JP2017062721A - Resource allocation changing apparatus, resource allocation changing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allocate an appropriate resource, in accordance with load estimation accuracy.SOLUTION: A resource allocation changing apparatus 10 includes: characteristic information storage means 15 which stores time determination information for determining allocation change time required for addition, in association with a processing resource to be added to processing in execution; load estimation means 12 which outputs an estimated value of the maximum load of the processing in execution in a predetermined time, and estimation accuracy of the estimated value; and allocation change determination means 13 which selects a processing resource having an allocation change time equal to or less than the upper limit, which is determined to be lower as the estimation accuracy is lower, in reference to the characteristic information storage means, to output the selection information to allocation change means for changing allocation of the selected processing resource.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a resource allocation change device, a resource allocation change method, and a program, in particular, in order to maintain processing performance at a high load, select and additionally allocate a plurality of processing resources to be allocated to a process. The present invention relates to a resource allocation changing device, a resource allocation changing method, and a program.

  Patent Document 1 provides a method of allocating a server computer in a data center to each network service in real time. In this method, the unit that determines the allocation first monitors the fluctuation of the request quantity that arrives at each network service, and predicts the value of the request quantity after a certain time has elapsed. Next, this unit controls the allocation amount of the server computer for the network service in accordance with the magnitude of the required amount prediction value. The amount of the server computer to be allocated is determined so that the response time to the user terminal is equal to or less than a predetermined threshold when the traffic indicated by the request amount prediction value arrives at the network service.

  Patent Document 2 discloses a cluster system management device that maintains the service level of a business by changing the allocation of computer resources when a load change occurs in the cluster system. This management apparatus first predicts the load on the cluster system and changes the allocation of computer resources in accordance with the prediction. When changing the resource allocation, this management device can process the predicted load amount while taking into consideration the constraints such as the amount of available computer resources and the time required to change the computer resource amount. Allocate more computer resources.

Japanese Patent No. 5666482 JP2011-258119A

  Neither the technique of patent document 1 nor the technique of patent document 2 considers the prediction accuracy of the load predicted at the time of resource allocation. The prediction accuracy of the predicted load varies depending on, for example, the load fluctuation state. When the prediction accuracy is low, there is a high possibility that the load prediction is incorrect.

  An object of the present invention is to solve the above-described problems and provide a resource allocation change device, a resource allocation change method, and a program that perform appropriate resource allocation according to load prediction accuracy.

  A resource allocation changing device according to an embodiment of the present invention includes a characteristic information storage unit that stores time determination information that can determine an allocation change time required for addition in association with a processing resource that can be added to a process being executed, Load prediction means for outputting the predicted value of the maximum load amount within a certain time of the process being executed and the predicted accuracy of the predicted value, and the allocation change time that is shorter than the upper limit value set shorter as the predicted accuracy is lower And a change allocation determination unit that outputs the selection information to a resource allocation change unit that changes the allocation of the selected processing resource.

  The resource allocation changing method according to one embodiment of the present invention stores, in the characteristic information storage means, time determination information that can determine an allocation change time required for addition in association with a processing resource that can be added to a process being executed, The process that outputs the predicted value of the maximum load amount within a certain time of the process being executed and the prediction accuracy of the prediction value, and has the allocation change time that is shorter than the upper limit value that is shorter as the prediction accuracy is lower A resource is selected with reference to the characteristic information storage means, and the allocation of the selected processing resource is changed.

  A program according to an embodiment of the present invention relates to a storage process for storing, in the characteristic information storage means, time determination information capable of determining an allocation change time required for addition in association with a processing resource that can be added to a process being executed, Load prediction processing for outputting the predicted value of the maximum load amount within a certain time of the process being executed and the prediction accuracy of the prediction value, and the allocation change time that is shorter than the upper limit value set shorter as the prediction accuracy is lower The computer executes the allocation change determination process for selecting the processing resource having a reference with reference to the characteristic information storage unit and outputting the selection information to the resource allocation changing unit for changing the allocation of the selected processing resource. Let

  The resource allocation changing device according to the present invention can perform appropriate resource allocation according to the prediction accuracy of load prediction.

