (Information and Communication Technology), the necessity of reducing the power consumed by computers, network devices, and data center devices has begun to be pointed out.

  With the spread of the Internet in recent years, many data centers have a network connection type, and perform processing in response to a job request from an external network such as the Internet. Some of these data centers employ a configuration including a plurality of servers arranged in parallel in order to process many accesses at once (see Patent Document 1).

As an operation management system for managing such a data center, Sente (registered trademark) provided by Nomura Research Institute, Ltd., OpenTP1001 (registered trademark) provided by Hitachi, Ltd., and a system provided by Fujitsu Ltd. There is a walker.
Further, a power consumption reduction device for a network connection device is known (see Patent Document 2).

JP 2008-225793 A JP 2007-97126 A

The Green Grid Association, [online], Internet <URL: http://www.thegreengrid.org>

  The current operation management system has a load balancer and distributes requests to individual servers even when the number of accesses is small. Many such load balancers distribute access evenly to the servers connected to it. In this case, access can be distributed in a manner that ignores the processing capacity and the degree of energy saving of each server.

  Regarding server characteristics, the processing capacity and the degree of energy saving often differ from server to server depending on the date of manufacture, type of equipment, manufacturer, equipment (memory, disk), power supply unit, and the like. For example, even when the same process is run on the same WEB (World Wide Web) server, the power consumed by the process may differ depending on the manufacturer. Therefore, the access distribution method in the current operation management system is not necessarily the optimum method from the viewpoint of energy saving.

  The present invention has been made in view of such problems, and an object thereof is to provide an operation management technique capable of reducing power consumption when a request is processed in parallel by a plurality of request processing units.

  One embodiment of the present invention relates to an operation management apparatus. This operation management device is generated in the future with reference to a load history holding unit for holding a load history of a request for an information processing device configured by connecting a plurality of request processing units via a network, and a load history holding unit. When the load predicted by the load predictor and the load predicted by the load predictor are smaller than the predetermined first threshold, at least one request processing unit is saved more power than the first state in which a request can be accepted. A state setting unit that sets the second state, and an index calculation unit that calculates an index indicating the operating status of the plurality of request processing units from information obtained from the network. The state setting unit selects a request processing unit to be set to the second state based on the index calculated by the index calculation unit.

The “request” may be a request for processing to the information processing apparatus, for example.
The “load” may be a value indicating the amount of request, for example.

  According to this aspect, the request processing unit to be set to the second state can be selected based on the index calculated by the index calculation unit.

  It should be noted that any combination of the above-described constituent elements, or those obtained by replacing the constituent elements and expressions of the present invention with each other between apparatuses, methods, systems, computer programs, recording media storing computer programs, and the like are also included in the present invention. It is effective as an embodiment of

  According to the present invention, it is possible to reduce power consumption when processing requests in parallel by a plurality of request processing units.

It is the schematic which shows an information processing system provided with the operation management apparatus which concerns on embodiment. It is explanatory drawing for demonstrating the flow of access in FIG. It is a block diagram which shows the function and structure of the load balancer of FIG. It is a data structure figure which shows the connection holding | maintenance part of FIG. It is a data structure figure which shows the 1st server group state holding | maintenance part of FIG. It is a block diagram which shows the function and structure of the operation management apparatus of FIG. It is a data structure figure which shows an example of the data hold | maintained at the communication information holding part of FIG. It is a data structure figure which shows an example of the data hold | maintained at the server information holding | maintenance part of FIG. It is a data structure figure which shows an example of the data hold | maintained at the 1st parameter | index holding | maintenance part of FIG. It is a data structure figure which shows an example of the data hold | maintained at the use condition holding | maintenance part of FIG. FIG. 7 is a data structure diagram illustrating an example of data held in a coefficient holding unit in FIG. 6. It is a data structure figure which shows an example of the data hold | maintained at the 2nd parameter | index holding | maintenance part of FIG. It is a data structure figure which shows an example of the data hold | maintained at the 3rd parameter | index holding | maintenance part of FIG. It is a data structure figure which shows the load log | history holding part of FIG. 15A to 15D are representative screen diagrams of the status screen. It is a typical screen figure of a warning screen. It is a typical screen figure of a state setting screen. It is a flowchart which shows a series of processes in the parameter | index calculation part of FIG. It is a chart which shows a series of processes in the learning part and state setting part of FIG. 6 along a time series. It is a data structure figure which shows an example of a load history holding | maintenance part in case the information processing system of FIG. 1 is used as a data center which provides the internet stock trading system of a securities company. FIG. 2 is a data structure diagram illustrating an example of a load history holding unit when the information processing system of FIG. 1 is used as a data center that provides a search service. It is a data structure figure which shows the load log | history holding part at the time of applying a certain embodiment to the shared data center of a university. It is a flowchart which shows a series of processes in the parameter | index calculation part which concerns on a modification.

  The present invention will be described below based on preferred embodiments with reference to the drawings. The same or equivalent components, members, and processes shown in the drawings are denoted by the same reference numerals, and repeated descriptions are omitted as appropriate.

  The operation management apparatus according to the embodiment is suitably used as a management server of an information processing system that processes access from a user with a plurality of servers arranged in parallel. The operation management apparatus predicts a load of a request coming from the network to the information processing system from past operation results, and when the predicted value is small, sleeps an unnecessary OS of the server group, or power supply itself of the server group Turn off. The power characteristics of the server group may vary depending on the model used, but the operation management apparatus measures the power characteristics of each server group by intercepting the network. As a result, for example, the OS can be put to sleep or the power can be turned off from a group of servers with low power efficiency, and the information processing system is comprehensively energy-saving.

  FIG. 1 is a schematic diagram showing an information processing system 2 including an operation management apparatus 1010 according to the embodiment. The information processing system 2 is a network connection type data center, for example, a data center that provides an Internet stock trading system of a securities company. The information processing system 2 includes an operation management device 1010, an information processing device 1020, and a load balancer 1030. The information processing apparatus 1020 includes a first server group 1022a, a second server group 1022b, a third server group 1022c, a fourth server group 1022d, and a fifth server group 1022e. The first server group 1022a includes a first front-end server 1024a, a first application server 1026a, and a first database server 1028a. The second server group 1022b, the third server group 1022c, the fourth server group 1022d, and the fifth server group 1022e each have a configuration equivalent to that of the first server group 1022a.

  Hereinafter, the first server group 1022a, the second server group 1022b, the third server group 1022c, the fourth server group 1022d, and the fifth server group 1022e may be collectively referred to as a server group 1022. The first front-end server 1024a, the second front-end server 1024b, the third front-end server 1024c, the fourth front-end server 1024d, and the fifth front-end server 1024e may be collectively referred to as a front-end server 1024. Further, the first application server 1026a, the second application server 1026b, the third application server 1026c, the fourth application server 1026d, and the fifth application server 1026e may be collectively referred to as an application server 1026. The first database server 1028a, the second database server 1028b, the third database server 1028c, the fourth database server 1028d, and the fifth database server 1028e may be collectively referred to as a database server.

The information processing system 2 is connected to the external network 1004 and receives a request from at least one user terminal 1006 that is also connected to the external network 1004. Here, the request is access from the user terminal 1006, for example. Access means that the user terminal 1006 and one server group 1022 of the information processing system 2 establish a connection and exchange a series of information. Accesses from different user terminals are different accesses, and different accesses can be made from the same user terminal 1006.
The external network 1004 is, for example, a local area network (LAN), a wide area network (WAN), or the Internet. The user terminal 1006 is a computer used by the user, such as a home desktop computer connected to the external network 1004 by wire or a laptop computer connected to the external network 1004 by wireless.

The load balancer 1030 is connected to the external network 1004 and is connected to each server group 1022 included in the information processing apparatus 1020 via the internal network 3. The load balancer 1030 is located between the information processing apparatus 1020 and the external network 1004 from the viewpoint of data flow, and mediates access to the information processing apparatus 1020 from the external network 1004.
The load balancer 1030 further assigns access from the user to a server group set in a state where access can be accepted (hereinafter referred to as an operating state) among the plurality of server groups 1022 included in the information processing apparatus 1020. Note that the operating state and the power saving state are basically defined as the state of the server group 1022.

  The operation management apparatus 1010 is a management server that manages the information processing apparatus 1020 and the load balancer 1030. In addition to the load prediction function described below, the operation management apparatus 1010 is equipped with a function for server operation management similar to the first (registered trademark) provided by Nomura Research Institute, Ltd., for example.

The operation management apparatus 1010 intercepts packets passing through the external network 1004, the load balancer 1030, and the internal network 3. The operation management apparatus 1010 also intercepts packets exchanged inside each server group 1022. This is realized, for example, by connecting the analysis apparatus 100 to a subnet that connects the front-end server 1024 and the application server 1026 or a subnet that connects the application server 1026 and the database server 1028. The internal network 3 includes these subnets that interconnect the servers of each server group 1022.
“Intercepting” may be acquiring the contents of the packet to be intercepted without annihilating it, for example, acquiring the packet regardless of the destination IP address included in the packet.
The operation management apparatus 1010 accumulates the contents of intercepted packets and sorts them by IP address, thereby deriving a work breakdown in the server or load balancer 1030 corresponding to each IP address.

  The operation management apparatus 1010 acquires and accumulates measured values of power supplied to each server from, for example, a bus bar type table tap PDU that supplies power to each server. The operation management apparatus 1010 calculates an index indicating the operating status of the server group 1022, for example, the power consumption per unit effective work amount in the server group 1022, from the accumulated measured power value and the breakdown of the work.

  The operation management apparatus 1010 calculates and records a load due to access from the external network 1004 to the information processing apparatus 1020, while predicting a load that will occur in the future based on a load requested in the past. When the predicted load becomes smaller than a predetermined value, the operation management apparatus 1010 is unnecessary for access processing among the server groups in the operating state based on the index indicating the calculated operating state of the server group 1022. A server group is selected, and the selected server group is set to a power saving state that is more power saving than the operating state.

Here, the load is a value representing the amount of work performed by the server group 1022, for example, and is determined based on the number of accesses per unit time. For example, the load may be a value having a mathematical relationship such as a proportional relationship with the number of accesses per unit time. The load may be a value obtained by imposing an upper limit value, a lower limit value, or both on the number of accesses per unit time. Hereinafter, a case where the load is simply the number of accesses per unit time (hereinafter referred to as the number of accesses) will be described.
If the load is equal to or greater than a predetermined value, the operation management apparatus 1010 sets all the server groups 1022 to the operating state.

The information processing apparatus 1020 processes access from the external network 1004. The information processing apparatus 1020 includes a plurality of server groups 1022, and each of the server groups 1022 functions as a request processing unit which is an access processing unit.
In the present embodiment, a case is considered in which communication is performed according to the TCP (Transmission Control Protocol) / IP (Internet Protocol) protocol.

