The present disclosure relates to a resource management device and a resource management method.
There is a node including one or more containers. The node is a computer on which one operating system (OS) operates. The container causes an application to be executed using computer resources (hereinafter referred to as “resources”) of the node. The node has finite resources. Thus, it is useful to determine the one or more containers included in the node in such a manner that a total of resources used by the containers is not larger than a total of resources of the node.
There is a method of determining whether a container can be mounted (see, for example, Patent Literature 1). A method disclosed in Patent Literature 1 manages, as to one or more containers, respective usage states of resources, and determines a container included in a node on the basis of the respective usage states of resources so as not to exceed a total of resources of the node. The usage state of resources indicates a temporal change in a use amount of the resources used by the container in the past. The use amount of the resources may vary over time.
Patent Literature 1: JP 2020-52730 A
A variation in the use amount of resources may be different for each container. Under a situation where the variation in the use amount of resources is different for each container, a time at which the use amount of resources reaches a peak (hereinafter referred to as a “peak time”) may be different for each container, In a case where the peak time is different for each container, if the container included in the node is determined on the basis of respective peak values of the use amount of resources, a large portion of resources of the node may be unused.
On the other hand, peak times of all the containers included in the node or some of the containers included in the node may overlap depending on respective variation states of the use amount of resources. When the containers included in the node are determined on the basis of the respective use amounts of resources at a freely-selected time in a case where the peak times overlap, a total of the use amounts of resources by all the containers included in the node may become larger than the total of the resources of the node in the future.
In the method disclosed in Patent Literature 1, nodes included in a node are determined on the basis of the respective usage states of resources. However, there is a problem that the nodes included in the node are not determined on the basis of respective variation states of the use amount of resources. Thus, a situation in which a large portion of resources of the node is unused or a situation in which the total of the use amounts of resources by all the containers included in the node becomes larger than the total of the resources of the node in the future may occur.
The present disclosure has been made to solve the problem as described above, and an object of the present disclosure is to obtain a resource management device and a resource management method capable of suppressing a total of use amounts of resources by all containers included in a node to be equal to or less than a total of resources of the node while suppressing a surplus of the resources of the node.
A resource management device according to the present disclosure includes a data acquiring unit to acquire monitoring data indicating a temporal change in a use amount of a resource used in the past by each of one or more containers included in a node among at least one resource of the node, and a prediction unit to predict a temporal change in a use amount of the resource to be used in the future by each of the containers from a corresponding one of pieces of the monitoring data acquired by the data acquiring unit, and calculate a temporal change in a total use amount of the resource to be used in the future by the one or more containers from a prediction result of the temporal change, Further, the resource management device includes a container deleting unit to delete one of the one or more containers when a timing at which the total use amount of the resource becomes larger than a total of the resource of the node is included in the temporal change in the total use amount of the resource calculated by the prediction unit.
According to the present disclosure, it is possible to suppress a total of use amounts of resources by all containers included in a node to be equal to or less than a total of resources of the node while suppressing a surplus of the resources of the node.
FIG, 7 is a configuration diagram illustrating a resource management device 12 according to a second embodiment.
Hereinafter, in order to describe the present disclosure in more detail, modes for carrying out the present disclosure will be described with reference to the accompanying drawings.
In
The node 2 is connected to the container distribution server 1 via a network.
The node 2 is a computer on which one operating system (OS) operates.
The node 2 includes containers 11-1 to 11-N distributed from the container distribution server 1 and the resource management device 12.
Further, the node 2 includes an OS 13 and a container runtime 14.
The node 2 illustrated in
The container 11-n (n=1, . . . , N) is an execution environment of an application. A resource on which the application operates is a resource virtually separated from resources managed by the OS 13.
Virtual separation of resources is performed by the container runtime 14 using a function held by the OS 13.
An example of the resources is a central processing unit (CPU) included in the node 2, a memory included in the node 2, a storage included in the node 2, a network connected to the node 2, or an external device connected to the node 2. An example of the external device is a printer, a communication device, or a sensor,
The resource management device 12 includes a data acquiring unit 31, a prediction unit 32, and a container deleting unit 33.