FIG. 1 is a configuration diagram of a resource allocation changing apparatus 10 according to the first embodiment. FIG. 2 is a configuration diagram of the computer device 40. FIG. 3 is a configuration diagram of the characteristic information 20 stored in the characteristic information storage unit 15. FIG. 4 is an operation flowchart of the resource allocation changing device 10. FIG. 5 is an operation flowchart of the resource allocation changing device 10 according to the second exemplary embodiment. FIG. 6 is a configuration diagram of the resource allocation changing apparatus 10 according to the third embodiment.

<First Embodiment>
[Constitution]
FIG. 1 is a configuration diagram of a resource allocation changing device 10 according to the present embodiment. The resource allocation change device 10 includes a load measurement unit 11, a load prediction unit 12, an allocation change determination unit 13, a resource allocation change unit 14, a characteristic information storage unit 15, a processing resource 16, and an active resource 17.

  The load measuring unit 11 measures the load of processing that the resource allocation changing device 10 is executing using the active resource 17. The load prediction unit 12 predicts the future load based on the measured load information, and further obtains the accuracy of the predicted value. When it is predicted that the future load will be high, the allocation change determination unit 13 determines that the processing resource 16 should be additionally allocated to the process being executed in order to maintain the processing performance.

  Here, the active resource 17 is, for example, a core that is being used for a process being executed, a processing time that is being consumed, and the process resource 16 is a resource that can be added when the load of the process being executed increases, In other words, the waiting core, the available processing time, and the waiting accelerator. The accelerator is a dedicated processor that performs part of core processing, for example, image processing.

  The resource allocation changing device 10 includes a plurality of processing resources 16 having different characteristics. Such information is described in the characteristic information 20 described later and stored in the characteristic information storage unit 15.

  The allocation change determination unit 13 selects an appropriate processing resource 16 to cope with future load fluctuations based on the load prediction information and prediction accuracy output by the load prediction unit 12 and adds them to the process being executed. Decide what to assign. At this time, the allocation change determination unit 13 refers to the characteristic information 20.

  The resource allocation changing device 10 includes a resource allocation changing unit 14 that allocates the resource to a process being executed in association with each of the plurality of processing resources 16. When the processing resource 16 is selected, the allocation change determination unit 13 instructs the resource allocation changing unit 14 associated with the selected processing resource 16 to allocate the processing resource 16 to the process being executed. That is, the allocation change determination unit 13 outputs selection information to the resource allocation change unit 14. The resource allocation changing unit 14 receives an instruction based on the selection information from the allocation change determining unit 13 and allocates the corresponding processing resource 16 to the process being executed.

  The allocation process performed by the resource allocation changing unit 14 is different for each associated processing resource 16. The resource allocation changing unit 14 associated with the waiting core performs, for example, the initial setting of the core and the setting change of the process dispatcher so that the core can execute the process of the processing being executed. The resource allocation changing unit 14 associated with the vacant processing time, for example, increases the usage rate upper limit value that can be consumed by the process being executed.

  Here, the load measuring unit 11, the load predicting unit 12, the allocation change determining unit 13, and the resource allocation changing unit 14 are configured by logic circuits. The characteristic information storage unit 15 is composed of a semiconductor storage device or a magnetic disk device.

  Further, the resource allocation changing device 10 can also be realized by a computer device 40 provided with a program 43.

  FIG. 2 is a configuration diagram of the computer device 40. The computer device 40 includes a processor 41, a main storage unit 42, and an auxiliary storage device 44 that are connected to each other via a bus 45. Here, for example, the main storage unit 42 is a semiconductor storage device, and the auxiliary storage device 44 is an HDD (Hardware Disk Drive) or an SDD (Solid State Drive). The main storage unit 42 stores a program 43.

  The program 43 is executed by the processor 41 to cause the processor 41 to function as the load measurement unit 11, the load prediction unit 12, the allocation change determination unit 13, or the resource allocation change unit 14. The auxiliary storage device 44 or the main storage unit 42 functions as the characteristic information storage unit 15.

  When the resource allocation changing device 10 is realized by the computer device 40, the processor 41 may be used as the active resource 17 and the processing resource 16 of the idle processing time of the active core.

