JP2011118525A - Server management apparatus, server management method, and server management program - Google Patents

Abstract

【Task】
When the load of the management target deviates from the threshold, a change in the load of the management target is predicted based on the past load fluctuation history, and control related to the physical server is executed according to the prediction result.
[Solution]
The load information collection unit 130 collects the loads of the physical server 110 and the virtual server 115, stores the loads in the load information table 132 in association with the time when each load was collected, and the similar load information selection unit 136 When the load to be managed deviates from the scale-out threshold value or the scale-in threshold value, past load information similar to the load at the current time is selected from the load information table 132, and the scale-out determination unit 134 or the scale-in determination unit 135 is selected. Assuming that the load to be managed changes according to the selected load information, and when it is predicted that the load to be managed will still deviate from any threshold value, scale out or scale in is executed.
[Selection] Figure 1

Description

  The present invention relates to management of a physical server and a virtual server operating on the physical server, and more particularly, to a server management apparatus, a server management method, and a server management program for controlling fluctuations in load between the physical server and the virtual server.

  In recent years, a virtualization technique for operating a plurality of virtual servers on one physical server has been widely introduced. By adopting this virtualization technology, it is possible to move a virtual machine operating on one physical machine to another physical machine for operation. This makes it possible to distribute the load to a plurality of physical servers so that the load is not concentrated on a specific physical server. In addition, by consolidating the processing load distributed over multiple physical servers into a single physical server, it is possible to turn off physical servers that are excluded from the aggregation target, thus reducing power consumption. It becomes.

As a technique for allocating a plurality of virtual servers having different loads to a smaller number of physical servers, there is Patent Document 1. In Patent Document 1, a problem called “bin packing problem” is generally applied to a virtual server allocation problem to create a virtual server migration plan.

By the way, in a cluster system, scale-in may be executed as a process of stopping request allocation for a virtual server operating on a physical server and stopping the virtual server. When there is no virtual server running on the physical server due to the scale-in, the physical server can be powered off, and power consumption can be reduced. In a cluster system, scale-out may be executed as a process of starting a new virtual server on a physical server and starting request allocation to the started virtual server. When the scale-out is executed, the number of virtual servers operating on the physical server increases, so that the processing amount of the cluster system can be increased.

However, since a load for migrating a virtual server is applied to scale-in / scale-out, it is preferable that the number of scale-in / scale-out is small. When scale-in / scale-out frequently occurs, a CPU (Central Processing Unit) load, an I / O (Input / Output) load, a transition time increase, and power consumption increases.

In addition, scale-in / scale-out may involve starting and stopping of a physical server. In this case, power consumption further increases. For example, when a temporary (sudden, instantaneous, exceptional) load fluctuation occurs, the physical server may be in a high load state or a low load state in a short time, and scale-in / scale-out may occur.

As a technique for preventing scale-in / scale-out due to frequent load fluctuations, there is Patent Document 2. In Patent Document 2, a virtual server is moved when the load of a physical server exceeds a predetermined threshold and the state continues for a certain period. Since the load on the physical server is reduced by the movement of the virtual server, it is possible to suppress scale-in / scale-out caused by temporary (for example, instantaneous) load fluctuations.

JP 2008-21001 A JP 2009-116380 A

However, since the technique disclosed in Patent Document 2 is designed to perform scale-in / scale-out of a virtual server when the state where the load of the physical server exceeds the threshold for a certain period of time continues, Although there is an advantage of suppressing excessive scale-in / scale-out against sudden load fluctuations that occur in the case of the load, if the load has exceeded the threshold for a long time, that is, the scale-in / scale-out Even if it is preferable to execute out, scale-in and scale-out are suppressed for a certain period. As a result, the state in which the power of the physical server is turned on for a certain period of time continues, and power may be consumed unnecessarily.

In the technique disclosed in Patent Document 2, if the determination period (the above-mentioned fixed time) is set short in order to shorten the scale-in / scale-out suppression time, the original scale-in / scale-out is suppressed. Even if it is preferable to do so, there is a risk of performing scale-in / scale-out.

In this regard, in Patent Document 2, there is no specific disclosure regarding adjustment of a certain period, which is a condition for executing scale-in / scale-out, and there remains a problem that it is left to the judgment of the user or developer.

  The present invention has been made in view of the above-described problems of the prior art, and its purpose is to change the load of the management target after that when the load of the management target deviates from the threshold. An object is to provide a server management apparatus, a server management method, and a server management program capable of performing prediction based on a history and executing control relating to a physical server according to the prediction result.

In order to achieve the above object, the present invention collects the load of a physical server and the load of a virtual server, and associates each collected load with the time at which each load is collected to load information as load information. Stored in a table, monitoring changes in the load of the management target among the loads along the time axis, and when the load of the management target deviates from a threshold, the load of the management target at that time is As the load, the load information table is searched, and load information related to a load similar to the load at the current time is selected from among the loads that are past the current time and collected at the same time as the current time. Assuming that the load of the management target changes according to the selected load information, the load of the management target tends to deviate from the threshold thereafter. Whether to predict, and executes control related to the physical server in accordance with the prediction result.