The OS 13 is located between an application program and hardware, and manages the resources of the node 2. Further, the OS 13 provides a standard interface for the application program.
The container runtime 14 virtually separates the resources of the node 2 by using a function held by the OS 13.
Further, the container runtime 14 provides the virtually separated resources to the container 11-n (n=1, . . . , N).
Furthermore, the container runtime 14 outputs, to the data acquiring unit 31 of the resource management device 12, monitoring data MDn indicating a temporal change in a use amount of a resource used in the past by the container 11-n.
The data acquiring unit 31 is implemented by, for example, a data acquiring circuit 41 illustrated in
The data acquiring unit 31 acquires, from the container runtime 14, the monitoring data MDn indicating a temporal change in a use amount of a resource used in the past by the container 11-n (n=1, . . . , N) among the resources of the node 2.
The data acquiring unit 31 outputs the monitoring data MDn to the prediction unit 32.
The prediction unit 32 is implemented by, for example, a prediction circuit 42 illustrated in
The prediction unit 32 acquires the monitoring data MDn from the data acquiring unit 31.
The prediction unit 32 predicts, from the monitoring data MDn, temporal change in a use amount of a resource to be used in the future by the container 11-n.
The prediction unit 32 calculates, from the prediction result of the temporal change, a temporal change in a total use amount of resources to be used in the future by the containers 11-1 to 11-N.
The container deleting unit 33 is implemented by, for example, a container deleting circuit 43 illustrated in
The container deleting unit 33 determines whether or not a timing at which the total use amount of resources becomes larger than a total of the resources of the node 2 is included in the temporal change in the total use amount of resources calculated by the prediction unit 32.
When it is determined that the timing is included, the container deleting unit 33 deletes one of the containers 11-1 to 11-N.
When it is determined that the timing is not included, the container deleting unit 33 does not delete any of the containers 11-1 to 11-N.
In
Each of the data acquiring circuit 41, the prediction circuit 42, and the container deleting circuit 43 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel-programmed processor, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof,
The components of the resource management device 12 are not limited to those implemented by dedicated hardware, and the resource management device 12 may be implemented by software, firmware, or a combination of software and firmware.
The software or firmware is stored in a memory of a computer as a program. The computer means hardware that executes a program, and corresponds to, for example, a CPU, a central processing device, a processing device, an arithmetic device, a microprocessor, a microcomputer, a processor, or a digital signal processor (DSP).
In a case where the resource management device 12 is implemented by software, firmware, or the like, a program for causing a computer to execute each processing procedure performed in the data acquiring unit 31, the prediction unit 32, and the container deleting unit 33 is stored in a memory 51. Then, a processor 52 of the computer executes the program stored in the memory 51.
Further,
Next, an operation of the resource management device 12 illustrated in
The node 2 includes the containers 11-1 to 11-N distributed from the container distribution server 1.
Here, in order to simplify the description, processing to determine, at a time when any container of the containers 11-1 to 11-N is distributed from the container distribution server 1, whether or not to mount the distributed container on the node 2 will be described later.
The data acquiring unit 31 acquires, from the container runtime 14 of the node 2, the monitoring data MDn indicating the temporal change in the use amount of the resource used in the past by the container 11-n (n=1, . . . , N) among the resources of the node 2 (step ST1 in
A use amount of the resource of the node 2 is, for example, a use rate of a CPU included in the node 2, a use amount of a memory included in the node 2, a use amount of a storage included in the node 2, a bandwidth use rate of a network connected to the node 2, or a use rate of a device connected to the node 2.
The data acquiring unit 31 outputs the monitoring data MDn to the prediction unit 32.
The prediction unit 32 acquires the monitoring data MDn (n=1, . . . , N) from the data acquiring unit 31.