  FIG. 3 is a configuration diagram of the characteristic information 20 stored in the characteristic information storage unit 15. The characteristic information 20 records an ID (IDentification), a maximum allocatable resource amount, a performance improvement rate, an allocation change time, and an accuracy lower limit value in association with each processing resource 16.

  The maximum allocatable resource amount designates the amount when the allocated amount of the processing resource 16 is variable. When the allocated amount of the processing resource 16 is not variable, for example, when there is only a used / unused state like an accelerator, the maximum allocatable resource amount is omitted.

  The performance improvement rate represents, for example, a multiple of the ratio between the maximum amount that can be allocated to the processing resource 16 and the amount that is currently allocated. This value indicates the upper limit of the improvement rate obtained by changing the allocation of the processing resource 16. The resource allocation change unit 14 recalculates the performance improvement rate when the allocation amount of the processing resource 16 is changed, and updates the characteristic information 20. For example, the resource allocation changing unit 14 calculates the performance improvement rate by the following calculation formulas (1) and (2).

When the processing resource 16 is “free processing time”: maximum allocatable time ÷ current allocation time (1)
When the processing resource 16 is “standby core”: the maximum number of allocatable cores / the current number of allocated cores (2)
The formula for calculating the performance improvement rate may be described in the characteristic information 20 or may be preliminarily incorporated as processing logic in the resource allocation changing unit 14. However, in the case of the processing resource 16 in which the maximum allocatable resource amount is omitted, the resource allocation changing unit 14 indicates that the processing resource 16 has already been allocated when the processing resource 16 is allocated. The performance improvement rate of 20 is changed to 1 (no improvement).

  The allocation change time is a time required for changing the allocation of the processing resource 16. This time is, for example, a time required for the initial setting of the accelerator. The unit is, for example, milliseconds. The allocation change time can be said to be a delay time until the resource allocation change determining unit 13 actually allocates the resource after the resource allocation change determining unit 13 determines the additional allocation of the processing resource 16.

  The lower limit of accuracy specifies the lower limit of the prediction accuracy of the predicted load value. Here, the prediction accuracy is a value that increases when the accuracy is high. The allocation change determination unit 13 sets the processing resource 16 as a selection candidate for the allocation change when the prediction accuracy of the predicted load value output by the load prediction unit 12 is equal to or higher than the lower limit of accuracy.

  As shown in FIG. 3, the accuracy lower limit value is specified to be larger as the allocation change time is longer. For example, the accuracy lower limit value of the processing resource 16 (standby accelerator) whose allocation change time is 50 milliseconds is 90, whereas the accuracy lower limit value of the processing resource 16 (free processing time) whose allocation change time is 1 millisecond. Is 10. In this case, for example, if the prediction accuracy is 95, the allocation change determination unit 13 makes the processing resource 16 whose allocation change time is 500 milliseconds or less as a candidate. Only the processing resource 16 whose time is 1 millisecond or less is selected as a candidate.

  That is, when the prediction accuracy is low, the allocation change determination unit 13 selects the processing resource 16 having a short allocation change time. The reason for doing this is that there is a possibility that the prediction will be lost when the prediction accuracy is low, so it is possible to respond quickly by changing the allocation of the processing resources 16 that require a shorter allocation change time and see the next load change. This is because it is better to change the allocation of the processing resources 16 again.

[Operation]
FIG. 4 is an operation flowchart of the resource allocation changing device 10.

  First, the load measurement unit 11 measures the load of the process being executed, and outputs the measured value to the load prediction unit 12 (S1). The load to be measured is, for example, a CPU (Central Processing Unit) usage rate of the operating core. The load to be measured may be the number of input transactions.

  The load prediction unit 12 obtains the predicted value of the maximum load amount, the prediction accuracy, and the limit arrival time from the load measurement value input from the load measurement unit 11, and outputs the result to the allocation change determination unit 13 (S2). For example, the load prediction unit 12 performs this prediction by a method using a moving average or an existing method using a Kalman filter.

  Here, the limit arrival time is a time until a limit load amount (hereinafter referred to as a limit load amount) determined according to the amount of the operating resource 17 is reached. For example, if the load amount is a CPU usage rate, the limit load amount is 90%. For example, if the load amount is the number of input transactions, it is a number obtained by multiplying the predetermined number of transactions per core number by the number of assigned cores.