  The server group 1022 is a so-called three-level server group, and one set of these constitutes an access processing unit. The front-end server 1024 is also called a WEB server, and a service that provides display of objects such as HTML (HyperText Markup Language) and images to the WEB browser of the user terminal 1006 operates in accordance with HTTP (HyperText Transfer Protocol). Server computer. The application server 1026 is a server computer that implements functions related to applications such as Java Servlet (Java Servlet, Java is a registered trademark) processing from the front-end server 1024. The database server 1028 is a server computer that stores data used by applications of the application server 1026. The front-end server 1024, the application server 1026, and the database server 1028 are realized using a known information processing technique. In the present embodiment, the front-end server 1024, the application server 1026, and the database server 1028 are separate servers, and are connected in series in this order.

  The front end server 1024 of the server group 1022 is connected to the operation management apparatus 1010 and the load balancer 1030 via the internal network 3. Each server group 1022 is managed by the operation management apparatus 1010.

  The servers included in each server group 1022 are managed by the OS individually or as a whole. The operating state of each server group is a state in which access from a user can be processed immediately. This state includes a state in which the server group 1022 can process a new access while processing an access from the user, and an idle state in which there is no access from the user and no work is generated. In the operating state, even in the idle state, at least the basic OS function and the network monitoring task are operating, and power is consumed accordingly. From the experience of the present inventor as a person skilled in the art, the power consumption in the idle state is in the range of approximately 30% to 70% during operation at the peak access number depending on the type of server device. In particular, the latest blade server consumes a large amount of power even in an idle state because of its high-density circuit configuration and many semiconductor elements installed in a narrow space. Some blade servers consume 70% of electric power during operation at the peak number of accesses in an idle state. When standard server equipment is used, it is about 60%. Here, the peak number of accesses is an upper limit value of the range of the number of accesses that the server group 1022 can operate without reducing its processing speed, and is predetermined for each server group 1022 based on its specifications. Here, the processing speed is a processing speed of access from the user in the server group 1022.

  The power saving state of each server group 1022 is a state where access from a user cannot be processed immediately. This state includes an OS sleep state in which power is supplied to the server group 1022 but the server group 1022 cannot process access from the user. When the server group 1022 receives the sleep introduction signal from the operation management apparatus 1010, the server group 1022 enters the OS sleep state. In this OS dormant state, the OS of the server group 1022 is hibernated and does not accept any access from the user. When the server group 1022 in the OS sleep state receives the sleep release signal from the operation management apparatus 1010, it returns to the operating state. It usually takes several seconds to several tens of seconds to return from the OS sleep state to the operating state. Further, from the experience of the present inventor as a person skilled in the art, the power consumption of the server group 1022 in the OS sleep state is approximately 5% to 10% during operation at the peak access number.

The power saving state of the server group 1022 further includes a power off state in which power to the server group 1022 is cut off. When the server group 1022 receives a power-off signal from the operation management apparatus 1010, the server group 1022 enters a power-off state. When the server group 1022 receives a power-on signal such as a WOL (Wake Up on LAN) signal from the operation management apparatus 1010, the server group 1022 returns to an operating state. It usually takes several minutes to return from the power-off state to the operating state.
In the present embodiment, the server group 1022 is set in the power saving state in consideration of the time taken for the server group 1022 to return from the power saving state to the operating state (hereinafter referred to as overhead) and the power consumption reduced in the power saving state. The number of server groups to be determined is determined. In other words, if only the server group as many as necessary to process the current number of accesses is operating, there is a possibility that a sudden increase in the number of accesses cannot be handled. On the other hand, the power consumption derived from standby power increases as the number of active server groups increases. Therefore, the number of server groups to be placed in the power saving state is determined so that their influences are antagonized.

  FIG. 2 is an explanatory diagram for explaining the flow of access in FIG. Hereinafter, a case where at least one packet is exchanged between the user terminal 1006 and a server group to which access is assigned in one access will be described. This packet includes a start point IP address that is a source IP address, an end point IP address that is a destination IP address, a sequence number, a port number, and a time stamp. In many cases, a plurality of packets are transferred between the user terminal 1006 and the server group for one access. In FIG. 2, it is assumed that the third server group 1022c is assigned to access. The IP address of the user terminal 1006 is “175.34.111.21”, the IP address of the load balancer 1030 is “100.10.10.10”, and the IP of the third front-end server 1024c included in the third server group 1022c. The address is “121.21.15.3”.

  The load balancer 1030 functions as a virtual server for the user terminal 1006. That is, in the external network 1004, when a user tries to access an information resource included in the information processing apparatus 1020, the URL (Uniform Resource Locator) of the information resource is the IP address “100.10.10.10” of the load balancer 1030. Is set to resolve names.

First, the user specifies the URL of the information resource that the information processing apparatus 1020 has to the web browser of the user terminal 1006. A first packet P1 having a start point IP address “175.34.11.21” and an end point IP address “100.10.10.10” is generated by the web browser of the user terminal 1006 and sent to the external network 1004. . The load balancer 1030 receives the first packet P1, selects the third server group 1022c from the server groups in the operating state, and allocates this access. The load balancer 1030 sends the second packet P2 with the end point IP address “121.21.15.3” to the internal network 3 while keeping the start point IP address of the first packet P1. The third front-end server 1024c of the third server group 1022c receives the second packet P2 addressed to itself, and performs processing based on the port number of the second packet P2 and data loaded in the packet. Information to be returned to the user terminal 1006 as a result of the processing is included in the third packet P3 with the start point IP address “121.21.15.3” and the end point IP address “1755.34.111.21”. The data is transmitted from the third front-end server 1024c to the internal network 3. The load balancer 1030 receives the third packet P3. The load balancer 1030 sends the fourth packet P4 having the start point IP address “100.10.10.10” to the external network 1004 while keeping the end point IP address of the third packet P3. The user terminal 1006 receives the fourth packet P4 addressed to itself from the external network 1004.
Hereinafter, the packet (first packet P1) sent from the user terminal 1006 to the load balancer 1030 is generically referred to as a packet, and the packet (second packet P2) sent from the load balancer 1030 to the information processing apparatus 1020 is generically assigned packet. That's it.

  FIG. 3 is a block diagram showing the function and configuration of the load balancer 1030. Each block shown here can be realized by hardware such as a computer (CPU) (central processing unit) and other elements and mechanical devices, and software can be realized by a computer program or the like. Here, The functional block realized by those cooperation is drawn. Therefore, it is understood by those skilled in the art who have touched this specification that these functional blocks can be realized in various forms by a combination of hardware and software.

The load balancer 1030 includes a first server group state holding unit 1322, a connection holding unit 1324, a request acquisition unit 1034, and a request allocation unit 1036.
The first server group state holding unit 1322 is a table that stores the state of the server group 1022 that would be currently set. Details of the first server group state holding unit 1322 will be described later.

  The connection holding unit 1324 is a table for guaranteeing that packets within the same access are sent to the same server group (hereinafter referred to as access identity). FIG. 4 is a data structure diagram showing the connection holding unit 1324. In the connection holding unit 1324, an entry 1118 corresponding to access is generated by a connection update unit 1368 described later. One entry corresponds to one access. The entry 1118 of the connection holding unit 1324 is assigned to the access by the user terminal IP address 1210 that is the IP address of the user terminal 1006 that has accessed, the load balancer IP address 1212 that is the IP address of the load balancer 1030, and the load balancer 1030. The assigned server group IP address 1214, which is the IP address of the front end server of the server group, and the sequence number 1216 for determining the difference in access when the user terminal 1006 is the same. Different sequence numbers are assigned to different accesses in the same user terminal. Hereinafter, the IP address of the front end server of the server group is simply referred to as the IP address of the server group.

  Returning to FIG. The request acquisition unit 1034 is connected to the external network 1004. The request acquisition unit 1034 acquires an access outbound packet from the user terminal 1006 that arrives from the external network 1004. At this time, the request acquisition unit 1034 determines whether the packet is addressed to itself based on the end point IP address included in the outgoing packet. The request acquisition unit 1034 passes the acquired outbound packet to the request allocation unit 1036.

  In this specification, “pass” means that processing for an information element is transferred from a certain functional block to a certain functional block. Speaking between the request acquisition unit 1034 and the request allocation unit 1036, for example, the request acquisition unit 1034 has a temporary memory (not shown) and stores the acquired outgoing packets in the request allocation unit 1036. The packet is appropriately transmitted in response to the request from the temporary memory and transmitted to the request allocation unit 1036 from the temporary memory. In addition, to pass, the first storage device 1032 has a storage area (not shown), the request acquisition unit 1034 writes the acquired outbound packet to the storage area, and the request allocation unit 1036 appropriately transmits the required outbound packet from the storage area. It may be read and processed.

  The request allocation unit 1036 allocates an outgoing packet from the user terminal 1006 to one of the server groups in the operating state. Particularly in the slope distribution mode, when the access from the user terminal 1006 is a new access, the request allocation unit 1036 allocates the new access to one of the server groups based on the index acquired from the operation management apparatus 1010. . The request allocation unit 1036 includes an identical connection determination unit 1362, a server group selection unit 1364, an address conversion unit 1366, and a connection update unit 1368.

  The same connection determination unit 1362 acquires the outgoing packet from the request acquisition unit 1034 and determines whether the outgoing packet is due to a new access. The same connection determination unit 1362 reads the start point IP address and sequence number of the acquired outbound packet. The same connection determination unit 1362 searches the entry of the connection holding unit 1324 using the read start point IP address and sequence number as a key, and if there is an entry that matches these, the assigned server group included in the entry The assigned server group IP address is acquired. The same connection determination unit 1362 passes the acquired allocation server group IP address and the outgoing packet to the address conversion unit 1366.

  If there is such a matching entry, the outgoing packet is one packet in the access assigned to the existing server group 1022 (the server group 1022 specified by the assigned server group IP address of the entry). It is. The address conversion unit 1366 converts the end point IP address of the passed packet to the assigned server group IP address acquired from the connection holding unit 1324 by the same connection determination unit 1362. The outgoing packet with the end point IP address converted in this way is sent from the address conversion unit 1366 to the internal network 3 as an assigned packet.

  If there is no matching entry, the identical connection determination unit 1362 passes the outgoing packet to the server group selection unit 1364. In this case, it can be said that the same connection determination unit 1362 has detected a new access.

  When the server group selection unit 1364 receives the outgoing packet corresponding to the new access from the same connection determination unit 1362, the server group selection unit 1364 refers to the first server group state holding unit 1322 and selects the server group 1022 to process the access. The first server group state holding unit 1322 is updated by the operation management apparatus 1010 based on the prediction of the number of accesses in the operation management apparatus 1010 and the calculated index.

  FIG. 5 is a data structure diagram showing the first server group state holding unit 1322. The first server group state holding unit 1322 holds the IP address 1202 of the server group 1022, the state 1204 of the server group 1022, the operation rate 1206 of the server group 1022, and the power cost 1207 of the server group 1022 in association with each other. To do.