ADVANTAGE OF THE INVENTION According to this invention, the control regarding a physical server can be performed efficiently with high precision.

1 is a block configuration diagram of a cluster system to which the present invention is applied. FIG. It is a block diagram of a structure information table. It is a block diagram of a load information table. It is a flowchart for demonstrating the process of the similar load information selection part which selects similar load information from the past load information. It is a flowchart for demonstrating the process of the scale out judgment part which controls scale out from the past load information. It is a characteristic view which shows the fluctuation state of the load in the case of performing scale-out. It is a flowchart for demonstrating the process of the scale-in determination part which controls scale-in from the past load information. It is a characteristic view which shows the fluctuation state of the load in the case of performing scale-in.

  In this embodiment, the physical server load and the virtual server load are collected respectively, and the collected loads are stored in the load information table as load information in association with the time when each load is collected. In the process of changing along the axis, when the load to be managed is out of the threshold among the loads, it is a past load from the current time at that time, and the load collected at the same time as the current time. From the load information table, load information related to the load similar to the load at the current time is selected, and the load to be managed is changed from the threshold value on the assumption that the load to be managed changes according to the selected load information. When it is predicted that there is a tendency, scale-out or scale-in is executed as control related to the physical server.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a cluster system to which the present invention is applied.

  In FIG. 1, the cluster system includes a server management apparatus 120, a plurality of physical servers 110, and a load distribution apparatus 102. Each physical server 110 is connected to the server management apparatus 120 and load distribution is performed. It is connected to the network 101 via the device 102.

Each physical server 110 includes a CPU (processing unit) 111, a memory (storage unit) 114, a hard disk (storage unit) 112, and a communication interface 113. The CPU 111 is connected to the server via the communication interface 113. Exchanges information with the management device 120.

In the memory 114 of each physical server 110, a plurality of virtual servers 115 are virtualized by the virtualization software 117 so as to operate on each physical server 110. That is, one or two or more virtual servers 115 operate on each physical server 110. Each virtual server 115 includes a load information collection unit 116 that collects the load of each virtual server 115.

When receiving a request from the network 101, the load distribution apparatus 102 allocates the received request to the virtual server 115 of each physical server 110 via each physical server 110.

The server management device 120 includes a memory (storage unit) 121, a CPU (processing unit) 122, a communication interface 123, and a hard disk (storage unit) 124. Information is exchanged with the physical server 110 and each virtual server 115.

  The memory 121 includes a load information collection unit 130, a configuration information collection unit 131, a load information table 132, a configuration information table 133, a scale-out determination unit 134, a scale-in determination unit 135, and a similar load information selection unit 136. And.

The load information collection unit 130, the configuration information collection unit 131, the scale-out determination unit 134, the scale-in determination unit 135, and the similar load information selection unit 136 are configured as programs executed by the CPU 122.

The configuration information collection unit 131 collects configuration information regarding the configuration of the virtual server 115 allocated on the physical server 110 at the current time from each physical server 110 by the execution of the CPU 122, and the collected configuration information is stored in the configuration information table 133. To accumulate.

FIG. 2 shows the configuration of the configuration information table 133. The configuration information table 133 is assigned in advance to identify the physical server ID (IDentifire) field 202 assigned in advance to identify each physical server 110 and the virtual server 115 included in each physical server 110. And a virtual server ID field 203.

  In each entry of the physical server ID field 202, as an identifier of the physical server 110, for example, when there are three physical servers 110, "physical server 1", "physical server 2", "physical server 3" Is stored.

Each entry of the virtual server ID field 203 stores the identifier of the virtual server 115. For example, the entries corresponding to “physical server 1” store “virtual servers 1 to 3” as virtual servers belonging to “physical server 1”, and the entry corresponding to “physical server 2” includes “physical server 1”. “Virtual servers 4 to 5” are stored as virtual servers belonging to “server 2”, and “virtual servers 6 to 7” are stored as virtual servers belonging to “physical server 3” in entries corresponding to “physical server 3”. Is stored.

The load information collection unit 130 collects and collects load information of the physical server 110 and the virtual server 115 at regular intervals in cooperation with the load information collection unit 118 in the physical server and the load information collection unit 116 in the virtual server. The load information is stored in the load information table 132.

Specifically, the load information collection unit 118 in each physical server 110 collects the load of each physical server 110 from each physical server 110 and the collected load at the time (time) when the load is collected. Correspondingly, load information is transferred to the server management apparatus 120 via the communication interface 113. The load information collection unit 116 in the virtual server 115 collects the load of each virtual server 115 from one or more virtual servers 115 operating on each physical server 110, and collects the collected load. The load information is transferred to the server management apparatus 120 via the communication interface 113 in association with the time (time).