  The predicted maximum load amount is the maximum load amount that is predicted to be reached within time T in the future. This fixed time T is predetermined, for example, a value of 10 seconds. The load prediction unit 12 uses the current predicted load amount Lest and the current predicted load increase rate Aest obtained from the approximate expression of the load fluctuation obtained from the input load information, for example, as the predicted maximum load Lmax. , Lmax = Aest × T + Lest may be calculated by a linear expression.

  The prediction accuracy is a numerical value indicating the accuracy of the predicted maximum load amount, and is, for example, an index calculated based on the error range of the predicted maximum load amount. The prediction accuracy decreases as the error range increases. For example, the prediction range is an index having a value range of 1 to 100. The load prediction unit 12 may calculate the error range using the average absolute deviation from the current load amount Lcur obtained from the input load information and the above-described current predicted load amount Lest.

  The allocation change determination unit 13 that has received the predicted value of the maximum load amount, the prediction accuracy, and the limit arrival time calculated in S2 from the load prediction unit 12 refers to the characteristic information 20 and allocates shorter than the limit arrival time. A processing resource 16 having a change time is selected (S3). The reason for making such a selection is that it is desired to complete the additional allocation of the processing resources 16 before the load reaches the limit load amount. When there are a plurality of processing resources 16 that meet the conditions, the allocation change determination unit 13 selects the plurality.

  Next, the allocation change determination unit 13 selects the processing resource 16 having the input prediction accuracy higher than the accuracy lower limit value from the processing resources 16 selected in S3 (S4). The reason is as described above.

  When the number of processing resources 16 selected in S4 is one (Y in S5), the allocation change determination unit 13 adds the resource to the process being executed to the resource allocation change unit 14 associated with the processing resource 16 Direct allocation. The instructed resource allocation changing unit 14 performs additional allocation of the associated processing resource 16 (S7).

  When there are a plurality of processing resources 16 selected in S4 (N in S5), the allocation change determination unit 13 selects the processing resource 16 having the maximum evaluation value based on the improvement effect and the allocation change time, and the processing resource 16 The resource allocation changing unit 14 associated with is instructed to allocate additional resources to the process being executed (S6). This evaluation value has a higher upper limit of the performance improvement rate described in the characteristic information 20 than the ratio between the predicted value of the maximum load amount and the current load, and the higher the processing resource 16 whose allocation change time is shorter, the higher the evaluation value. Designed to be The allocation change determination unit 13 calculates the evaluation value using, for example, the following mathematical formula (3).

min (performance improvement rate, predicted maximum load / current load) ÷ (log (allocation change time + A) + B)
... (3)
Here, min is a function that takes a small value of two values. The constant terms A and B are parameters for adjusting the processing resources 16 to be selected with an appropriate balance. Here, A is 1 or more and B is 0 or more. Further, the logarithm of the allocation change time is taken so that even a configuration change time such as an accelerator that is very long can be selected.

[effect]
The resource allocation changing device 10 according to the present embodiment can allocate an appropriate processing resource 16 according to the prediction accuracy of load prediction. This is because the allocation change determination unit 13 selects a processing resource 16 having a short allocation change time when the prediction accuracy is low. As a result, when the prediction accuracy is low, the resource allocation change device 10 responds quickly by changing the allocation of the processing resource 16 that requires a shorter allocation change time, and after seeing the next load change, the processing resource 16 again. It is possible to change the allocation.

  Furthermore, the resource allocation changing device 10 according to the present embodiment can cope with load fluctuations at an appropriate speed. This is because the allocation change determination unit 13 selects a processing resource 16 whose allocation change time is equal to or shorter than the limit arrival time.

<Modification of the first embodiment>
The process being executed may be executed by another device connected to the resource allocation changing device 10, for example, a server device. In this case, the processing resource 16 and the operating resource 17 are provided in the other device. The resource allocation changing unit 14 may be included in the resource allocation changing device 10 or another device.