The prediction unit 32 predicts, from the monitoring data MDn, a temporal change in the use amount of the resource to be used in the future by the container 11-n (step ST2 in
Hereinafter, prediction processing performed in the prediction unit 32 will be specifically described.
The prediction unit 32 acquires, from the monitoring data MDn, a use amount un(t) of a resource at a certain time t and a use amount un(t+τ) of the resource at a time (t+τ) later than the time t by a delay time τ.
The prediction unit 32 calculates an autocorrelation function Rn(τ) for the use amount of the resource, from the use amount un(t) of the resource and the use amount un(t+τ) of the resource. Since the calculation processing of the autocorrelation function Rn(τ) itself is a known technique, detailed description thereof will be omitted.
The prediction unit 32 calculates M autocorrelation functions Rn(τ) by changing the delay time τ. M is an integer equal to or more than 2. Hereinafter, the M autocorrelation functions Rn(τ) are expressed as Rn,m(τ). Here, m=1, . . . , M.
The prediction unit 32 specifies, as a period pn of the use amount of the resource used in the past by the container 11-n, the delay time τ of the autocorrelation function Rn,m(τ) having the highest positive correlation among M autocorrelation functions Rn,m(τ).
The prediction unit 32 predicts the temporal change in the use amount of the resource to be used in the future by the container 11-n on the basis of the temporal change in the use amount within the period pn.
Specifically, the prediction unit 32 predicts that the temporal change in the use amount of the resource used by the container 11-n from a time (t0−pn·j) that is j periods before the current time t0 to a time (t0−pn·(j−1)) that is (j−1) periods before the current time t0 is a temporal change in the use amount of the resource to be used by the container 11-n from a time (t0+pn(i−1)) that is (i−1) periods after the current time t0 to a time (t0+pn·i) that is i periods after the current time t0. Each of i and j is an integer equal to or more than 1.
In the example of
Here, the prediction unit 32 predicts the temporal change in the use amount of the resource for one period. However, this is merely an example, and the prediction unit 32 may predict the temporal change in the use amount of the resource for a plurality of periods on the basis of the temporal change in the use amount within the period pn.
In the resource management device 12 illustrated in
In the resource management device 12 illustrated in
However, in some cases, there is no periodicity in the use amount of the resource used in the past by the container 11-n. In a case where there is no periodicity in the use amount of the resource, each of the M autocorrelation functions Rn,1(τ) to Rn,M(τ) is approximately 0.
When there is no periodicity in the use amount of the resource, the prediction unit 32 predicts a temporal change in the use amount of the resource to be used in the future by the container 11-n on the basis of the peak value of the use amount of the resource used in the past by the container 11-n.
That is, the prediction unit 32 predicts that the use amount of the resource to be used in the future by the container 11-n has a continuous peak value of the use amount of the resource used in the past.
The prediction unit 32 calculates, from a prediction result of the temporal change, the temporal change in the total use amount of the resource to be used in the future by the containers 11-1 to 11-N (step ST3 in
Hereinafter, calculation processing by the prediction unit 32 will be specifically described.
The prediction unit 32 compares the periods p1 to pN of the use amount of the resource used in the past by the containers 11-1 to 11-N with each other, and specifies a longest period pmax among the periods p1 to pN. Here, for convenience of description, it is assumed that the container related to the longest period pmax is the container 11-1.
The prediction unit 32 specifies a use amount u1(t) of a resource to be used from the current time t0 to a time t0+p1, which is one period after the current time, among the use amounts of the resource to be used in the future by the container 11-1, t0≤t≤t0+p1=t0+pmax.
The prediction unit 32 specifies a use amount un(t) of a resource to be used from the current time t0 to the time t0+p1 among the use amounts of the resource to be used in the future by the container 11-n (n=2, . . . , N).
The prediction unit 32 calculates a total resource use amount utotal(t) which is a total of use amounts u1(t) to uN(t) of the resource as expressed in the following expression (1).
utotal(t)=u1(t)+ . . . +uN(t) (1)
In the example of
The prediction unit 32 outputs the total resource use amount utotal(t) to the container deleting unit 33 as a temporal change in the total use amount of the resource.