  The operation rate 1206 of the server group 1022 indicates the ratio of the number of accesses that the server group 1022 is currently processing to the peak access number of the server group 1022 in% units. This operation rate is calculated and updated by an operation rate update unit (not shown) from a predetermined peak access number of the server group 1022 and the access number currently assigned to the server group 1022 known from the connection holding unit 1324. May be. Alternatively, an operating rate update unit (not shown) may acquire the operating rate from the server group 1022 in the operating state and update the operating rate of the first server group state holding unit 1322. Alternatively, the operation rate of the first server group state holding unit 1322 may be updated by the operation management apparatus 1010.

A power cost 1207 of the server group 1022 represents power consumption with respect to a unit work amount for WEB communication. The unit work amount may be, for example, a unit data amount or one access. When the power cost of the server group 1022 represents the power consumption with respect to the unit data amount for WEB communication, this power cost of 1 (W / GB) means that 1 GB of WEB communication data is processed in the WEB server. 1W is consumed. Considering the amount of data for WEB communication as a load on the WEB server, the power cost indicates the power consumed by the WEB server when the same load (unit data amount) is applied to each WEB server.
The value of the power cost of the first server group state holding unit 1322 is updated by the operation management apparatus 1010.

  Returning to FIG. The server group selection unit 1364 extracts a server group in an operating state with reference to the state of the server group from the server groups registered in the first server group state holding unit 1322. A server that causes the server group selection unit 1364 to process a new access depending on whether the request allocation unit 1036 is set to the power saving mode, the normal mode, or the slope distribution mode. The algorithm for selecting a group is different. Each case will be described below.

1. Power Saving Mode In the power saving mode, the server group selection unit 1364 selects a server group for processing a new access from the server group in the operating state so that the operating rate of the server group in the operating state is 100%. . For example, in the example of FIG. 5, the server group selection unit 1364 selects the third server group 1022 c as a server group for processing a new access. Further, for example, the first server group 1022a, the second server group 1022b, and the third server group 1022c are set in an operating state, the operating rate of the first server group 1022a is 100%, and the operating rate of the second server group 1022b is When 80% and the availability factor of the third server group 1022c is 0%, the server group selection unit 1364 selects the second server group 1022b as a server group for processing a new access.

2. Normal Mode In the normal mode, the server group selection unit 1364 selects an optimal server group for processing a new access according to a load distribution algorithm set in advance by the administrator of the information processing system 2. The load distribution algorithm used here is the round robin method in which the server group 1022 is selected in order, the minimum connection method in which the server group with the smallest number of accesses being processed is selected, or the earliest response. This is a known algorithm such as a fastest method for selecting a server group.

3. Inclined distribution mode In the inclined distribution mode, the server group selection unit 1364 refers to the first server group state holding unit 1322, and a server group with a low power cost as a server group for processing a new access from a server group in an operating state. Select with priority.

  The server group selection unit 1364 passes the IP address of the selected server group and the outgoing packet corresponding to the new access to the address conversion unit 1366. The address conversion unit 1366 converts the end point IP address of the received outgoing packet into the IP address of the server group selected by the server group selection unit 1364. The outgoing packet with the end point IP address converted in this way is sent from the address conversion unit 1366 to the internal network 3 as an assigned packet.

Each time the server group selection unit 1364 selects a server group 1022 for a new access, the connection update unit 1368 acquires information related to the selection from the server group selection unit 1364 and corresponds to the connection holding unit 1324. Add an entry. Information regarding this selection includes the user terminal IP address of the user terminal 1006 of the user who has made a new access, the load balancer IP address, the assigned server group IP address of the server group selected for the new access, and the sequence And a number.
Also, the connection update unit 1368 deletes entries that are no longer necessary.

  FIG. 6 is a block diagram showing the function and configuration of the operation management apparatus 1010. Each block shown here can be realized by hardware such as a computer (CPU) (central processing unit) and other elements and mechanical devices, and software can be realized by a computer program or the like. Here, The functional block realized by those cooperation is drawn. Therefore, it is understood by those skilled in the art who have touched this specification that these functional blocks can be realized in various forms by a combination of hardware and software.

  The operation management apparatus 1010 includes an index calculation unit 100, an index transmission unit 154, a display control unit 120, a load history holding unit 1112, a second server group state holding unit 1114, a learning unit 1130, and a state setting unit 1140. And a divergence determination unit 1150 and an override unit 1160.

The index calculation unit 100 calculates an index indicating the operating status of the plurality of server groups 1022 from information obtained from the external network 1004, the load balancer 1030, and the internal network 3.
The index calculation unit 100 includes a network interface unit 102, a communication information holding unit 104, a classification unit 132, a totaling unit 130, a first index holding unit 108, a coefficient holding unit 114, a first analysis unit 116, The usage status acquisition unit 110, the usage status holding unit 112, the second index holding unit 118, the second analysis unit 152, and the third index holding unit 158 are included.

The network interface unit 102 includes an intercepting unit 122, an extracting unit 124, a filter unit 126, and a registration unit 128.
The intercepting unit 122 intercepts packets from the external network 1004, the load balancer 1030, and the internal network 3. The extraction unit 124 extracts the start point IP address, the end point IP address, the port number, and the sequence number from the packet intercepted by the interception unit 122.

Hereinafter, the load balancer 1030, the server group 1022, or the server of the server group 1022 may be referred to as a node.
In the filtering mode in which the filter function is turned on, the filter unit 126 selects only packets including the IP address of the measurement target node based on the start point IP address and the end point IP address extracted by the extraction unit 124. To the registration unit 128. More specifically, the filter unit 126 registers the extraction source packet only when at least one of the start point IP address and the end point IP address extracted by the extraction unit 124 matches the IP address of the measurement target node. To the unit 128.

In particular, the filter unit 126 sequentially selects a measurement target node from a plurality of nodes included in the information processing system 2. That is, the filter unit 126 switches the IP address of the measurement target node in order among the IP addresses of a plurality of nodes included in the information processing system 2.
In the filtering off mode in which the filter function is turned off, the filter unit 126 outputs the packet from which the extraction point 124 has extracted the start point IP address, the end point IP address, the port number, and the sequence number to the registration unit 128.

  For the packet output by the filter unit 126, the registration unit 128 includes a time related to the packet, a start point IP address extracted from the packet by the extraction unit 124, and an end point IP address extracted from the packet by the extraction unit 124. The communication information holding unit 104 associates the communication usage corresponding to the port number extracted from the packet by the extraction unit 124 with the data amount of the packet and the sequence number extracted from the packet by the extraction unit 124. Register with.

FIG. 7 is a data structure diagram showing an example of data held in the communication information holding unit 104. The communication information holding unit 104 holds the time 202, the start point IP address 204, the end point IP address 206, the communication application 208, the data amount 210, and the sequence number 211 in association with each other.
The time 202 is a time related to the packet. For example, if the packet includes a time stamp, the time 202 is a time indicated by the time stamp. Alternatively, the time when the packet was intercepted by the intercepting unit 122 may be used.
The start point IP address 204, the end point IP address 206, and the sequence number 211 are the start point IP address, the end point IP address, and the sequence number extracted by the extraction unit 124, respectively.
The communication application 208 is a communication application corresponding to the port number extracted by the extraction unit 124.
The data amount 210 indicates the amount of data loaded in the packet in units of bytes (B).

  Returning to FIG. The classification unit 132 classifies the packet based on at least one of the start point IP address and the end point IP address extracted by the extraction unit 124 and the port number. Hereinafter, a case will be described in which the classification unit 132 classifies packets based on the start point IP address and the port number. However, even when the classification unit classifies a packet based on the end point IP address and the port number, and when the classification unit classifies the packet based on the start point, the end point IP address, and the port number, the following explanation is valid. It will be apparent to those skilled in the art who have touched the specification.

  The classification unit 132 sorts the data held in the communication information holding unit 104 with respect to the start point IP address. Hereinafter, the start point IP address registered in the communication information holding unit 104 is referred to as a registered IP address. The classification unit 132 classifies the packet for each registered IP address according to the communication use. In particular, the classification unit 132 refers to the server information holding unit 138, and acquires information on the original use that is a predetermined communication use for a node corresponding to the registered IP address (hereinafter referred to as a registered node). The classification unit 132 classifies the packet into a packet corresponding to the original use and a packet corresponding to a non-original use which is a communication use other than the original use. For example, if the registered node is a WEB server, the original use is WEB communication corresponding to the port number “80”. The non-original use is, for example, a communication use other than the original use, such as FTP transfer corresponding to the port number “20” if the registered node is a WEB server.

FIG. 8 is a data structure diagram showing an example of data held in the server information holding unit 138. The server information holding unit 138 holds the node name 214 of the node, the IP address 218 of the node, and the usage 224 of the node in association with each other.
When the classification unit 132 acquires information on the original use of a node by referring to the server information holding unit 138, the classification unit 132 may acquire the original use by referring to the use of the node.

Returning to FIG. The totaling unit 130 totals packets corresponding to the intended use for at least one registered node. In particular, the totaling unit 130 totals the data amount of the packet corresponding to the original use of the registered node for each period of a predetermined length, that is, the time zone, and totals the data amount of the packet corresponding to the non-original use.
For example, consider a case where the registered nodes corresponding to the registered IP address “121.21.15.1” are counted in the time zone “02/15/10 11:10:00 to 11:20:00”. The registered node from the server information holding unit 138 in FIG. 8 is the first front-end server 1024a, which is a WEB server. From the classification result in the classification unit 132, the totaling unit 130 has the start point IP address “121.21.15.1”, the communication application “WEB communication”, and the time “02/15/10 11:10:00”. Add up the data amount of packets included in "11:20:00" to get the total result for the intended use. In addition, the totaling unit 130 has a start IP address “121.21.15.1”, a communication usage not “WEB communication”, and a time zone “02/15/10 11:10:00 to 11:20:00 Are added together to obtain the total result for the non-original use.

The totaling unit 130 calculates the number of accesses for each time period with reference to the start point IP address, the end point IP address, and the sequence number for at least one registered node. The totaling unit 130 divides the calculated number of accesses by the length of the time zone, thereby calculating the number of accesses in the registered node at the time zone.
For the load balancer 1030, the aggregation unit 130 calculates the number of accesses from the outside of the information processing system 2 to the entire information processing apparatus 1020 (hereinafter referred to as the total number of accesses). That is, the totaling unit 130 totals the access to the load balancer 1030 by totaling packets having the IP address of the terminal external to the information processing system 2 as the start IP address and the IP address of the load balancer 1030 as the end point IP address. Calculate the number.
The aggregation unit 130 registers the aggregation result and the calculation result in the first index holding unit 108.

FIG. 9 is a data structure diagram illustrating an example of data held in the first index holding unit 108. The first index holding unit 108 includes the node name 228 of the registered node, the original use 230 of the registered node, the main port number 232 that is a port number corresponding to the original use, the original use data amount 234, and the non-original use. A data amount 236, a period of a predetermined length used in the counting unit 130, that is, a measurement period 238 corresponding to a time zone, and the number of accesses 239 are stored in association with each other.
The original use data amount 234 is a data amount obtained as a result of adding the data amount of the packet corresponding to the original use of the registered node within the measurement period.
The non-original use data amount 236 is a data amount obtained as a result of adding the data amount of the packet corresponding to the non-original use of the registered node within the measurement period.