  The prediction accuracy may be a value that increases when the accuracy is low. That is, the prediction accuracy may be expressed by the magnitude of error. In this case, the characteristic information 20 includes the accuracy upper limit value, and the magnitude relationship regarding the accuracy is reversed from the above.

  The characteristic information 20 may not include the accuracy lower limit value or the accuracy upper limit value. In this case, the allocation change determination unit 13 may include a table including an upper limit value of the allocation change time determined for each of the plurality of prediction accuracy intervals. In this table, the lower the prediction accuracy, the smaller the upper limit value of the allocation change time. In this table, for example, the prediction accuracy interval is divided into less than 10, less than 10 and less than 50, more than 50 and less than 90, and more than 90, and the upper limit value is 0.5 (milliseconds), 2, 20, 600 in each interval. Determine. The allocation change determination unit 13 may refer to the allocation change time in the table and the characteristic information 20 and select the processing resource 16 having an allocation change time equal to or less than the input upper limit value corresponding to the prediction accuracy as a selection candidate.

  The load prediction unit 12 stores a load prediction value corresponding to time, date, or day of the week, and a prediction error obtained by analyzing past statistical information, and may output the value as a prediction value. In this case, the resource allocation change device 10 may not include the load measurement unit 11.

  The characteristic information 20 may include the current allocated resource amount and the performance improvement rate calculation formula without including the performance improvement rate. In this case, the allocation changing unit 14 updates the current allocated resource amount when the allocation of the processing resource 16 is changed. When necessary, the allocation change determination unit 13 refers to the maximum allocatable resource amount such as the maximum allocatable time of the characteristic information 20, the current allocated resource amount, and the performance improvement rate calculation formula to determine the performance improvement rate. It may be calculated.

<Second Embodiment>
In the resource allocation changing device 10 of the present embodiment, the operation of the allocation change determining unit 13 is different from that of the first embodiment.

  FIG. 5 is an operation flowchart of the resource allocation changing device 10 according to the present embodiment. In FIG. 5, the operations from S1 to S7 are the same as those in the first embodiment.

  In the present embodiment, the allocation change determination unit 13 that has received the predicted value of the maximum load amount, the prediction accuracy, and the limit arrival time calculated in S2 from the load prediction unit 12 has a predetermined maximum predicted load amount. It is determined whether the value is exceeded (S11).

  If it has exceeded (Y in S11), the allocation change determination unit 13 selects the processing resource 16 having the highest performance improvement rate, and moves the resource allocation changing unit 14 associated with the processing resource 16 to the process being executed. Is assigned (S12). The instructed resource allocation changing unit 14 performs additional allocation of the associated processing resource 16 (S7).

  When the predicted maximum load amount does not exceed a predetermined constant value (N in S11), the assignment change determination unit 13 executes the processes after S3 as in the first embodiment.

  Obviously, if the future load is expected to be much higher than the current situation, it may be better to choose one with a high performance improvement rate even if the limit arrival time is exceeded. The resource allocation changing device 10 according to the present embodiment can appropriately cope with such a situation.

<Third Embodiment>
FIG. 6 is a configuration diagram of the resource allocation changing device 10 according to the present embodiment. The resource allocation changing device 10 according to the present embodiment includes a characteristic information storage unit 15, a load prediction unit 12, and an allocation change determination unit 13.

  The characteristic information storage unit 15 stores time determination information that can determine the allocation change time required for addition in association with the processing resource 16 that can be added to the process being executed. The load prediction unit 12 outputs the predicted value of the maximum load amount within a certain time of the process being executed and the prediction accuracy of the predicted value. The allocation change determination unit 13 selects a processing resource 16 having an allocation change time that is shorter than the upper limit value shorter as the prediction accuracy is lower with reference to the characteristic information storage unit 15, and selects the selected processing resource. Is output to the resource allocation changing unit 14 that changes the allocation.

  The resource allocation changing device 10 according to the present embodiment can allocate an appropriate processing resource 16 according to the prediction accuracy of load prediction. This is because the allocation change determination unit 13 selects a processing resource 16 having a short allocation change time when the prediction accuracy is low.

[Industrial applicability]
The present invention can be applied to a Web server computer or an AP (Application Program) server computer that receives and processes a user request from a network. The present invention can also be applied to an IoT (Internet of Things) / M2M (Machin to Machin) server computer that receives and processes data from a sensor.