In the resource management device 12 illustrated in
The container deleting unit 33 acquires the total resource use amount utotal(t) from the prediction unit 32.
For example, the container deleting unit 33 determines whether or not there is a timing at which the total resource use amount utotal(t) becomes larger than a threshold Th between the current time t0 and the time t0+p1. The threshold Th indicates the total of the resources of the node 2. The threshold Th may be stored in the internal memory of the container deleting unit 33 or may be given from the outside of the resource management device 12.
When there is a timing at which the total resource use amount utotal(t) becomes larger than the threshold Th between the current time t0 and the time t0+p1 (in the case of step ST4; YES in
When there is no timing at which the total resource use amount utotal(t) becomes larger than the threshold Th between the current time t0 and the time t0+p1 (in the case of step ST4: NO in
When it is determined that the resource use amount larger than the total of the resources of the node 2 is included in the total resource use amount utotal(t), the container deleting unit 33 deletes one of the containers 11-1 to 11-N (step ST5 in
Hereinafter, container deletion processing by the container deleting unit 33 will be specifically described.
When it is determined that the resource use amount larger than the total of the resources of the node 2 is included in the total resource use amount utotal(t), the container deleting unit 33 specifies a use amount un(Tim) of a resource to be used by the container 11-n at a time Tim when the resource use amount becomes larger than the total of the resources of the node 2,
The container deleting unit 33 specifies a largest use amount umax(Tim) of a resource among the use amounts u1(Tim) to uN(Tim) of the resource to be used by the containers 11-1 to 11-N, respectively.
The container deleting unit 33 deletes a container related to the largest use amount umax(Tim) of the resource among the containers 11-1 to 11-N.
If the total resource use amount utotal(Tim) is larger than the threshold Th even when the container related to the largest use amount umax(Tim) of the resource is deleted, the container deleting unit 33 further deletes the container related to the second largest use amount of a resource.
The container deleting unit 33 deletes the containers in descending order of the use amount of the resource until the total resource use amount utotal(Tim) becomes equal to or less than the threshold Th.
When it is determined that the timing does not exist, the container deleting unit 33 does not perform the container deletion processing.
In the resource management device 12 illustrated in
In the resource management device 12 illustrated in
In the resource management device 12 illustrated in
Further, the container deleting unit 33 calculates a cost CSTn of the container 11-n by performing weighted addition or the like of the use amount un(Tim) of the resource at the time Tim, the time tdn of distribution to the node 2, and the operation priority wn of the container 11-n. The cost CSTn is larger as the use amount un(Tim) of the resource is larger, is larger as the time tdn of distribution to the node 2 is closest to the current time t0, and is larger as the operation priority wn is smaller.
Then, the container deleting unit 33 may determine the container having the largest cost CSTn among the containers 11-1 to 11-N as the container to be deleted.
In the first embodiment described above, the resource management device 12 is configured to include the data acquiring unit 31 that acquires monitoring data indicating the temporal change in the use amount of a resource used in the past by each of one or more containers 11-n (n=1, . . . , N) included in the node 2 among the resources of the node 2, and the prediction unit 32 that predicts the temporal change in the use amount of a resource to be used in the future by the container 11-n from the corresponding one pieces of the monitoring data acquired by the data acquiring unit 31 and calculates the temporal change in the total use amount of the resource to be used in the future by the containers 11-1 to 11-N from the prediction result of the temporal change. Further, the resource management device 12 includes the container deleting unit 33 that deletes one of the containers 11-1 to 11-N when a timing at which the total use amount of the resource becomes larger than the total of the resources of the node 2 is included in the temporal change in the total use amount of the resource calculated by the prediction unit 32. Therefore, the resource management device 12 can suppress the total of the resource use amounts by the containers 11-1 to 11-N included in the node 2 to be equal to or less than the total of the resources of the node 2 while suppressing the surplus of the resources of the node 2.