  Returning to FIG. The usage status acquisition unit 110 acquires information on the usage status of the node from the node of the information processing system 2 at a predetermined time interval or in real time through the internal network 3. For example, if the node is a front-end server (WEB server) 1024 or an application server 1026, power consumption (unit: W), device temperature (unit: ° C), and CPU usage rate (unit:%). Yes, if the node is the database server 1028, the data transfer amount (unit: MB / second), power consumption, and device temperature. If the node is a network device such as the load balancer 1030, the data transfer amount, power consumption, and device. Temperature.

The usage status acquisition unit 110 may acquire data of a system performance monitor installed in the basic OS of the server.
The usage status acquisition unit 110 may acquire, as power consumption, a measured value of power supplied to each node from a bus bar table tap type PDU that supplies power to each node.
The usage status acquisition unit 110 includes a commercially available device (a temperature sensor or an optical fiber temperature sensor) that can measure the temperature of the node, and a data collection function (a commercially available software package) that stores and reports the temperature data. Or a server-embedded special hardware package), the temperature of the device may be acquired.

  Alternatively, the usage status acquisition unit 110 uses the following three reporting functions to obtain power consumption, device temperature, and CPU usage rate from nodes registered and certified by Energy Star for Server 2 of the US EPA. get. In other words, in Energy Star for Server 2 conducted by EPA in the United States, registered and certified devices can provide three reporting functions at the microcode (firmware) level of the hardware itself, and the basic OS and other It is supposed to have a hardware management function that does not affect the measurement software, the application, and the processing capacity of the original server (main body CPU / MPU (Micro Processing Unit)). The three reporting functions are 1) the actual power consumption value of the server, 2) the usage rate of the processing processor such as the main body CPU / MPU of the server, and 3) the actual operating temperature of the server.

  The usage status acquisition unit 110 registers the acquired usage status information, the time zone corresponding to the time when the usage status is acquired, the node name of the acquisition destination, and the IP address in association with each other in the usage status holding unit 112. . When the usage status acquisition unit 110 acquires information at a predetermined time interval, for example, the time zone is set to a period from the acquired time until the predetermined time interval elapses. Alternatively, when the usage status acquisition unit 110 acquires information in real time, a period having a predetermined length is set as a time zone, and information on the usage status is averaged over the period.

  FIG. 10 is a data structure diagram illustrating an example of data held in the usage status holding unit 112. The usage status holding unit 112 has acquired the node name 244 of the node whose usage status has been acquired by the usage status acquisition unit 110, the IP address 245, and the recording time zone 246 corresponding to the time when the usage status was acquired. The power consumption 248, the acquired device temperature 250, the acquired CPU usage rate 252 and the acquired data transfer amount 253 are stored in association with each other.

  Generally, in a WEB server, it is possible to measure how much CPU usage is required to process and send data for a predetermined amount, for example, 1 kB of WEB communication. That is, for the WEB server, the relationship between the data amount for WEB communication and the CPU usage rate can be measured in advance. The coefficient holding unit 114 holds a coefficient indicating the relationship (for example, unit of% · minute / GB). When this coefficient is 1, it means that, for example, in order to process and send data for 1 GB of WEB communication, the WEB server must operate for 1 minute at a CUP usage rate of 1%.

  FIG. 11 is a data structure diagram illustrating an example of data held in the coefficient holding unit 114. The coefficient holding unit 114 holds the server name 240 of the WEB server and the coefficient 242 in association with each other.

Returning to FIG. The first analysis unit 116 includes a first calculation unit 134, a second calculation unit 136, and a first synthesis unit 142.
The first calculation unit 134 is a measurement period corresponding to the original use data amount, that is, the data amount for WEB communication, for a WEB server (hereinafter referred to as a registered WEB server) among the registered nodes registered in the first index holding unit 108. The time average data amount (for example, unit of GB / min) for WEB communication is obtained by dividing by the length of, for example, the length in minutes. The first calculation unit 134 acquires a coefficient corresponding to the registered WEB server from the coefficient holding unit 114. The first calculation unit 134 multiplies the registration web server by the time average data amount for web communication and a coefficient, and obtains a value (for example, unit of%) obtained as a result of the multiplication by the registration web server and the measurement period. The CPU usage rate estimated value (hereinafter referred to as the first estimated value) by the processing related to WEB communication in FIG.

  For example, when paying attention to the node name “first front-end server” registered in the first index holding unit 108 in FIG. 9, the first calculation unit 134 detects the measurement period “02/15/10 11:10:00 to 11”. : 20: 00 ",“ 500 (GB) ”is obtained as the data amount for WEB communication. The first calculator 134 divides “500 (GB)” by the length of the measurement period “10 (minutes)” to obtain a time average data amount “50 (GB / minute)” for WEB communication. The first calculation unit 134 refers to the coefficient holding unit 114 in FIG. 11 and obtains a coefficient “1.5 (% · min / GB)”. The first calculation unit 134 multiplies the time average data amount “50 (GB / min)” for the WEB communication by the coefficient “1.5 (% · min / GB)” to obtain “75 (%)”. The first calculation unit 134 calculates this value “75 (%)” during the measurement period “02/15/10 11:10:00 to 11:20:00” on the WEB server of the node name “first front-end server”. It is set as the 1st guess value by the process regarding the WEB communication.

The second calculation unit 136 searches the usage status holding unit 112 using the server name and measurement period of the registered WEB server targeted by the calculation of the first estimated value in the first calculation unit 134 as keys, and the corresponding power consumption And device temperature and CPU usage rate are extracted. The second calculation unit 136 subtracts the first estimated value from the CPU usage rate extracted from the usage status holding unit 112, and estimates the CPU usage rate by processing other than the processing related to WEB communication (hereinafter referred to as second estimated value). And
Processing other than processing related to WEB communication is, for example, processing related to other communication, virus scanning by a virus checker, and writing of data to a hard disk drive. Or, you may be idle and not working.

  The second calculation unit 136 divides the extracted power consumption into the ratio between the first estimated value and the second estimated value, and estimates the value corresponding to the first estimated value for the power used for processing related to WEB communication. The value corresponding to the value and the second estimated value is the estimated value of the power used for the processing that is not so. The device temperature is the same in that the difference between the device temperature and the outside air temperature is divided into the ratio between the first estimated value and the second estimated value.

  For example, focusing on the node name “first front-end server” registered in the first index holding unit 108 in FIG. 9, the second calculation unit 136 receives the node name “first” from the usage status holding unit 112 shown in FIG. 1 front-end server ”and measurement period“ 02/15/10 11: 10: 00 ~ 11: 20: 00 ”Power consumption“ 30 (W) ”, device temperature“ 35 (° C) ”, CPU usage The rate “80 (%)” is extracted. The second calculation unit 136 subtracts the first estimated value “75 (%)” from the CPU usage rate “80 (%)” to obtain the second estimated value “5 (%)”. The second calculation unit 136 divides the power consumption “30 (W)” into a ratio between the first estimated value “75 (%)” and the second estimated value “5 (%)”, that is, 15: 1, and relates to WEB communication. The estimated power value “28 (W)” used for processing and the estimated power value “2 (W)” used for other processing are obtained.

  The first synthesizing unit 142 uses the time and node name as keys, the data held in the first index holding unit 108, the data held in the usage status holding unit 112, and the calculation result (registration) by the second calculation unit 136 The power consumption derived from the WEB communication process, the rising temperature derived from the WEB communication process, and the CPU usage rate derived from the WEB communication process) are registered in the second index holding unit 118 in association with each other.

FIG. 12 is a data structure diagram illustrating an example of data held in the second index holding unit 118. The second index holding unit 118 includes a node name 254, an original use 256, a main port number 258, an original use data amount 260, a non-original use data amount 262, a measurement period 264, power consumption 266, The device temperature 268, the CPU usage rate 270, the data transfer amount 271, the WEB communication-derived power consumption 272, the WEB communication-derived rising temperature 274, and the WEB communication-derived CPU usage rate 276 are stored in association with each other.
The WEB communication-derived power consumption 272 is an estimated value of power used for processing related to WEB communication, calculated by the second calculation unit 136. The WEB communication-derived increased temperature 274 is a portion that is predicted to have been increased by the process related to WEB communication among the difference between the device temperature and the outside air temperature calculated by the second calculation unit 136. Here, the outside air temperature is 25 (° C.). The WEB communication-derived CPU usage rate 276 is a first estimated value calculated by the first calculation unit 134.

Returning to FIG. The second analysis unit 152 uses the data related to the WEB server (front-end server 1024) among the data stored in the second index storage unit 118 for a time interval longer than a predetermined time interval in the usage status acquisition unit 110, for example, 1 Statistical processing is performed at time intervals of months and half years.
The second analysis unit 152 uses the average power consumption that is the time average of power consumption and the maximum value of power consumption within the statistical processing target period for the WEB server among the nodes registered in the second index holding unit 118. A certain maximum power consumption is calculated.
The second analysis unit 152 calculates the time average of the power consumption derived from WEB communication for the WEB server among the nodes registered in the second index holding unit 118 as the time average of the original use data amount (data amount for WEB communication). The power cost (for example, unit of W / GB) is calculated.
The second analysis unit 152 registers the server name, average power consumption, maximum power consumption, and power cost in the third index holding unit 158 in association with each other.

  FIG. 13 is a data structure diagram illustrating an example of data held in the third index holding unit 158. The third index holding unit 158 holds the server name 304, the average power consumption 306, the maximum power consumption 308, and the power cost 310 in association with each other.

  Returning to FIG. The display control unit 120 causes the monitor 140 to display a screen showing various parameters for each node based on at least one of the first index holding unit 108, the second index holding unit 118, and the third index holding unit 158.

  The index transmission unit 154 transmits information regarding the power cost registered in the third index holding unit 158 to the load balancer 1030 in order to update the first server group state holding unit 1322.

  The learning unit 1130 includes an update unit 1132 and a load prediction unit 1134. The learning unit 1130 stores the history of the total number of accesses in the load history holding unit 1112 by the update unit 1132, and predicts the total number of accesses that can occur in the future by the load prediction unit 1134 based on the storage. As a result, the longer the operation results, the smarter the operation management from the viewpoint of energy saving.

  The update unit 1132 acquires the total number of accesses from the first index holding unit 108. The updating unit 1132 updates the load history holding unit 1112 based on the acquired total number of accesses. Therefore, the load history holding unit 1112 holds the history of the total number of accesses, that is, the total number of accesses requested in the past.

FIG. 14 is a data structure diagram showing the load history holding unit 1112. The load history holding unit 1112 includes a time zone 1218, an attribute 1220 to which the time zone belongs, the total number of accesses 1222 acquired in the time zone, and the number of active server groups 1224 set to the operating state in the time zone. And the average operating rate 1226 for that time period are stored in association with each other. The attribute 1220 to which the time zone belongs is, for example, a day of the week such as Monday, Tuesday, a weekday holiday, a year-end / New Year holiday, a morning / afternoon, or any combination thereof. The average operation rate 1226 is an average value of the operation rates of the server group set in the operation state.
The update unit 1132 may refer to the second server group state holding unit 1114 in order to obtain the number 1224 of active server groups set to the operation state.
The time zone 1218 is a period having a time interval that serves as a reference for updating the mode of the operation management apparatus 1010.