  While the present invention has been described with reference to the embodiments, the present invention is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

DESCRIPTION OF SYMBOLS 10 Resource allocation change apparatus 11 Load measurement part 12 Load prediction part 13 Assignment change determination part 14 Resource allocation change part 15 Characteristic information storage part 16 Processing resource 17 Active resource 20 Characteristic information 40 Computer apparatus 41 Processor 42 Main memory part 43 Program 44 Auxiliary storage 45 bus

Claims (9)

Characteristic information storage means for storing time discrimination information capable of discriminating allocation change time required for addition in association with processing resources that can be added to the processing being executed;
Load prediction means for outputting a predicted value of the maximum load amount within a predetermined time of the process being executed, and a prediction accuracy of the predicted value;
The processing resource having the allocation change time that is shorter than the upper limit defined as the prediction accuracy is lower is selected with reference to the characteristic information storage unit, and the allocation of the selected processing resource is changed. A resource allocation changing device comprising: an allocation change determining means for outputting to a resource allocation changing means for performing The characteristic information storage means stores an upper limit value of a performance improvement rate when a processing resource is added, and the allocation change time corresponding to the addable processing resource,
The load prediction means predicts a limit arrival time that is a time to reach a limit load determined according to the amount of processing resources allocated to the process being executed within the predetermined time,
The allocation change determination means is a processing resource in which the allocation change time is shorter than the limit arrival time, and the upper limit of the performance improvement rate is equal to or greater than a ratio between the predicted value of the maximum load amount and the current load. The resource allocation changing device according to claim 1, wherein   3. The resource allocation change means for adding a processing resource when selection information for an associated processing resource is input from the assignment change determination means, the resource assignment changing means being associated with the addable processing resource. Resource allocation change device.   When the predicted value of the maximum load amount exceeds a predetermined value, the allocation change determining means has the highest upper limit value of the performance improvement rate regardless of the prediction accuracy of the predicted value and the limit arrival time. The resource allocation changing device according to any one of claims 2 to 3, wherein an addable processing resource is selected as a processing resource to be added to the execution of the processing.   The resource allocation change according to any one of claims 2 to 4, wherein the time determination information is any one of the allocation change time and an upper and lower accuracy value determined according to the length of the allocation change time. apparatus. In association with the processing resource that can be added to the process being executed, the time information that can determine the allocation change time required for addition is stored in the characteristic information storage means,
Outputting the predicted value of the maximum load amount within a certain time of the process being executed and the predicted accuracy of the predicted value;
A resource that changes the allocation of the selected processing resource by selecting the processing resource having the allocation change time that is shorter than the upper limit value determined as the prediction accuracy is lower with reference to the characteristic information storage unit Assignment change method. The upper limit value of the performance improvement rate when a processing resource is added to the processing resource that can be added, and the allocation change time are stored in the characteristic information storage,
Predicting a limit arrival time that is a time to reach a limit load determined according to the amount of processing resources allocated to the process being executed within the predetermined time;
The processing resource in which the upper limit of the performance improvement rate is equal to or greater than the ratio of the predicted value of the maximum load amount and the current load among the processing resources whose allocation change time is shorter than the limit arrival time is selected. Resource allocation change method.   When the predicted value of the maximum load amount exceeds a predetermined value, the addable processing resource having the highest upper limit value of the performance improvement rate regardless of the prediction accuracy of the predicted value and the limit arrival time, The resource allocation changing method according to any one of claims 6 to 7, which is selected as a processing resource to be added to the execution of the processing. A storage process for storing in the characteristic information storage means time determination information that can determine the allocation change time required for addition in association with a processing resource that can be added to the process being executed;
A load prediction process for outputting a predicted value of a maximum load amount within a predetermined time of the process being executed and a prediction accuracy of the predicted value;
The processing resource having the allocation change time that is shorter than the upper limit defined as the prediction accuracy is lower is selected with reference to the characteristic information storage unit, and the allocation of the selected processing resource is changed. A program that causes a computer to execute allocation change determination processing that is output to resource allocation changing means.

Images