The node 2 illustrated in
When the new container 11-(N+1) is a container that the node 2 has never included in the past, the data acquiring unit 31 of the resource management device 12 acquires a scheduled use amount of the resource used by the new container 11-(N+1) from the container distribution server 1.
The container deleting unit 33 adds a temporal average value of the total resource use amount by the containers 11-1 to 11-N and the scheduled use amount of the resource used by the new container 11-(N+1).
Then, the container deleting unit 33 determines whether or not an addition result of the temporal average value and the scheduled use amount is equal to or less than the total of the resources of the node 2.
When the addition result is equal to or less than the total of the resources of the node 2, the container deleting unit 33 outputs a permission command indicating that it is permitted to include the new container 11-(N+1) to the container runtime 14. Thus, the container runtime 14 causes the node 2 to include the containers 11-1 to 11-N and also include the container 11-(N+1).
When the addition result is larger than the total of the resources of the node 2, the container deleting unit 33 outputs a non-permission command indicating that it is not permitted to include the new container 11-(N+1) to the container runtime 14. Thus, the container runtime 14 causes the node 2 to include only the containers 11-1 to 11-N.
When the new container 11-(N+1) is a container that the node 2 has included in the past, the prediction unit 32 may predict a temporal change in the use amount of the resource to be used in the future by the container 11-(N+1), similarly to the containers 11-1 to 11-N.
Then, the prediction unit 32 calculates a temporal change in the total use amount of the resource to be used in the future by the containers 11-1 to 11-(N+1).
When the temporal change in the total use amount of the resource calculated by the prediction unit 32 does not include a timing at which the total use amount of the resource becomes larger than the total of the resources of the node 2, the container deleting unit 33 outputs a permission command indicating that it is permitted to include the new container 11-(N+1) to the container runtime 14. Thus, the container runtime 14 causes the node 2 to include the containers 11-1 to 11-N and also include the container 11-(N+1).
When the temporal change in the total use amount of the resource calculated by the prediction unit 32 includes a timing at which the total use amount of the resource becomes larger than the total of the resources of the node 2, the container deleting unit 33 outputs a non-permission command indicating that it is not permitted to include the new container 11-(N+1) to the container runtime 14. Thus, the container runtime 14 causes the node 2 to include only the containers 11-1 to 11-N.
In the resource management device 12 illustrated in
Specifically, it is as follows.
The prediction unit 32 detects a container having no periodicity in the use amount of the resource among the containers 11-1 to 11-N. In the container having no periodicity in the use amount of the resource, each of the M autocorrelation functions Rn,1(τ) to Rn,M(τ) is substantially 0 even if the delay time τ is changed.
In a case where there are H containers having no periodicity in the use amount of the resource, the prediction unit 32 calculates a correlation coefficient C between the use amounts of the resource used in the past by the respective H containers so as not to include the use amount of the resource to be used in the future by a container other than one container among the H containers in the total use amount of the resource. H is an integer equal to or more than 2, and H≤N.
Among the containers 11-1 to 11-N, when containers having no periodicity in the use amount of the resource are, for example, the containers 11-1 and 11-2, a correlation coefficient C1-2 between the use amount of the resource used in the past by the container 11-1 and the use amount of the resource used in the past by the container 11-2 is calculated. Since the calculation processing for the correlation coefficient C1-2 itself is a known technique, detailed description thereof will be omitted.
When the correlation coefficient C1-2 is a negative correlation, the prediction unit 32 specifies a container having the highest peak value of the use amount of the resource used in the past among the containers 11-1 and 11-2,
For example, when the container having the highest peak value is the container 11-1, the prediction unit 32 specifies the use amount u1(t) of the resource as the temporal change in the use amount of the resource to be used in the future by the container 11-1. The use amount u1(t) of the resource has a continuous peak value of the use amount of the resource used in the past by the container 11-1.