  Returning to FIG. The load prediction unit 1134 predicts the total number of accesses that will occur in the future based on the total number of accesses requested in the past. The load prediction unit 1134 predicts the total number of accesses for each future time zone. The state setting unit 1140, which will be described later, also sets the state of the server group 1022 as necessary for each time period. In particular, the load predicting unit 1134 predicts the total number of accesses with reference to the load history holding unit 1112 for a time zone (hereinafter referred to as a time zone to be predicted) for which the total number of accesses is predicted. Hereinafter, the total number of accesses predicted by the load prediction unit 1134 is referred to as a predicted total number of accesses.

  The load prediction unit 1134 extracts a trend of the total number of accesses according to the purpose for which the information processing system 2 is used from the load history holding unit 1112, and the time zone to be predicted based on the extracted trend The total number of accesses is predicted. In particular, the load prediction unit 1134 extracts the tendency of the total number of accesses using the time zone attribute as a key.

The load predicting unit 1134 obtains the total number of accesses corresponding to the attribute to which the prediction target time zone belongs from the load history holding unit 1112 and sets the total access number for the time zone.
For example, when a day of the week is set as a key, the load prediction unit 1134 refers to the load history holding unit 1112 and calculates an average value of the total number of accesses for each day of the week. Of these average values, the load prediction unit 1134 sets the average value of the day of the week that matches the day of the week to be predicted as the predicted total access count. Also, for example, when the type of weekday holiday is set as a key, the load prediction unit 1134 refers to the load history holding unit 1112 and calculates the average value of the total number of accesses on weekdays and the average value of the total number of accesses on holidays. The load predicting unit 1134 uses the average value of these average values as the predicted total access number on weekdays if the time zone to be predicted is a weekday and on holidays if it is a holiday.
Note that although the average value is used here as an example, other statistical parameters such as a representative value may be used.

  The load prediction unit 1134 passes the predicted total number of accesses to the deviation determination unit 1150 and the load comparison unit 1142 of the state setting unit 1140.

  When the predicted total number of accesses is smaller than the predetermined access threshold Mo, the state setting unit 1140 sets at least one server group 1022 included in the information processing apparatus 1020 in the power saving state in the prediction target time zone.

  The mo access threshold value Mo in the state setting unit 1140 is set in a range where the performance of the information processing apparatus 1020 does not deteriorate. Here, the performance means, for example, how many accesses can be processed at what speed. Alternatively, the performance may be a response time such as a time taken to process one access. Further, the performance of the information processing apparatus 1020 is reduced. For example, the number of accesses exceeding the peak access number is assigned to a certain server group 1022, and as a result, the processing speed of the server group 1022 is reduced. That is, the access processing speed of the entire 1020 decreases.

  For example, it is assumed that the peak access counts of the first server group 1022a to the fifth server group 1022e are all 2000. In this case, if the access threshold Mo is set to 9000, at least one server group must be set in the power saving state in the prediction target time zone even if the predicted total number of accesses is 8500. Here, it is assumed that the fifth server group 1022e is set to the power saving state. The total peak access number of the remaining four server groups 1022a to 22d is 8000, which is smaller than the predicted total access number 9500. Therefore, in this case, when the time zone to be predicted arrives, there is a possibility that at least one server group among the remaining four server groups 1022a to 22d has to process accesses exceeding the peak access number. In this case, the processing speed of the server group decreases. As a result, the processing speed of the entire information processing apparatus 1020 may decrease. In order to avoid such a situation, the access threshold value Mo is set in a range where the performance of the information processing apparatus 1020 does not deteriorate. In the above example, the access threshold Mo may be set to 8000 or less.

  The state setting unit 1140 determines a server group to process access on the premise that the maximum performance of the server group 1022 is exhibited, and sets the remaining server group to the power saving state in the prediction target time zone. In this case, maximizing the performance of the server group 1022 is, for example, causing the server group 1022 to perform access processing with the peak number of accesses, in other words, using the server group at an operation rate of 100%. Further, when it is predicted that the performance of the information processing apparatus 1020 will deteriorate due to a change in the predicted total number of accesses, the state setting unit 1140 selects at least one server group that is set in the power saving state in the prediction target time zone. Set to operational state.

Regarding the setting of the server group 1022 to the power saving state or the operating state, the state setting unit 1140 defines a range of the predicted total number of accesses according to the number of server groups to be in the operating state in the prediction target time zone. . For example, if the predicted total number of accesses is in the range from 0 to the first threshold value T1, the state setting unit 1140 sets only one server group to the operating state and sets the other server groups to the power saving state in the prediction target time zone. . Table 1 shows the number of server groups that are in an active state, the number of server groups that are in an OS sleep state, the number of server groups that are in a power-off state, and the predicted total number of accesses for the state setting in the state setting unit 1140 The relationship with the range is shown. T2 is a second threshold value, T3 is a third threshold value, and T1 <T2 <T3. Individual threshold values are set in advance by an administrator of the information processing system 2.

  The first threshold value T1, the second threshold value T2, and the third threshold value T3 are set on the assumption that the server group 1022 is used at an operation rate of 100%. That is, in the example in which all the peak access numbers of the first server group 1022a to the fifth server group 1022e are 2000, T1 = 2000, T2 = 4000, and T3 = 6000. In this case, the minimum number of server groups necessary for processing the access of the predicted total access count is set in the operating state in the prediction target time zone. For example, when the predicted total number of accesses between the first threshold value T1 and the second threshold value T2 increases and exceeds the second threshold value T2, if it remains as it is, at least one in the prediction target time zone. Because the operating rate of the server group is predicted to exceed 100%, the number of server groups that are in the operating state in the prediction target time zone is increased by one to avoid a decrease in the processing speed of the information processing apparatus 1020.

The state setting unit 1140 includes a load comparison unit 1142, an active server group determination unit 1144, and a state signal generation unit 1146.
The load comparison unit 1142 determines the magnitude relationship between the predicted total number of accesses acquired from the load prediction unit 1134 and the first threshold value T1, the second threshold value T2, the third threshold value T3, and the access threshold value Mo. Determine. This magnitude relationship is, for example, information “T2 <predicted total access count <T3”. The load comparison unit 1142 passes information on the magnitude relationship to the active server group determination unit 1144.

Based on this information, the active server group determination unit 1144 determines the necessity of switching the state in the next time zone according to the strategy shown in Table 1. When it is necessary to switch the state, the active server group determination unit 1144 sets the server group to be in the active state, the server group to be in the OS sleep state, and the power-off state based on the index acquired from the index calculation unit 100. Determine the server group. The active server group determination unit 1144 updates the second server group state holding unit 1114 and the first server group state holding unit 1322 of the load balancer 1030 based on this determination in accordance with the arrival of the next time zone. The active server group determination unit 1144 passes information on a server group that needs to be switched to the state signal generation unit 1146 when the next time zone arrives. The active server group determination unit 1144 suspends or terminates the process when it is not necessary to switch the state, and waits for the next information.
The second server group state holding unit 1114 is a table similar to the first server group state holding unit 1322 shown in FIG. Note that either the operation management apparatus 1010 or the load balancer 1030 may include a server group state table and refer to a table provided by a person who does not include the table.

  The active server group determination unit 1144 first determines the number of server groups to be set in the active state, the OS sleep state, and the power-off state from the strategy shown in Table 1. Next, the active server group determination unit 1144 refers to the second server group state holding unit 1114, and it is necessary to switch the state of the server group in the next time zone. In other words, the next time zone predicted by the load prediction unit 1134 It is determined whether or not the number of server groups in each state corresponding to the predicted total number of accesses in the server matches the number of server groups in each current state registered in the second server group state holding unit 1114. If they match, the active server group determination unit 1144 interrupts or ends the processing.

If they do not match, the active server group determination unit 1144 determines a server group to be in each state in the next time zone. The active server group determination unit 1144 acquires data held in the third index holding unit 158 of the index calculation unit 100. The active server group determination unit 1144 sets an energy saving stoppable order indicating the order in which the power saving state is set for each server group 1022 based on the acquired data. The order of possible energy saving stops is “1”, which is the highest and decreases as the number increases.
In the active server group determination unit 1144, the average power consumption, the maximum power consumption, and the power cost of the front-end server 1024 held in the third index holding unit 158 are changed to the average power consumption of the server group 1022 to which the front-end server 1024 belongs, It is considered as the maximum power consumption and power cost.

表２のストラテジでは、サーバ群１０２２の電力コストが大きいほど優先的にそのサーバ群１０２２が省電力状態とされるように省エネ停止可能順位が設定される。電力コストが遜色ない２つのサーバ群が存在する場合は、平均消費電力や最高消費電力が高いほうのサーバ群の省エネ停止可能順位をより高く設定する（表２の第４サーバ群と第５サーバ群）。 Table 2 shows the relationship among the server group name, the average power consumption, the maximum power consumption, the power cost, and the energy saving stoppable order when the energy saving stoppable order is determined based on the power cost of the server group 1022.

In the strategy of Table 2, the energy saving stoppage priority order is set so that the higher the power cost of the server group 1022 is, the more preferentially the server group 1022 is put into the power saving state. If there are two server groups with similar power costs, the higher energy saving stoppage order is set for the server group with the higher average power consumption or maximum power consumption (the fourth server group and the fifth server in Table 2). group).

表３のストラテジでは、サーバ群１０２２の平均消費電力と最高消費電力との和が大きいほど優先的にそのサーバ群１０２２が省電力状態とされるように省エネ停止可能順位が設定される。
情報処理装置１０２０において各サーバ群１０２２が主にＷＥＢ通信をおこなっており他の用途の処理は無視できる程度であり、またロードバランサ１０３０において稼動状態にあるサーバ群にアクセスが均等に割り当てられる設定となっている場合は、稼動状態にあるサーバ群には同じ負荷がかけられれていると考えることができる。したがって、稼動状態にある各サーバ群の平均消費電力は、同じ負荷がかけられれている場合に消費される電力の平均値と同一視することができ、また、稼動状態にある各サーバ群の最高消費電力は、同じ負荷がかけられれている場合に消費される電力の最高値と同一視することができる。この場合、表３のストラテジでは、サーバ群１０２２に同じ負荷がかけられた場合に消費される電力が大きいほど優先的にそのサーバ群１０２２が省電力状態とされるように省エネ停止可能順位が設定されているといえる。 Table 3 shows the relationship among the server group name, the average power consumption, the maximum power consumption, the power cost, and the energy saving stoppable order when the energy saving stoppable order is determined based on the power consumption of the server group 1022.