As long as the load information of the virtual server 115 can be acquired via the virtualization software 117, the load information collection unit 118 may be provided outside the virtual server 115.

  The load information collection unit 130 of the server management device 120 receives the load information collected by the load information collection unit 116 via the communication interface 123, stores the received load information in the load information table 132, and collects the load information. The load information collected by the unit 118 is received via the communication interface 123, and the received load information is stored in the load information table 132.

At this time, the load information collection unit 130 indirectly collects the load of each physical server 110 and the load of the virtual server 115 operating on each physical server 110, and collects each collected load. It is stored in the load information table 132 as load information in association with time (time) and date.

Here, the load includes the CPU usage rate of the physical server 115 and the virtual server 115, the memory usage rate, the disk usage rate, the network bandwidth usage rate, the number of queues for processing applications running on the virtual server 115, and the connection to the database. Number, occurrence frequency of Full GC (Garbage Collection), number of sessions, etc.

For each load, a scale-in threshold value and a scale-out threshold value are set in advance as threshold values.

FIG. 3 shows the configuration of the load information table 132. The load information table 301 includes a time field 302, a server ID field 303 of the physical server 110 that collected the load, a CPU usage rate field 304, and a memory usage rate field 305.

  Each entry in the time field 302 stores the time (time) when the physical server 110 or the virtual server 115 collected the load together with the date. Each entry in the server ID field 303 stores the server ID of the physical server 110 or virtual server 115 that collected the load, for example, “physical server 1”, “physical server 2”, “physical server 3”, and the like. Each entry of the CPU usage rate field 304 stores the CPU usage rate of the physical server 110 or virtual server 115 from which the load is collected, for example, “10%”, “20%”, “15%”, and the like. Each entry of the memory usage rate field 305 stores the memory usage rate of the physical server 110 or the virtual server 115 that collected the load, for example, “15%”, “10%”, “25%”, and the like.

  The scale-out determination unit 134 and the scale-in determination unit 135 monitor the change of each load due to the collection of the load information collection unit 130 along the time axis, and among the loads, for example, the load of each physical server 110 is a threshold value. If the load of each physical server 110 is determined to be out of the threshold value, each physical server 110 is loaded according to the load information selected by the similar load information selection unit 136 (load information indicating the past load fluctuation history). Assuming that the load of the server 110 changes, it is predicted whether or not the load of each physical server 110 tends to deviate from the threshold value thereafter, and a control unit that executes control related to each physical server 110 according to the prediction result Consists of one element.

  Specifically, for example, when the load on each physical server 110 exceeds the scale-out threshold (when the load exceeds the scale-out threshold), the scale-out determination unit 134 determines that the load on each physical server 110 has deviated from the threshold. , Output a positive determination result to the similar load information selection unit 136, and predict whether the load of each physical server 110 tends to exceed the scale-out threshold thereafter. When obtained, scale-out is executed as control for increasing the number of virtual servers 115 operating on each physical server 110.

  For example, when the load on each physical server 110 falls below the scale-in threshold (less than the scale-in threshold), the scale-in determination unit 135 determines that the load on each physical server 110 has deviated from the threshold, and gives a positive determination result. When outputting to the similar load information selection unit 136 and predicting whether the load of each physical server 110 tends to be lower than the scale-in threshold after that, when a positive result is obtained in this prediction, As control for reducing the number of virtual servers 115 operating on the physical server 110, scale-in is executed.

  On the other hand, when the similar load information selection unit 136 makes a positive determination in the scale-out determination unit 134 or the scale-in determination unit 135, the load at that time, for example, the load of each physical server 110 is set as the load at the current time. The load information table 132 is searched, and load information related to a load similar to the load at the current time is selected from the load information table 132 from among the loads that are past the current time and collected at the same time as the current time. To do.

Hereinafter, the processing of the similar load information selection unit 136 will be described with reference to the flowchart of FIG.

First, the similar load information selection unit 136, for example, a similar load related to a load similar to the load at the current time on the condition that the load of the physical server 110 exceeds the scale-out threshold or falls below the scale-in threshold. A process for selecting information is started (step 401), and a plurality of days as candidate days are selected from the days belonging to the past load information (step 402). For example, the similar load information selection unit 136 selects the same day last month, the same day last week, and the third day of the previous day as a plurality of candidate days from the days belonging to the past load information.

Next, the similar load information selection unit 136 adopts the following method when selecting similar load information regarding a load similar to the load at the current time. For example, the difference between the load until a certain time T before the current time t _TODAY and the load between the same time in the past (from t′-T to t ′) is obtained. At this time, assuming that the load at time t is L _t− , the difference Diff (t ′) between the load up to time t _TODAY and the load up to time t ′ can be calculated as follows.