The prediction unit 32 does not specify the use amount u2(t) of the resource to be used in the future by the container 11-2 which is a container other than the container having the highest peak value.
When calculating the total resource use amount utotal(t), the prediction unit 32 includes the use amount u1(t) of the resource to be used in the future by the container 11-1 in the total resource use amount utotal(t), but does not include the use amount u2(t) of the resource to be used in the future by the container 11-2.
When the correlation coefficient C1-2 between the use amount of the resource used in the past by the container 11-1 and the use amount of the resource used in the past by the container 11-2 is a negative correlation, it is unlikely that the container 11-1 and the container 11-2 use the resource at the same timing. Thus, the prediction unit 32 does not include the use amount u2(t) of the resource to be used in the future by the container 11-2 in the total resource use amount utotal(t).
Here, an example is illustrated in which, among the containers 11-1 to 11-N, containers having no periodicity in the use amount of the resource are the containers 11-1 and 11-2. However, this is merely an example, and there may be three or more containers having no periodicity in the use amount of the resource. For example, it is assumed that containers having no periodicity in the use amount of the resource are containers 11-1 to 11-3.
In this case, the prediction unit 32 calculates the correlation coefficient C1-2 between the use amount of the resource used in the past by the container 11-1 and the use amount of the resource used in the past by the container 11-2, and calculates a correlation coefficient C2-3 between the use amount of the resource used in the past by the container 11-2 and the use amount of the resource used in the past by the container 11-3.
Further, the prediction unit 32 calculates a correlation coefficient C3-1 between the use amount of the resource used in the past by the container 11-3 and the use amount of the resource used in the past by the container 11-1.
When each of the correlation coefficients C1-2, C2-3, and C3-1 has a negative correlation, the prediction unit 32 specifies a container having the highest peak value of the use amount of the resource used in the past among the containers 11-1 to 11-3.
For example, when the container having the highest peak value is the container 11-1, the prediction unit 32 specifies the use amount u1(t) of the resource as the temporal change in the use amount of the resource to be used in the future by the container 11-1, The use amount u1(t) of the resource has a continuous peak value of the use amount of the resource used in the past by the container 11-1.
The prediction unit 32 does not specify the use amounts u2(t) and u3(t) of the resource to be used in the future by the respective containers 11-2 and 11-3 which are containers other than the container having the highest peak value.
When calculating the total resource use amount utotal(t), the prediction unit 32 includes the use amount u1(t) of the resource to be used in the future by the container 11-1 in the total resource use amount utotal(t), but does not include the use amounts u2(t) and u3(t) of the resource to be used in the future by the respective containers 11-2 and 11-3.
In a second embodiment, a resource management device 12 in which a prediction unit 35 acquires a prediction result of a temporal change in the use amount of the resource to be used in the future by the container 11-n (n=1, . . . , N) from a learning model 34 will be described.
The resource management device 12 illustrated in
The configuration of the node 2 including the resource management device 12 illustrated in
The learning model 34 is implemented by, for example, a learning circuit 44 illustrated in
At the time of learning, monitoring data MDn indicating a temporal change in the use amount of the resource used in the past by the container 11-n (n=1, . . . , N) and training data GDn indicating a temporal change in the use amount of the resource to be used in the future by the container 11-n are given to the learning model 34. Then, the learning model 34 learns the temporal change in the use amount of the resource to be used in the future by the container 11-n.
At the time of inference, when the monitoring data MDn indicating the temporal change in the use amount of the resource used in the past by the container 11-n is given, the learning model 34 outputs a prediction result indicating the temporal change in the use amount of the resource to be used in the future by the container 11-n.
The prediction unit 35 is implemented by, for example, a prediction circuit 45 illustrated in
The prediction unit 35 acquires the monitoring data MDn (n=1, . . . , N) from the data acquiring unit 31.
The prediction unit 35 gives the monitoring data MDn to the learning model 34 and acquires, from the learning model 34, a prediction result indicating the temporal change in the use amount of the resource to be used in the future by the container 11-n.