In the strategy of Table 3, the energy saving stoppage priority order is set so that the server group 1022 is preferentially put into the power saving state as the sum of the average power consumption and the maximum power consumption of the server group 1022 increases.
In the information processing apparatus 1020, each server group 1022 mainly performs WEB communication, and processing for other uses is negligible. In addition, the load balancer 1030 is configured so that access is evenly allocated to the server group in the operating state. In this case, it can be considered that the same load is applied to the server group in the operating state. Therefore, the average power consumption of each server group in the operating state can be equated with the average value of the power consumed when the same load is applied, and the maximum power of each server group in the operating state is the same. The power consumption can be equated with the highest value of power consumed when the same load is applied. In this case, according to the strategy shown in Table 3, the energy saving stoppage priority order is set so that the larger the power consumed when the same load is applied to the server group 1022, the higher the power consumption of the server group 1022 is. It can be said that.

When one of the server groups in the operating state needs to be switched to the power saving state, the active server group determining unit 1144 has the highest order of possible energy saving stoppages set in Table 2 or Table 3 among the server groups in the operating state. A high server group is selected as a server group that should be in a power saving state.
When it is necessary to switch one of the server groups in the power saving state to the operating state, the active server group determination unit 1144 has the energy saving stop possibility order set in Table 2 or Table 3 among the server groups in the power saving state. The lowest server group is selected as the server group that should be in operation.

  In the above-described algorithm for determining the server group 1022 in the active server group determination unit 1144, access identity may be taken into consideration particularly when the operating state is switched to the power saving state. For example, when switching the operating state to the power saving state, the active server group determining unit 1144 updates the second server group state holding unit 1114 and the first server group state holding unit 1322 as the next time zone arrives. While restricting assignment of new access to the server group that should be switched to the power saving state, refer to the connection holding unit 1324 of the load balancer 1030 and confirm that there is no access to the server group that should be switched to the power saving state. The information on the server group that needs to be switched from one state to another may be passed to the state signal generation unit 1146. As a result, access identity can be guaranteed.

The status signal generator 1146 is connected to the first server group 1022a to the fifth server group 1022e via the internal network 3.
The state signal generation unit 1146 is one of a dormancy introduction signal, a dormancy release signal, a power off signal, and a power on signal corresponding to switching for the server group based on the information of the server group that needs to be switched. Send. For example, when it is necessary to change the third server group 1022c from the operating state (OS sleep state) to the OS sleep state (operating state), the state signal generation unit 1146 sends a sleep introduction signal (sleep release signal) to the third server group 1022c. ). Further, when it is necessary to change the third server group 1022c from the operating state (power off state) to the power off state (operating state), the state signal generation unit 1146 sends a power off signal (power on) to the third server group 1022c. Signal).
The server group set from the power saving state to the operating state by the status signal generation unit 1146 is recorded in the first server group state holding unit 1322 of the load balancer 1030 by the operating server group determination unit 1144 as being in the operating state. Therefore, a new access is allocated by the request allocation unit 1036 of the load balancer 1030.

  The operation management apparatus 1010 is based on the total number of accesses held in the first index holding unit 108 in addition to the above-described load prediction mode in which the state setting unit 1140 sets the state of the server group 1022 based on the predicted total number of accesses. The state setting unit 1140 has an adaptive mode in which the state of the server group 1022 is adaptively set. Hereinafter, this adaptive mode will be described.

When the difference between the predicted total number of accesses and the actual total number of accesses is large, that is, when the absolute value of the difference is larger than a predetermined divergence value, the divergence determination unit 1150 indicates to the state setting unit 1140 the actual total number of accesses. Let the server group state be set based on
The deviation determination unit 1150 acquires the total number of accesses corresponding to the current time zone from the first index holding unit 108. The divergence determination unit 1150 also acquires the predicted total number of accesses predicted for the current time zone from the load prediction unit 1134. Then, the deviation determination unit 1150 calculates the difference between the two. When the absolute value of the difference is larger than the deviation value, the deviation determination unit 1150 causes the load comparison unit 1142 of the state setting unit 1140 to use the total number of accesses from the first index holding unit 108 instead of the predicted total number of accesses. . In this case, the load comparison unit 1142 compares the total number of accesses from the first index holding unit 108 with the first threshold value T1, the second threshold value T2, the third threshold value T3, and the access threshold value Mo. Determine the relationship. The state setting unit 1140 performs processing similar to the above-described processing using the magnitude relationship. Thereby, the state setting unit 1140 can adaptively set the state of the server group 1022 according to the actual total number of accesses when the difference between the predicted total number of accesses and the actual total number of accesses is large. .

The operation management apparatus 1010 also has a manual setting mode in which the administrator of the information processing system 2 can set the state of the server group 1022 manually.
The override unit 1160 receives a manual setting by an administrator from an input device 1012 such as a keyboard attached to the operation management device 1010. When the override unit 1160 receives this manual setting, the override unit 1160 causes the state setting unit 1140 to set the state of the server group 1022 based on the manual setting. This makes it possible to cope with sudden changes in the usage environment.

  The display control unit 120 acquires the total number of accesses in the current time zone from the first index holding unit 108 and the current state of the server group 1022 from the second server group state holding unit 1114. The display control unit 120 displays status screens 1400 a to 1400 d indicating the total number of accesses and the state of the server group 1022 on the monitor 140 associated with the operation management apparatus 1010. When the absolute value of the difference between the total number of accesses predicted by the divergence determination unit 1150 and the actual total number of accesses is determined to be larger than the divergence value, the display control unit 120 displays a warning screen 1402 on the monitor 140. Display. Further, the display control unit 120 causes the monitor 140 to display a state setting screen 1404 for manual setting by the administrator.

15A to 15D are representative screen diagrams of the status screens 1400a to 1400d. Here, it is assumed that the peak access counts of the first server group 1022a to the fifth server group 1022e are all 2000. Further, consider a case where the strategy shown in Table 2 is used in the active server group determination unit 1144 and the inclination distribution mode is used in the server group selection unit 1364. FIGS. 15A to 15D correspond to cases where the total number of accesses in the current time zone is 1600, 2800, 4400, and 7000, respectively.
FIG. 15A is a representative screen diagram of the status screen 1400a corresponding to the case where the total number of accesses in the current time zone is 1600. The status screen 1400a includes a total access number area 1406, a graph 1408, and an override button 1410. The total access number area 1406 indicates the total number of accesses in the current time zone. A graph 1408 indicates the operation rate of each server group 1022. In the graph 1408, “Δ” indicates an OS sleep state, and “X” indicates a power-off state. Override button 1410 triggers override portion 1160 when depressed. When the override button 1410 is pressed, the display control unit 120 displays a state setting screen 1404 on the monitor 140. The same applies to FIGS. 15B to 15D.

FIG. 16 is a representative screen diagram of the warning screen 1402.
FIG. 17 is a representative screen diagram of the state setting screen 1404. The state setting screen 1404 includes a setting area 1412 and a setting button 1414. In the state immediately after the state setting screen 1404 is opened, the current state of each server group is indicated in the setting area 1412 by a radio button method. In the setting area 1412, the first server group 1022a to the fifth server group 1022e are displayed with the names server group A to server group E, respectively. The administrator uses the input device 1012 such as a mouse to select the state of each server group and presses the setting button 1414. Then, the state of each server group 1022 set in the setting area 1412 is sent to the override unit 1160 as a desired manual setting. Accordingly, it is possible to set whether each of the first server group 1022a to the fifth server group 1022e is in an operating state, an OS sleep state, or a power-off state.

  In the load balancer 1030 and the operation management apparatus 1010 described above, examples of the holding unit are a hard disk and a memory. Based on the description in this specification, each unit is realized by a CPU (not shown), an installed application program module, a system program module, a memory that temporarily stores the contents of data read from the hard disk, and the like. It is understood by those skilled in the art who have touched this specification that they can do this.

  The operation of the operation management apparatus 1010 having the above configuration will be described. FIG. 18 is a flowchart showing a series of processing in the index calculation unit 100. The intercepting unit 122 intercepts the packet from the network (S402). The extraction unit 124 extracts the start point address and the port number from the intercepted packet (S404). The registration unit 128 registers the extracted start point address, port number, and packet data amount in the communication information holding unit 104 (S406). The classification unit 132 sorts the data held in the communication information holding unit 104 by the start point address (S408). The classification unit 132 classifies the sorted data into an original use and a non-original use (S410). The tabulation unit 130 tabulates the classified data (S412). The second analysis unit 152 calculates the power cost of each server group 1022 from the aggregation result (S414).

  FIG. 19 is a chart showing a series of processes in the learning unit 1130 and the state setting unit 1140 in time series. Consider a case in which the time interval for the update unit 1132 to acquire the total number of accesses from the first index holding unit 108 is 15 minutes. At 9:15 on September 16, 2009, the update unit 1132 acquires the total number of accesses from the first index holding unit 108 (S502). The updating unit 1132 updates the load history holding unit 1112 based on the acquired total number of accesses (S504). On the other hand, the load prediction unit 1134 refers to the load history holding unit 1112 when it does not overlap with the update of the load history holding unit 1112 in step S504 within the time zone of 9: 15-9: 30 on September 16, 2009. The total number of accesses in the next time zone (1009: 30-9: 45) is predicted (S506). The state setting unit 1140 compares the predicted total number of accesses with the access threshold value Mo, the first threshold value T1, the second threshold value T2, and the third threshold value T3 (S508). The state setting unit 1140 determines whether or not it is necessary to switch the state of the server group 1022 in the next time zone based on the size comparison (S510). When switching of the state is not necessary (N in S510), the state of the server group 1022 is not set even when the next time zone comes. When the state needs to be switched (Y in S510), the state setting unit 1140 determines the state of the server group 1022 in the next time zone based on the energy-saving stoppage priority order (S512). When the next time zone arrives, that is, at 9:30 on September 16, 2009, the state setting unit 1140 sets the state of the server group as determined in step S512. In addition, the state setting unit 1140 updates the second server group state holding unit 1114 and the first server group state holding unit 1322 of the load balancer 1030 in parallel (S514). The operation management apparatus 1010 repeats this process for each time zone.

  According to the operation management apparatus 1010 according to the present embodiment, the trend of the total number of accesses according to the purpose for which the information processing system 2 such as a data center is used, for example, securities business or search engine is grasped and used. The total number of accesses that can occur in the future can be predicted. When the predicted total number of accesses is small, at least one server group is set in the power saving state in the prediction target time zone.

  As described above, the power consumption of the server group in the operating state is approximately 60% of the peak time even in the idle state. On the other hand, the power consumption of the server group in the power saving state is approximately 0 to 10% at the peak time. In the present embodiment, when there are server groups that are predicted to have a small number of accesses and are therefore predicted not to be processed, the server groups are set in a power saving state instead of an idle state. As a result, power consumption of the entire information processing apparatus 1020 can be reduced, wasteful use of power can be suppressed, and energy saving can be achieved.

  In addition, the operation management apparatus 1010 according to the present embodiment grasps the tendency of the total number of accesses according to the purpose for which the information processing system 2 is used, and predicts the total number of accesses that can occur in the future using the trend. Therefore, optimal power consumption reduction and energy saving can be realized for each purpose of use.