Using this calculation method, the similar load information selection unit 136 uses the load at the current time as a reference, the load at the same time t _LAST-MONTH on the same day last month, the load at the same time t _{LAST-WEEK on} the same day last week, The difference from the load at the same time t _YESTERDAY of the previous day is calculated. At this time, the similar load information selection unit 136 calculates the difference between the load at the current time and the load on each candidate day for all candidate dates (step 403).

Next, the similar load information selection unit 136 selects a candidate date having the smallest load difference based on each calculation result from the plurality of candidate dates, and sets the selected candidate date load information as similar load information. Selection is made (step 404), and the processing in this routine is terminated (step 405).

Thereby, when the load on the physical server 110 exceeds the scale-out threshold or falls below the scale-in threshold, the similar load information selection unit 136 sets the load on the physical server 110 at that time as the current time load. The load information table 132 is searched, and similar load information related to a load similar to the load at the current time is retrieved from the load information table 132 from among the loads that are past the current time and collected at the same time as the current time. You can choose.

Next, the process of the scale-out determination unit will be described with reference to the flowchart of FIG.

When the scale-out determination unit 134 starts the determination of scale-out (step 501), the scale-out determination unit 134 monitors each load collected by the load information collection unit 130 along the time axis, and among the loads, for example, the latest of the physical server 110 It is determined whether or not the load exceeds the scale-out threshold (step 502).

If the scale-out determination unit 134 determines in step 502 that the latest load on the physical server 110 is equal to or less than the scale-out threshold (No), the scale-out determination unit 134 ends the processing in this routine without executing the scale-out. (Step 508) If it is determined that the latest load of the physical server 110 has exceeded the scale-out threshold (Yes), the load information table 132 is searched, and whether past load information exists based on the search result. It is determined whether or not (step 503).

When the scale-out determination unit 134 determines in step 503 that past load information does not exist (NO), the scale-out determination unit 134 proceeds to the processing in step 507 and executes scale-out.

That is, when it is determined that there is no past load information, it is not possible to determine whether to execute scale-out based on the past load information, but it is determined that the load of the physical server 110 exceeds the scale-out threshold. Consider and scale out.

On the other hand, when the scale-out determination unit 134 determines in step 503 that the past load information exists (Yes), the similar load information related to the load most similar to the load at the current time among the past load information. Is requested to the similar load information selection unit 136.

In response to the request from the scale-out determination unit 134, the similar load information selection unit 136 searches the load information table 132, and from among the past load information of the physical server 110 and the past load information of the physical server 110. The load information that is most similar to the load information at the current time is selected as the similar load information. The scale-out determination unit 134 selects a time that is the same as the current time among the times belonging to the similar load information selected by the similar load information selection unit 136 (step 504).

For example, as shown in FIG. 6, the scale-out determination unit 134 increases the load on the physical server 110 above the scale-out threshold th1 at time 602 in the process of changing the load on the physical server 110 along the time axis. The similar load information selection unit 136 is requested to select similar load information related to the load most similar to the load at the current time.

The similar load information selection unit 136 searches the load information table 132 in response to the request from the scale-out determination unit 134, and loads the load information at the current time (time 602) from the past load information of the physical server 110. Is selected as the similar load information.

The scale-out determination unit 134 selects, as t ₁ , a time that is the same as the current time among the past days belonging to the similar load information selected by the similar load information selection unit 136 (step 504).

Here, the time of the past day is the same time as the current time among the past times than the current time, and is the same day of the previous month, the same day of the previous week, and the time of the previous day. For example, if the current time is 2:00 pm, the time of the past day is 2:00 pm on the same day last month, 2:00 pm on the same day last week, and 2:00 pm on the previous day.

When past load information exists in the load information table 134, the scale-out determination unit 134 changes the load of the physical server 110 according to past load information (similar load information selected by the similar load information selection unit 136). Assuming that the future load on the physical server 110 is predicted, it is determined whether or not to perform scale-out based on the prediction result.

At this time, if the scale-out for increasing the number of virtual servers 115 is executed, the computational resources for processing the request increase, so the load on the physical server 110 can be reduced. When the physical server 110 is activated due to out, power consumption increases correspondingly. Therefore, it is preferable not to scale out as much as possible.

Therefore, in the cluster system to which the present invention is applied, the temporary load increase does not need to be scaled out. For example, as shown in FIG. 6, the load on the physical server 110 changes along the time axis. When the load on the physical server 110 exceeds the scale-out threshold th1 at time 602, the similar load information selection unit 136 searches the load information table 132 and includes the past load information on the physical server 110. The load information most similar to the load information at the current time (time 602) is selected as past similar load information.