Similarly to the prediction unit 32 illustrated in
In the resource management device 12 illustrated in
In
Each of the data acquiring circuit 41, the learning circuit 44, the prediction circuit 45, and the container deleting circuit 43 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel-programmed processor, ASIC, FPGA, or a combination thereof.
The components of the resource management device 12 are not limited to those implemented by dedicated hardware, and the resource management device 12 may be implemented by software, firmware, or a combination of software and firmware.
In a case where the resource management device 12 is implemented by software, firmware, or the like, a program for causing a computer to execute each processing procedure performed in the data acquiring unit 31, the learning model 34, the prediction unit 35, and the container deleting unit 33 is stored in the memory 51 illustrated in
Furthermore,
Next, an operation of the resource management device 12 illustrated in
At the time of learning, the monitoring data MDn indicating the temporal change in the use amount of the resource used in the past by the container 11-n (n=1, . . . , N) and the training data GDn indicating the temporal change in the use amount of the resource to be used in the future by the container 11-n are given to the learning model 34.
If the learning model 34 is, for example, a learning model including a neural network, the learning model 34 updates a weighting factor of the neural network in such a manner that output data ODn indicating a prediction result when the monitoring data MDn is given becomes the temporal change in the use amount indicated by the training data GDn.
The prediction unit 35 acquires, from the data acquiring unit 31, the monitoring data MDn indicating the temporal change in the use amount of the resource used in the past by the container 11-n (n=1, . . . , N).
The prediction unit 35 provides the monitoring data MDn to the learning model 34.
When the monitoring data MDn is given, the learning model 34 outputs the output data ODn indicating a prediction result corresponding to the monitoring data MDn.
The prediction unit 35 acquires the output data ODn indicating the prediction result from the learning model 34.
Similarly to the prediction unit 32 illustrated in
The prediction unit 35 outputs the total resource use amount utotal(t) to the container deleting unit 33 as a temporal change in the total use amount of the resource.
Similarly to the resource management device 12 illustrated in
Further, the resource management device 12 illustrated in
In the resource management device 12 illustrated in
However, this is merely an example, and monitoring data MDn′ may be given to the learning model 34 in addition to the monitoring data MDn and the training data GDn at the time of learning. The monitoring data MDn′ indicates a temporal change in the use amount of the resource used in the past by a container among the containers 11-1 to 11-N different from the container 11-n.
The learning model 34 updates the weighting factor of the neural network in such a manner that the output data ODn indicating the prediction result when the monitoring data MDn and the monitoring data MDn′ are given becomes the temporal change in the use amount indicated by the training data GDn.
The prediction unit 35 gives the monitoring data MDn and the monitoring data MDn′ to the learning model 34, and acquires the output data ODn indicating the prediction result from the learning model 34.
The temporal change in the monitoring data MDn and the temporal change in the monitoring data MDn′ may affect each other. In such a case, at the time of inference, the prediction unit 35 may obtain a highly accurate prediction result from the learning model 34 by giving the monitoring data MDn and the monitoring data MDn′ to the learning model 34 rather than giving only the monitoring data MDn to the learning model 34.
In a third embodiment, a resource management device 12 in which a prediction unit 36 extracts, from each piece of monitoring data MD1 to MDN acquired by a data acquiring unit 31, a part of the data when a state of the node 2 is the same state will be described.
The resource management device 12 illustrated in
The configuration of the node 2 including the resource management device 12 illustrated in
The prediction unit 36 is implemented by, for example, a prediction circuit 46 illustrated in
The prediction unit 36 acquires the monitoring data MDn (n=1, . . . , N) from the data acquiring unit 31.
The prediction unit 36 extracts, from each piece of the monitoring data MD1 to MDN, a part of the data when the state of the node 2 is the same state.
An example of a part of data when the state of the node 2 is the same state is a part of data when the state of movement of the node 2 is the same state, a part of data when the operating state of the node 2 is the same state, or a part of data when the state of another system connected to the node 2 is the same state.