  Furthermore, according to the operation management apparatus 1010 according to the present embodiment, the index calculation unit 100 intercepts the network and calculates an index indicating a difference in characteristics of each server group 1022. When setting at least one server group 1022 in the power saving state, the state setting unit 1140 selects a server group to be set in the power saving state based on the index calculated by the index calculation unit 100. Therefore, it is possible to select a server group in the power saving state from among the server groups in the operating state in the form of incorporating the analysis result of the difference in characteristics of each server group 1022. As a result, it is possible to reflect the difference in the characteristics of each server group 1022 in the selection as compared with the case of selecting the server group in which the power saving state is considered after considering all the server groups 1022 to be equivalent in characteristics. It is possible to control the state of the server group 1022 that is more suitable from the viewpoint. As a result, energy saving of the information processing system 2 can be further promoted.

  In addition, the active server group determination unit 1144 preferentially selects a server group having a large power cost as a server group in which the power saving state is set. As a result, the power cost of the server group that is in the operating state is smaller than the power cost of the server group that is in the power saving state, so that less work is consumed effectively when viewed from the information processing system 2 as a whole. Can be processed with electricity.

  Further, according to the operation management apparatus 1010 according to the present embodiment, packets are intercepted from the network, and the intercepted packets are classified by the start IP address and the port number. Therefore, the communication data amount can be measured for each starting point IP address and each port number. Accordingly, it is possible to trace how much work corresponding to which communication application (e-mail communication, WEB browsing, etc.) has been performed for a desired node to be analyzed from a packet flowing on the network without directly inquiring the node. Since the node is not inquired, there is almost no burden on the node side accompanying the analysis of the workload. Since the operation management apparatus 1010 intercepts the network, the contribution of the operation management apparatus 1010 itself to the measurement is small and can be grasped.

  Further, in the present embodiment, the update unit 1132 updates the load history holding unit 1112 based on the total number of accesses calculated by the index calculation unit 100. Therefore, history data for predicting the total number of accesses can be obtained without increasing the load on the server side.

  In the present embodiment, the active server group determination unit 1144 determines the power saving state based on the index calculated by the index calculation unit 100 when the predicted total number of accesses is predicted to be greater than a predetermined value. At least one server group is selected from among a plurality of server groups set to. The status signal generation unit 1146 sets at least one server group selected by the active server group determination unit 1144 to the operating state. Thereby, first, it is possible to avoid a situation where the server group 1022 has to be used at a peak access number or more, and to avoid a delay in access processing. Furthermore, a server group that is in an active state can be selected from among the server groups that are in a power saving state in a form that incorporates an analysis result of the difference in characteristics of each server group 1022. As a result, it is possible to reflect the difference in the characteristics of each server group 1022 in the selection as compared with the case of selecting the server group to be in an operating state after regarding all the server groups as equivalent in characteristics. More favorable state control of the server group 1022 is possible.

  In addition, the operating server group determination unit 1144 preferentially selects a server group with a low power cost as a server group that is in an operating state. As a result, the power cost of the server group that is in the operating state is smaller than the power cost of the server group that is in the power saving state, so that less work is consumed effectively when viewed from the information processing system 2 as a whole. Can be processed with electricity.

  In the present embodiment, the load balancer 1030 preferentially assigns the acquired access to a server group with a low power cost. Therefore, as a result, the power cost of the server group to which more accesses are allocated is smaller than the power cost of the server group that is not so, so that effective work can be processed with less power consumption when viewed from the entire information processing system 2. it can.

  Moreover, in the operation management apparatus 1010 according to the present embodiment, the load history holding unit 1112 stores the acquired total access count and the attribute to which the time zone from which the total access count belongs, in association with each other. . Therefore, the tendency of the total number of accesses can be grasped on the basis of attributes. Further, the operation management apparatus 1010 obtains the total number of accesses corresponding to the attribute to which the prediction target time zone belongs from the load history holding unit 1112 and sets the total number of accesses predicted for the time zone. Therefore, the accuracy of the predicted total number of accesses can be improved by appropriately setting the attributes according to the purpose of use of the information processing system 2.

  FIG. 20 is a data structure diagram showing an example of the load history holding unit 1112a when the information processing system 2 according to the present embodiment is used as a data center that provides an Internet stock trading system of a securities company. Here, the number of operating servers and the average operating rate are omitted for the sake of clarity. As a tendency of the total number of accesses in this case, access may be concentrated within the trading hours of the stock exchange. In addition, some work is required to perform so-called batch processing for a while after the stock exchange closes. Other than that, there is almost no access during times when the stock exchange is closed, such as holidays. Even so, the brokerage firm's customers refer to their account information. Therefore, as shown in FIG. 20, as a time zone attribute, an attribute “transaction time” in a time zone corresponding to the trading time of a stock exchange, an attribute “batch processing” in a time zone in which batch processing is performed, “Holiday” may be given in the case of “out of hours” or weekends and holidays in other time zones. By assigning such attributes, it is possible to more appropriately predict the total number of accesses occurring in the data center that provides the securities company's Internet stock trading system.

  FIG. 21 is a data structure diagram showing an example of the load history holding unit 1112b when the information processing system 2 according to the present embodiment is used as a data center that provides a search service. Here, the number of operating servers and the average operating rate are omitted for the sake of clarity. As a tendency of the total number of accesses in this case, a holiday is more frequently used than a weekday, and a weekday is more frequently used in the afternoon than in the morning. Therefore, as shown in FIG. 21, a combination of a weekday holiday and a morning / afternoon may be given as the attribute of the time zone. By assigning such attributes, the total number of accesses occurring in the data center that provides the search service can be predicted more appropriately.

  In the operation management apparatus 1010 according to the present embodiment, in the power saving mode (predicted total access count <access threshold Mo), as shown in Table 1, a server group that is in an OS sleep state and a server that is in a power-off state Both groups are provided. As a result, when the total number of accesses suddenly increases in the prediction target time zone, it can be dealt with by returning the server group in the OS sleep state in which the overhead for returning after shifting to the adaptive mode is small to the operating state. Further, as long as it can cope with such a situation, the other server groups are in a power-off state in which power is not consumed, and the power consumption of the entire information processing apparatus 1020 is further reduced. In Table 1, only one server group that is in the OS sleep state is secured, but this number may be determined according to the balance between overhead and power consumption, and it is mentioned in the present specification that it can be appropriately increased or decreased. It will be understood by those skilled in the art.

  In the operation management apparatus 1010 according to the present embodiment, the state setting unit 1140 determines a server group to be in an operating state in a prediction target time zone on the premise that the maximum performance of the server group is exhibited. According to this server group determination method, a larger number of server groups can be put into a power saving state with respect to a given predicted total number of accesses. Therefore, the power consumption of the entire information processing apparatus 1020 can be further reduced. It is also true that the power consumption of the server group differs depending on the operating rate, but considering the fact that about 60% of the power is consumed even in the idle state as described above, power reduction by lowering the operating rate It is considered that the power reduction effect by setting the idle state to the power saving state is greater than the effect.

  Further, the access threshold Mo is set in a range where the performance of the information processing apparatus 1020 does not deteriorate. As a result, when the predicted total number of accesses is large, the parallel processing capability of the information processing apparatus 1020 is fully exhibited in the normal mode, and when the predicted total number of accesses decreases, the power consumption mode is reduced while shifting to the power saving mode to maintain the performance. Can be reduced.

  Further, in the operation management apparatus 1010 according to the present embodiment, the IP address indicates the subject of work, and the port number indicates the communication use. Therefore, the breakdown of the work being performed at the node is derived using them. Effective work can be measured. In this respect, it is greatly different from the conventional technique in which it is practically difficult to measure the effective work amount. According to the operation management apparatus 1010, by analyzing the amount of work in a node, it is possible to measure how much effective work the node has performed, and a good index for saving the information processing system 2 can be obtained.

  Also, in the operation management apparatus 1010 according to the present embodiment, the intercepted packet information held in the communication information holding unit 104 is classified into an original use and a non-original use. Therefore, as shown in the first index holding unit 108 in FIG. 9, it is possible to obtain the original use data amount and the non-original use data amount for a desired analysis target node. Thereby, it can be tracked how much the node to be analyzed is used for the original purpose from the ratio of both. This is one of the good indexes for saving energy in the information processing system 2. For example, the efficiency of the information processing system 2 can be improved by finding a node that is not used for the original purpose and applying an appropriate measure.

  In the operation management apparatus 1010, the second index holding unit 118 has a form in which the first index holding unit 108 and the usage status holding unit 112 are associated with each other. Therefore, when considering the energy saving of the information processing system 2, it is possible to grasp the workload at the node and the usage status in association with each other. In other words, it is possible to analyze how much CPU processing power is used, how much power is consumed, and how much temperature has been increased by what and when the node has done what work.

  For example, by looking at the breakdown of the workload of the server whose CPU usage rate is 100%, it is possible to know how much of the CPU usage rate of 100% is actually used for effective work. If most of the CPU usage rate is not used for effective work, the server should be considered for energy saving. In the conventional technology using PUE (Power Usage Effectiveness) (see Non-Patent Document 1) or the like, it is impossible to make such a determination because the breakdown of work cannot be seen, but it is possible to use the operation management apparatus 1010. It becomes.

  Also, for example, if there is a server with both low power consumption and effective workload, it looks like a server that is good for energy saving because of low power consumption when looking only at the usage status, and no measures are taken against this server. It is likely not. However, if the operation management apparatus 1010 is used to present the usage status and the workload so that they can be compared, it can be seen that the effective workload is also low. Therefore, the efficiency of the data center is improved and further energy saving is promoted. Therefore, it is possible to take measures such as turning work from another high-load server by virtualization, for example. Therefore, the contribution to energy saving of the data center is large.

  In addition, for example, when there is a server with a large exhaust heat although there is no problem in the effective work amount, it can be found. Therefore, if such a server exists in the hot aisle, it can be proposed to the server owner that it is better to move it to another location.

  Also, in the operation management apparatus 1010, in the filtering mode, the filter unit 126 selects only packets including the IP address of the measurement target node based on the start point IP address and the end point IP address extracted by the extraction unit 124. And output to the registration unit 128. Therefore, the amount of data held in the communication information holding unit 104 can be reduced within a range that does not affect measurement.

  In addition, the filter unit 126 sequentially selects a measurement target node from a plurality of nodes included in the information processing system 2. That is, not all the nodes are measured at one time, but the nodes to be measured are changed in order, for example, the first server group 1022a on Monday, the second server group 1022b on Tuesday, and the like. As a result, it is possible to reduce all the data held in the communication information holding unit 104 within a range that does not affect the measurement and set all nodes of the network as measurement targets.