After the past similar load information is selected, as shown in FIG. 6, when the load on the physical server 110 at the current time exceeds the scale-out threshold th1 at time 602, the scale-out determination unit 134 assumed to vary according to past similar load information of the server 110, of the time of the previous day belonging to past similar load information, a time at which the current time (time 602) and the same time, with the t _1, time A fixed time 605 from t ₁ to time t ₂ (t ₂ = t ₁ + load drop search time) is set as a load drop search time (step 505), and between time t ₁ and time t _{2 when} the fixed time 606 elapses. Then, it is checked whether or not the past load of the physical server 110 falls and becomes equal to or smaller than the scale-out threshold th1 (step 506).

Here, FIG. 6 is a characteristic diagram of past similar load information selected by the similar load information selection unit 136, and a graph 601 represents past load characteristics of the physical server 110. FIG. For this reason, when the load on the physical server 110 at the current time exceeds the scale-out threshold th1, the scale-out determination unit 134 assumes that the load changes thereafter according to the graph 601 that is past similar load information, and the time of the graph 601 among the, the time at which the current time (time 602) and the same time, is set to t _1.

When the past load of the physical server 110 temporarily increases from time 602 and then decreases and then decreases to the scale-out threshold th1 or less from time 603 (Yes), the scale-out determination unit 134 The server 110 determines that the increase in the past load is temporary, suppresses the scale-out, and terminates the processing in this routine without executing the scale-out (step 508).

Next, when the load on the physical server 110 exceeds the scale-out threshold th1 at time 604, the scale-out determination unit 134 requests the similar load information selection unit 136 to select similar load information again. As a result, the similar load information selection unit 136 searches the load information table 132, and from the past load information of the physical server 110, the load information most similar to the load information at the current time (time 604) is stored in the past. As similar load information.

After the past similar load information is selected, the scale-out determination unit 134 changes according to the past similar load information of the physical server 110 when the load of the physical server 110 at the current time exceeds the scale-out threshold th1. then assuming, of times of day in the past that belong to past similar load information, a time at which the current time (time 604) and the same time with the t _1, the load descends search time for a predetermined time 606 from the time t ₁ and then, during the time t ₂ when the predetermined time 606 has elapsed, past load of the physical server 110 checks whether the scale-out threshold th1 or less lowered (step 506).

In step 506, past load of the physical server 110, between times t ₂ for a predetermined time 606 has elapsed from the time t _1, not the scale-out threshold th1 or less, when subsequently also predicted to be on the rise ( No), the scale-out determination unit 134 determines that the increase in the past load on the physical server 110 is not temporary, and performs scale-out as control for increasing the number of virtual servers 115 operating on the physical server 110. Is executed (step 507).

As described above, when the past load of the physical server 110 exceeds the scale-out threshold value th1, the scale-out determination unit 134 tends to exceed the scale-out threshold value th1 over a certain period of time 606. When predicted, scale-out is executed as control for increasing the number of virtual servers 115 operating on the physical server 110. When the past load of the physical server 110 exceeds the scale-out threshold th1, If the number of virtual servers 115 operating on the physical server 115 is predicted to be less than or equal to the scale-out threshold th1 after the increase, the execution of control for increasing the number of virtual servers 115 operating on the physical server 115 is suppressed.

Next, the processing of the scale-in determination unit will be described with reference to the flowchart of FIG.

When the scale-in determination is started (step 701), the scale-in determination unit 135 monitors changes in loads by the collection of the load information collection unit 130 along the time axis. For example, the latest load of the physical server 110 is It is determined whether or not the scale-in threshold has been exceeded (step 702).

In Step 702, when the scale-in determination unit 135 determines that the load on the physical server 110 is equal to or greater than the scale-in threshold (No), the scale-in determination unit 135 ends the processing in this routine without executing the scale-in (Step S702). 711), when it is determined that the load of the physical server 110 has fallen below the scale-in threshold (Yes), the load information table 132 is searched, and whether or not past load information exists based on the search result is determined. Determination is made (step 703).

When the scale-in determination unit 135 determines in step 703 that the past load information does not exist (NO), the scale-in determination unit 135 cannot determine whether to execute the scale-in based on the past load information. Considering that the load is below the scale-in threshold value, the process proceeds to step 710 to execute scale-in, and then the process in this routine is terminated (step 711).

On the other hand, when the scale-in determination unit 135 determines in step 703 that past load information exists (Yes), the scale-in determination unit 135 selects a load most similar to the load at the current time from the past load information. Is requested to the similar load information selection unit 136.

The similar load information selection unit 136 searches the load information table 132 in response to the request from the scale-in determination unit 135, and is most similar to the load information at the current time from the past load information of the physical server 110. the load information is similar load selected as information to select a time t ₁ comprising the current time and the time of the time in the past day belong to the selected similar load information (step 704).

For example, as illustrated in FIG. 8, the scale-in determination unit 135 is configured such that the load on the physical server 110 falls below the scale-in threshold th2 at time 802 in the process in which the load on the physical server 110 changes along the time axis. The similar load information selection unit 136 is requested to select a load most similar to the load at the current time.