An example of the state of movement of the node 2 is a state in which the node 2 is moving at a high speed, a state in which the node 2 is moving at a medium speed, a state in which the node 2 is moving at a low speed, or a state in which the node 2 is stopped.
An example of the operating state of the node 2 is an activated state in which the node 2 is activated, a normal state in which the activation of the node 2 is completed, a safe mode state in which the node 2 is operating in the safe mode, a sleep state in which the operation of the node 2 is paused, or an end state in which the operation of the node 2 is ended.
An example of the state of the other system connected to the node 2 is a state of movement of the other system or an operating state of the other system, An example of the other system is a communication system connected to the node 2 or an in-vehicle system connected to the node 2.
The prediction unit 36 predicts the temporal change in the use amount of the resource to be used in the future by the container 11-n from each part of the data.
Similarly to the prediction unit 32 illustrated in
In
Each of the data acquiring circuit 41, the prediction circuit 46, and the container deleting circuit 43 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel-programmed processor, ASIC, FPGA, or a combination thereof.
The components of the resource management device 12 are not limited to those implemented by dedicated hardware, and the resource management device 12 may be implemented by software, firmware, or a combination of software and firmware.
In a case where the resource management device 12 is implemented by software, firmware, or the like, a program for causing a computer to execute each processing procedure performed in the data acquiring unit 31, the prediction unit 36, and the container deleting unit 33 is stored in the memory 51 illustrated in
Furthermore,
Next, an operation of the resource management device 12 illustrated in
The prediction unit 36 acquires the monitoring data MDn (n=1, . . . , N) from the data acquiring unit 31.
The prediction unit 36 extracts, from each piece of the monitoring data MD1 to MDN, a part of the data when the state of the node 2 is the same state.
That is, the prediction unit 36 extracts, from each piece of the monitoring data MD1 to MDN, as a part of the data when the state of the node 2 is the same state, for example, a part of the data when the state of movement of the node 2 is the same state, a part of the data when the operating state of the node 2 is the same state, or a part of the data when the state of another system connected to the node 2 is the same state.
In the example of
The prediction unit 36 predicts a temporal change in the use amount of the resource to be used in the future by the container 11-n from a part of data included in the monitoring data MDn. In the example of
Since the prediction processing by the prediction unit 36 itself is similar to the prediction processing by the prediction unit 32 illustrated in
Similarly to the prediction unit 32 illustrated in
In the resource management device 12 illustrated in
In the resource management device 12 illustrated in
An example of the part of the data when the environment of the node 2 is the same environment is a part of data when an environment related to weather around the node 2 is the same environment, or a part of data when an environment related to a use time of the resource is the same environment.
An example of the environment related to the weather around the node 2 is an environment in which the weather is sunny, an environment in which the weather is rainy, an environment in which the weather is cloudy, or an environment in which the weather is snowy.
An example of the environment related to the use time of the resource is an environment in which the use time of the resource is included in the morning time period, an environment in which the use time of the resource is included in the daytime time period, an environment in which the use time of the resource is included in the evening time period, or an environment in which the use time of the resource is included in the night time period.
Also in this case, the prediction unit 36 predicts a temporal change in the use amount of the resource to be used in the future by the container 11-n from a part of data included in the monitoring data MDn.
Note that, in the present disclosure, free combinations of the embodiments, modifications of any components of the embodiments, or omissions of any components in the embodiments are possible.
The present disclosure is suitable for a resource management device and a resource management method.
1: container distribution server, 2: node, 11-1 to 11-N: container, 12: resource management device, 13: OS, 14: container runtime, 31: data acquiring unit, 32: prediction unit, 33: container deleting unit, 34: learning model, 35: prediction unit, 36: prediction unit, 41: data acquiring circuit, 42: prediction circuit, 43: container deleting circuit, 44: learning circuit, 45: prediction circuit, 46: prediction circuit, 51: memory, 52: processor
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/JP2022/003706 | 2/1/2022 | WO |