  Consider a server that is being virtualized. Conventionally, the processor usage rate of each server image (individual guest OS) on the parent (host) virtualized OS is mainly obtained and compared using the VMOTION function or the like. Then, the guest OS is relocated to another virtual host OS server that is vacant in terms of processor utilization. However, in the case where the guest OS is moved between the virtualization servers using only the processor usage rate as a scale in this way, the breakdown of work actually performed by the guest OS is not taken into consideration. Therefore, it cannot be said that the relocation based on the effective work volume is performed.

When analyzing the work in the virtualized server using the operation management apparatus 1010, the work amount is known for each virtualization platform, so that it is possible to transfer the guest OS with high accuracy based on the effective work amount.
In addition, for example, it is possible to know how much the situation has been improved by virtualization. That is, the degree of improvement can be known by comparing the effective work amount and power consumption before and after virtualization. In addition, after virtualization, the current situation is that little attention is paid to the work performed by each guest OS, but the analysis apparatus 100 keeps track of the effective work amount for each guest OS, so that it is not used with time, for example. Guest OS can be specified. Therefore, the performance of other guest OSs can be improved by removing the guest OS thus identified. This contributes to improving the efficiency of the data center.

In the past, system operators have mainly evaluated server performance using server processor usage and memory utilization in system operation, and server integration and application Relocation, distribution, integration, and server update (exchange / replace with new equipment). However, particularly in the recent economic downturn, there is a need to further reduce the operational cost of the entire data center by further effective use of servers and further reduction of power consumption.
Therefore, by using the operation management apparatus 1010, it is possible to analyze in detail whether the work at the server is inward (to the server) or outward (from the server), and the work is a work to the outside of the data center. You can also analyze whether it is work for other servers and applications in the data center. Therefore, based on this analysis, high-accuracy server integration, distribution, application integration, distribution, equipment update, enhancement, abolition, etc. can be performed for energy saving.

  The configuration and operation of the operation management apparatus 1010 according to the embodiment have been described above. This embodiment is an exemplification, and it is understood by those skilled in the art that various modifications can be made to each component and combination of processes, and such modifications are within the scope of the present invention.

  In the embodiment, the description has been made based on the flow of packets from the user terminal 1006 to the server group 1022. However, when returning a packet from the server group 1022 to the user terminal 1006, the load balancer 1030 refers to the connection holding unit 1324 as appropriate. It will be apparent to those skilled in the art who have touched this specification that address translation is possible.

  In the embodiment, a case where the access threshold value Mo, the first threshold value T1, the second threshold value T2, and the third threshold value T3 are threshold values that determine the number of servers in the active state will be described. However, it is not limited to this. For example, each threshold value may have hysteresis. In this case, even if the predicted total number of accesses fluctuates in the vicinity of the threshold value, it is not necessary to switch the state of the server group every time the threshold value is crossed, so that the overhead associated with the state switching can be reduced. As a result, the overall response of the information processing system 2 can be improved.

  In the embodiment, the state in which the server group 1022 is placed has been described in the case where the strategy shown in Table 1 is used in the state setting unit 1140. However, the present invention is not limited to this. For example, the OS sleep state may be used as the power saving state, and the power off state may not be used. In this case, since there is no relatively long overhead for power on / off, a faster response can be expected. Further, the processing is simplified. As another example, the power-off state may be used as the power saving state, and the OS sleep state may not be used. In this case, power consumption can be further reduced.

  In the embodiment, the front end server 1024, the application server 1026, and the database server 1028 are separate servers, and the case where they are connected in series in this order has been described. However, the present invention is not limited to this. Each server group may include at least one server. For example, the server group may include one server having all functions of a front-end server, an application server, and a database server. The server group may include a server having any two functions of the functions of the three servers and a server having the remaining functions.

  In the embodiment, the case where the load prediction unit 1134 uses the attribute as a key has been described, but it may be a time zone. The load prediction unit 1134 refers to the load history holding unit 1112, acquires the load history of the same time in the same month in the past for the prediction target time zone, and based on this, the prediction total of the prediction target time zone Determine the number of accesses. For example, when the load prediction unit 1134 predicts the total number of accesses at 9:15 to 9:30 on July 28, 2009, the load prediction unit 1134 at 9:15 to 9:30 on July 28, 2008 The total access number (5400 in the case of FIG. 14) is acquired from the load history holding unit 1112 and is set as the predicted total access number.

In the embodiment, as an example in which the information processing system 2 is used, a securities business or a search engine is given, but the present invention is not limited to this. The technical idea according to the present embodiment can be applied to, for example, management of computer center equipment used by an unspecified number of users. An example of this computer center is a server and storage system used by many students and researchers in academic relations such as universities.
FIG. 22 is a data structure diagram showing the load history holding unit 1112c when an embodiment of the present invention is applied to a shared data center of a university. In this case, it is desirable to select an attribute according to the university schedule in order to grasp the tendency of the workload more accurately.

  In the embodiment, the case where the request processing unit is a server group has been described. However, the present invention is not limited to this. The technical idea according to the present embodiment can be applied to, for example, GSLB (Global Server Load Balance). There, the request processing unit may itself be a system having a plurality of servers arranged in parallel. Further, in the embodiment, the case where the request is an access from a user has been described. However, the request is not limited to this, and may be different depending on a system to which the technical idea according to the present embodiment is applied. It can be said that the request is an instruction of processing to the processing subject.

  In the embodiment, the average power consumption, the maximum power consumption, and the power cost of the front-end server (WEB server) 1024 of the server group 1022 are adopted as the average power consumption, the maximum power consumption, and the power cost of the server group 1022, respectively. Although the case has been described, the present invention is not limited to this. For example, the power cost of each server included in the server group 1022 may be calculated, and an average value thereof may be adopted as the power cost of the server group 1022.

  In the embodiment, for example, as illustrated in FIG. 18, the case has been described in which the index calculation unit 100 performs data aggregation (S412) from packet interception (S402) as a series of processes, but the present invention is not limited thereto. . For example, the index calculation unit may perform from packet interception to data classification at any time, accumulate the classification result data, and aggregate the classification result data when a predetermined condition is satisfied. Here, the predetermined condition is, for example, that a predetermined period has elapsed since the previous aggregation, or that the amount of classification result data has reached a predetermined amount.

FIG. 23 is a flowchart showing a series of processes in the index calculation unit according to the modification. The index calculation unit according to the modification includes an intercept unit, an extraction unit, a registration unit, a communication information holding unit, a classification unit, a classification result holding unit, and a counting unit. The intercepting unit intercepts the packet from the network (S502). The extraction unit extracts the start point address and the port number from the intercepted packet (S504). The registration unit registers the extracted start point address, port number, and packet data amount in the communication information holding unit (S506). The classification unit sorts the data held in the communication information holding unit with the start point address (S508). The classification unit classifies the sorted data into an original use and a non-original use (S510). The classification unit registers the classification result in the classification result holding unit (S512). If the predetermined condition is not satisfied (N in S514), the process returns to step S502. Thereby, the classification result is accumulated in the classification result holding unit until a predetermined condition is satisfied. When the predetermined condition is satisfied (Y in S514), the totaling unit totals the data accumulated in the classification result holding unit (S516).
According to the present modification, the amount of classification result data used as a total population can be adjusted by changing predetermined conditions. Therefore, it is possible to flexibly adjust the amount of classification result data according to the accuracy required for the index obtained as a result of aggregation.

  Although the present invention has been described based on the embodiments, the embodiments merely show the principle and application of the present invention, and the embodiments are defined in the claims. Needless to say, many modifications and arrangements can be made without departing from the spirit of the present invention.

  2 Information processing system, 100 index calculation unit, 102 network interface unit, 104 communication information holding unit, 108 first index holding unit, 110 usage status acquisition unit, 112 usage status holding unit, 114 coefficient holding unit, 116 first analysis unit 118 Second index holding unit, 120 Display control unit, 130 Totaling unit, 132 Classification unit, 138 Node information holding unit, 140 Monitor, 152 Second analysis unit, 158 Third index holding unit, 1006 User terminal, 1010 Operation management Apparatus, 1020 information processing apparatus, 1030 load balancer, 1130 learning unit, 1140 state setting unit, 1150 deviation determination unit, 1160 override unit.

Claims (4)

A load history holding unit for holding a load history of requests for an information processing apparatus configured by connecting a plurality of servers via a network;
A load prediction unit that predicts a load that will occur in the future with reference to the load history holding unit;
A state setting unit configured to set at least one server in a second state of power saving rather than the first state capable of accepting a request when the load predicted by the load prediction unit is less than a predetermined first threshold; ,
A usage status acquisition unit that acquires the temperature of the server from information obtained from the network,
The state setting unit preferentially selects a server having a high temperature acquired by the use state acquisition unit as a server to be set to the second state. A load history holding unit for holding a load history of requests for an information processing apparatus configured by connecting a plurality of servers via a network;
A load prediction unit that predicts a load that will occur in the future with reference to the load history holding unit;
A state setting unit configured to set at least one server in a second state of power saving rather than the first state capable of accepting a request when the load predicted by the load prediction unit is less than a predetermined first threshold; ,
Based on the information obtained from the network, the amount of work of the server is classified into the amount of work for the original purpose predetermined for the server and the amount of work for the non-original purpose that is a use other than the original purpose. An index calculation unit that calculates an increase in the temperature of the server due to processing of one unit of work for use, and
The operation management apparatus, wherein the state setting unit preferentially selects a server having a large increase value calculated by the index calculation unit as a server to be set to the second state. A load management device that obtains a request for an information processing device configured by connecting a plurality of servers through a network from outside, and assigns the obtained request to at least one of the plurality of servers;
An analysis device that acquires the temperature of the server from information obtained from the network,
The analysis device includes:
A unit selection unit that selects at least one server that does not require execution of processing corresponding to the request among a plurality of servers;
A unit setting unit that sets at least one server selected by the unit selection unit from an operating state in which a request can be received to a power saving state that is lower than the operating state; and
The unit selection unit sets the order in which the power saving state is set for each server so that the higher the temperature acquired by the analysis device is, the higher the temperature is, the power saving state is set.
When the requested load is less than a predetermined value, the unit selection unit selects a server in the operating state that has the highest power saving state as a server that does not need to execute processing corresponding to the request. An information processing system characterized by A load management device that obtains a request for an information processing device configured by connecting a plurality of servers through a network from outside, and assigns the obtained request to at least one of the plurality of servers;
From the information obtained from the network, for each server, the amount of work for the server is determined in advance for the amount of work for the original use and the amount of work for the non-original use that is a use other than the original use. An analysis device that classifies and calculates an increase in the temperature of the server due to the processing of one unit of work for intended use,
The analysis device includes:
A unit selection unit that selects at least one server that does not require execution of processing corresponding to the request among a plurality of servers;
A unit setting unit that sets at least one server selected by the unit selection unit from an operating state in which a request can be received to a power saving state that is lower than the operating state; and
The unit selection unit sets the order in which the power saving state is set for each server so that the increase value calculated by the analysis device is larger and the power saving state is preferentially set.
When the requested load is less than a predetermined value, the unit selection unit selects a server in the operating state that has the highest power saving state as a server that does not need to execute processing corresponding to the request. An information processing system characterized by

Images