The similar load information selection unit 136 searches the load information table 132 in response to the request from the scale-in determination unit 135, and loads the load information at the current time (time 802) from the past load information of the physical server 110. Is selected as the past similar load information.

After the past similar load information is selected, the scale-in determination unit 135 sets a time that is the same as the current time among the past days belonging to the similar load information selected by the similar load information selection unit 136, It is selected as t ₁ (step 704).

Thereafter, Scale In determining unit 135, based on the load information and the time t ₁ of the last selected in a similar load information selecting unit 136, predicts future changes in the load of the physical server 110, based on the prediction result Whether or not to perform scale-in is determined. At this time, since power is always consumed while the physical server 110 is running, it is desirable to reduce power consumption by performing scale-in as much as possible.

Therefore, in the cluster system to which the present invention is applied, the scale-in determination unit 135 executes the scale-in when the load decreases for a certain period of time. If there is a temporary increase in load while the load is decreasing, it is determined that the load should decrease over a long period of time and that the scale-in should be executed. .

Specifically, for example, as shown in FIG. 8, when the load on the physical server 110 changes along the time axis, the load on the physical server 110 falls below the scale-in threshold th2 at time 802. The scale-in determination unit 135 requests the similar load information selection unit 136 to select similar load information.

In response to the request from the scale-in determination unit 135, the similar load information selection unit 136 searches the load information table 132 to obtain the load information at the current time (time 802) from the past physical server load information. The most similar load information is selected as past similar load information.

After the past similar load information is selected, the scale-in determination unit 135 changes according to the past similar load information of the physical server 110 when the load of the physical server 110 at the current time falls below the scale-in threshold th2. Assuming that the time t, which is the same time as the current time (time 802), of the past days belonging to the similar load information in the past is t = t ₁ and from time t ₁ to time t ₂ the predetermined time 807 as a load increase search time (step 705), during the time t ₂ when the predetermined time 807 from the time t ₁ has elapsed, past load of the physical server 110, increased scale in threshold th2 or more as a It is checked whether or not (step 706).

Here, FIG. 8 is a characteristic diagram of past similar load information selected by the similar load information selection unit 136, and a graph 801 represents past load characteristics of the physical server 110. FIG. Therefore, when the load of the physical server 110 at the current time falls below the scale-in threshold th2, the scale-in determination unit 135 assumes that the load changes thereafter according to the graph 801 that is past similar load information, and the time of the graph 801 of, the time t as the current time (time 802) and the same time, is set to t _1.

In step 706, during the time t ₂ for a predetermined time 807 has elapsed from the time t _1, past load of the physical server 110, together below the scale in threshold th2, that then, if the lower scale in threshold th2 When predicted (No), the scale-in determination unit 135 executes scale-in (step 710).

On the other hand, when it is predicted in step 706 that the past load on the physical server 110 temporarily decreases from time 802 and then increases and then increases to the scale-in threshold th2 or more from time 803 (Yes ) scale in determining unit 135, a time at which the current time (time 803) and the same time, with the t _3, the predetermined time 808 from time t ₃ to time t ₄ and the load descends search time (step 707 ), during the time t ₄ when a certain time 808 elapses from the time t _3, past load of the physical server 110 checks whether or not lower than the scale in threshold th2 is lowered (step 708).

In step 708, during the time t ₄ when a predetermined time 808 from the time t ₃ has elapsed, past load of the physical server 110, when it is predicted that there is no fall below Scale In threshold th2 (No), Scale In decision The unit 135 ends the processing in this routine without executing the scale-in (step 711).

In step 708, during the time t ₄ when a certain time 808 elapses from the time t _3, past load of the physical server 110, when it is predicted that there is a fall below Scale In threshold th2 (Yes), Scale In decision parts 135, past load of the physical server 110, the time t falls below the scale in threshold th2 as t = t ₁ (step 709), the process returns to step 706.

Next, when the past load of the physical server 110 falls below the scale-in threshold th2 at time 804, the scale-in determination unit 135 requests the similar load information selection unit 136 to select similar load information again.

The similar load information selection unit 136 searches the load information table 132 in response to the request from the scale-in determination unit 135, and loads the load information at the current time (time 804) from the past load information of the physical server 110. Is selected as past similar load information.

After the past similar load information is selected, the scale-in determination unit 135 changes according to the past similar load information (graph 801) when the load of the physical server 110 at the current time falls below the scale-in threshold th2. then assuming, of times of day in the past that belong to past similar load information, certain time t comprising the current time (time 804) and the same time, with the t _1, from time t ₁ to time t ₂ time 809 and load increase search time (step 705), during the time t ₁ to time t ₂ to a predetermined time 809 has elapsed, or past load of the physical server 110, a scale-in threshold th2 or elevated Whether or not is checked (step 706).

When the past load of the physical server 110 temporarily decreases from time 804, then increases, and then increases to the scale-in threshold th2 or more from time 805, the scale-in determination unit 135 determines the current time (time time the time 805) and the same time, with the t _3, the predetermined time 810 from time t ₃ to time t ₄ and the load descends search time (step 707), a fixed time 810 from the time t ₃ has elapsed between t _4, past load of the physical server 110 checks whether or not lower than the scale in threshold th2 is lowered (step 708).

Between time t ₃ of the time t ₄ when a predetermined time 808 has elapsed, past load of the physical server 110, when below the scale in threshold th2 at time 806, the scale-in determining unit 135, the time 806 to the time t Set (step 709), the process returns to step 706.

In step 706, the scale-in determining unit 135 sets the time 806 to the time t, while the time t time t _2, the last load of the physical server 110, together below the scale in threshold th2, even then, When it is predicted that the value falls below the scale-in threshold th2, that is, when the past load of the physical server 110 is predicted not to increase to the load at the time t ₁ (time 804), the scale-in Is executed (step 710).

  At this time, the scale-in determination unit 135 sets the time 804 as the scale-in start time, and executes the scale-in at the time 804.

That is, the scale-in determination unit 135 determines that the decrease in the past load on the physical server 110 is not temporary, and performs scale-in as control for reducing the number of virtual servers 115 operating on the physical server 110. Run at time 804.

As described above, when the scale-in determination unit 135 predicts that the load on the physical server 110 is below the scale-in threshold th2 and then tends to fall below the scale-in threshold th2 over a certain period of time. Even if the load on the physical server 110 may temporarily exceed the scale-in threshold th2 after that, if it is predicted that the load will fall below the scale-in threshold th2 over a long period of time, As a control to reduce the number of operating virtual servers 115, scale-in is executed, and when the load on the physical server 110 falls below the scale-in threshold th2, then temporarily decreases, then the scale-in threshold th2 or higher When it is predicted that the number of virtual servers 115 operating on the physical server 110 will be reduced, the execution of control that reduces the number of virtual servers 115 is suppressed.

  At this time, when the virtual server 115 does not exist on the physical server 110, the scale-in determination unit 135 can reduce power consumption by controlling the power of the physical server 110 to be off.

According to the present embodiment, when the load on the physical server 110 deviates from the threshold value, a subsequent change in the load on the physical server 110 is predicted based on the past load fluctuation history, and according to the prediction result, the physical server 110 Since the scale-out or the scale-in is executed as the control regarding the scale-out, the scale-out or the scale-in can be executed efficiently and with high accuracy.

  Further, according to the present embodiment, when it is determined that the currently occurring load fluctuation is temporary, scale-in or scale-out is suppressed, so unnecessary configuration changes can be reduced. Can do.

  In this embodiment, when the scale-out determination unit 134 determines whether to perform scale-out, or when the scale-in determination unit 135 determines whether to perform scale-in, the physical server 110 Describes whether the load exceeds the scale-out threshold th1 or whether the load falls below the scale-in threshold th2, but determines whether to perform scale-out or scale-in As a load to be managed, instead of the load of the physical server 110, the load of the entire physical server, the load of the virtual server 115 operating on the physical server 110, or the entire virtual server operating on the physical server 110 Can also be used.

  In this embodiment, when setting the scale-out threshold value th1 and the scale-in threshold value th2 as the threshold values of the load to be managed, both values can be the same, or both values can be different values. .

101 Network 102 Load Balancing Device 110 Physical Server 111 CPU
112 Hard disk 113 Communication interface 114 Memory 115 Virtual server 116 Load information collection unit 117 Virtualization software 118 Load information collection unit 120 Server management device 121 Memory 122 CPU
123 Communication interface 124 Hard disk 130 Load information collection unit 131 Configuration information collection unit 132 Load information table 133 Configuration information table 134 Scale-out determination unit 135 Scale-in determination unit 136 Similar load information selection unit 201 Configuration information table 202 Physical server ID
203 Virtual server ID
301 Load information table 302 Time 303 Server ID
304 CPU usage rate 305 Memory usage rate

Claims (13)

Load information is collected as load information by collecting the load of the physical server and the load of one or more virtual servers operating on the physical server, and associating the collected loads with the time at which the loads were collected. Load information collection unit to be stored in the table;
A determination unit that monitors changes in the load of the management target among the loads collected by the load information collection unit along a time axis, and determines whether or not the load of the management target deviates from a threshold value;
When the determination unit makes an affirmative determination, the load information table is searched using the load to be managed at that time as the load at the current time, and the load is a load earlier than the current time, and the current time and A load information selection unit that selects, from the load information table, load information related to a load similar to the load at the current time from among the loads collected at the same time,
The determination unit
Assuming that the load of the management target changes according to the load information selected by the similar load information selection unit, predict whether or not the load of the management target will tend to deviate from the threshold, and A server management apparatus that executes control related to the physical server according to a result. The server management apparatus according to claim 1,
The determination unit
When the load to be managed exceeds a scale-out threshold, the load to be managed is out of the threshold, and an affirmative determination result is output to the similar load information selection unit, and the load to be managed is Then, it is predicted whether or not it tends to exceed the scale-out threshold, and when the result of the prediction is affirmative, control is performed to increase the number of the virtual servers operating on the physical server. A server management device. The server management apparatus according to claim 1,
The determination unit
When the load to be managed falls below the scale-in threshold, the load to be managed is out of the threshold, and an affirmative determination result is output to the similar load information selection unit, and the load to be managed is Then, it is predicted whether or not it tends to fall below the scale-in threshold, and when a positive result is obtained in the prediction, control is performed to reduce the number of the virtual servers operating on the physical server. A server management device. The server management apparatus according to claim 2,
The determination unit
When it is predicted that the load to be managed will tend to exceed the scale-out threshold for a certain period thereafter, control for increasing the number of the virtual servers operating on the physical server is executed.
When the load to be managed is temporarily increased after that and is predicted to be equal to or less than the scale-out threshold, execution of control for increasing the number of the virtual servers operating on the physical server is suppressed. Server management device. The server management apparatus according to claim 3,
The determination unit
When it is predicted that the load to be managed will tend to be lower than the scale-in threshold for a certain period thereafter, or the load to be managed may temporarily become greater than or equal to the scale-in threshold thereafter. Even if it is predicted that there is a tendency to fall below the scale-in threshold over a long period of time, control is performed to reduce the number of the virtual servers operating on the physical server,
When the load to be managed is predicted to be equal to or larger than the scale-in threshold after the temporary decrease after that, the execution of control for reducing the number of virtual servers operating on the physical server is suppressed. Server management device. The server management apparatus according to claim 5,
The determination unit is a server management apparatus configured to control power-off of the physical server when the virtual server does not exist on the physical server. Load information is collected as load information by collecting the load of the physical server and the load of one or more virtual servers operating on the physical server, and associating the collected loads with the time at which the loads were collected. Load information collection unit to be stored in the table;
A determination unit that monitors changes in the load of the management target among the loads collected by the load information collection unit along a time axis, and determines whether or not the load of the management target deviates from a threshold value;
When the determination unit makes an affirmative determination, the load information table is searched using the load to be managed at that time as the load at the current time, and the load is a load earlier than the current time, and the current time and A load management information selection unit that selects, from the load information table, load information related to a load similar to the load at the current time from among the loads collected at the same time, and a server management method comprising:
The determination unit
Assuming that the load of the management target changes according to the load information selected by the similar load information selection unit, predicting whether the load of the management target tends to deviate from the threshold after that,
Executing a control on the physical server according to the prediction result of the step. The server management method according to claim 7,
The determination unit
Outputting a positive determination result to the similar load information selection unit, assuming that the load to be managed is out of the threshold when the load to be managed exceeds a scale-out threshold;
Predicting whether the load to be managed tends to exceed the scale-out threshold thereafter;
Executing a control to increase the number of the virtual servers operating on the physical server when a positive result is obtained in the prediction of the step. The server management method according to claim 7,
The determination unit
When the load to be managed falls below the scale-in threshold, the load to be managed is out of the threshold, and a step of outputting a positive determination result to the similar load information selection unit and the load to be managed are Predicting whether or not the scale-in threshold tends to be lower than the scale-in threshold afterwards;
Executing a control to reduce the number of the virtual servers operating on the physical server when a positive result is obtained in the prediction of the step. The server management method according to claim 8, wherein
The determination unit
Executing a control to increase the number of the virtual servers operating on the physical server when the load to be managed is predicted to tend to exceed the scale-out threshold for a certain period thereafter;
When the load to be managed is temporarily increased after that and is predicted to be equal to or less than the scale-out threshold, suppressing the execution of control for increasing the number of virtual servers operating on the physical server; A server management method comprising: The server management method according to claim 9, wherein
The determination unit
When it is predicted that the load to be managed will tend to be lower than the scale-in threshold for a certain period thereafter, or the load to be managed may temporarily become greater than or equal to the scale-in threshold thereafter. Even if it is predicted that there is a tendency to fall below the scale-in threshold over a long period of time, a step of executing control to reduce the number of the virtual servers operating on the physical server;
When the load to be managed is predicted to be equal to or larger than the scale-in threshold after the temporary decrease after that, suppressing the execution of control for reducing the number of virtual servers operating on the physical server; A server management method comprising: The server management method according to claim 11,
The determination unit
A server management method comprising a step of controlling the physical server to be turned off when the virtual server does not exist on the physical server.   The server management program for making a computer perform each step of any one of Claims 7-13.