CN111198754B - Task scheduling method and device - Google Patents

Abstract

The invention discloses a task scheduling method and device. The method comprises the following steps: the resource use data of each node in the current period is obtained, the resource use data of each node in the next period is predicted by taking the resource use data in the current period as the input of the bidirectional cyclic neural network model, and then the tasks in each node can be scheduled according to the resource use data of each node in the next period. The embodiment of the invention adopts the bidirectional cyclic neural network model, so that the resource use data of each node in the next period can be predicted more accurately, and the accuracy of scheduling the tasks in each node is further improved.

Description

Method and device for task scheduling

technical field

The present invention relates to the field of software technology, in particular to a task scheduling method and device.

Background technique

The Kubernetes system is an open source container cluster management project designed and developed by Google. It is designed to provide an automated, scalable, and extensible operating platform for container clusters. Using the Kubernetes system can easily manage containerized applications and solve communication problems between containers.

The design of the task scheduling method in the Kubernetes system needs to start from the perspective of maximizing resource utilization, so that it can trigger dynamic scheduling in advance before resource bottlenecks occur. In order for the Kubernetes system to respond before a resource bottleneck occurs, it is necessary to predict the resource demand of the application in a certain period of time in the future, and then perform dynamic resource scheduling based on the predicted value. The current forecasting method mainly uses a simple regression model to predict the future resource demand. However, these methods are easily interfered by the subjective factors of the forecaster, have relatively high quality requirements for historical data, and have low forecasting accuracy.

Based on this, there is an urgent need for a task scheduling method, which is used to solve the problem of low scheduling accuracy due to low prediction accuracy of future resource demand in the process of task scheduling in the prior art.

Contents of the invention

Embodiments of the present invention provide a task scheduling method and device to solve the technical problem of low scheduling accuracy due to low prediction accuracy of future resource demand in the process of task scheduling in the prior art.

An embodiment of the present invention provides a task scheduling method, the method comprising:

Obtain the resource usage data of each node in the current cycle;

The resource usage data of each node in the current cycle is used as the input of the two-way cyclic neural network model to predict the resource usage data of each node in the next cycle; the two-way cyclic neural network model is based on the resources of each node in the historical cycle Obtained by using data for training, the historical period is a period before the current period;

According to the resource usage data of each node in the next period, the tasks in each node are scheduled.

In this way, the embodiment of the present invention uses the resource usage data in the current cycle as the input of the bidirectional cyclic neural network model to predict the resource usage data of each node in the next cycle, and then schedule tasks in each node accordingly. The use of a bidirectional cyclic neural network model can more accurately predict the resource usage data of each node in the next cycle, thereby improving the accuracy of scheduling tasks in each node; further, the embodiments of the present invention can automatically monitor and collect , process data, and conduct learning analysis and make predictions through the two-way cyclic neural network model, the whole process is intelligent, without manual intervention; furthermore, in the embodiment of the present invention, it can The resource usage data within a week flexibly adjusts the thresholds based on scheduling, so that task scheduling is more flexible and reasonable, and tasks can be scheduled in a timely, accurate, and dynamic manner.

In a possible implementation manner, the resource usage data includes at least one resource usage amount and a resource usage status, and the resource usage status is determined according to the at least one resource usage amount;

The bidirectional cyclic neural network model is obtained by training according to at least one resource usage and resource usage status of each node in the historical period.

In a possible implementation manner, the at least one resource usage includes memory usage, CPU usage, disk usage, and IO throughput.

In a possible implementation manner, the bidirectional cyclic neural network model is obtained by training according to at least one resource usage and resource usage status of each node in the historical period, including:

Obtaining at least one resource usage and resource usage status of each node in any historical period;

Using at least one resource usage and resource usage status of each node in the first historical period as an input parameter of the training sample, and using at least one resource usage and resource usage status of each node in the second historical period as the The output parameters of the training samples; the first historical period is the previous period of the second historical period;

Using the training samples to train the bidirectional cyclic neural network model to obtain the bidirectional cyclic neural network model.

In a possible implementation, the bidirectional cyclic neural network model includes a forward neural network layer and a reverse neural network layer;

The method also includes:

Receive the user's model update instruction;

According to the model update instruction, the number of layers of the forward neural network layer and/or the reverse neural network layer in the bidirectional recurrent neural network model is modified.

An embodiment of the present invention provides a task scheduling device, the device comprising:

The collection unit is used to obtain the resource usage data of each node in the current cycle;

The processing unit is configured to use the resource usage data of each node in the current cycle as the input of the bidirectional cyclic neural network model, and predict the resource usage data of each node in the next cycle; the bidirectional cyclic neural network model is based on the historical cycle The resource usage data of each node in the network is obtained through training, and the historical cycle is a cycle before the current cycle;

A scheduling unit, configured to schedule the tasks in each node according to the resource usage data of each node in the next period.

In a possible implementation manner, the resource usage data includes at least one resource usage amount and a resource usage status, and the resource usage status is determined according to the at least one resource usage amount;

The bidirectional cyclic neural network model is obtained by training according to at least one resource usage and resource usage status of each node in the historical period.

In a possible implementation manner, the at least one resource usage includes memory usage, CPU usage, disk usage, and IO throughput.

In a possible implementation manner, the processing unit is specifically configured to:

Obtaining at least one resource usage and resource usage status of each node in any historical period;

Using at least one resource usage and resource usage status of each node in the first historical period as an input parameter of the training sample, and using at least one resource usage and resource usage status of each node in the second historical period as the The output parameters of the training samples; the first historical period is the previous period of the second historical period;

Using the training samples to train the bidirectional cyclic neural network model to obtain the bidirectional cyclic neural network model.

In a possible implementation, the bidirectional cyclic neural network model includes a forward neural network layer and a reverse neural network layer;

The receiving unit is also used to receive a user's model update instruction;

The processing unit is further configured to modify the number of layers of the forward neural network layer and/or the reverse neural network layer in the bidirectional recurrent neural network model according to the model update instruction.

Embodiments of the present application also provide an apparatus, which has a function of implementing the task scheduling method described above. This function can be implemented by hardware executing corresponding software. In a possible design, the device includes: a processor, a transceiver, and a memory; the memory is used to store computer-executed instructions, and the transceiver is used to realize the communication between the device and other The entity communicates, the processor and the memory are connected through the bus, and when the device is running, the processor executes the computer-executable instructions stored in the memory, so that the device executes the task scheduling method described above.

An embodiment of the present invention also provides a computer storage medium, where a software program is stored in the storage medium, and when the software program is read and executed by one or more processors, the tasks described in the above-mentioned various possible implementation modes are realized Scheduling method.

An embodiment of the present invention also provides a computer program product including instructions, which, when run on a computer, cause the computer to execute the task scheduling method described in the foregoing various possible implementation manners.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings that need to be used in the description of the embodiments.

FIG. 1 is a schematic diagram of a system architecture applicable to an embodiment of the present invention;

FIG. 2 is a schematic flowchart corresponding to a task scheduling method provided by an embodiment of the present invention;

Fig. 3 is a schematic structural diagram of a bidirectional recurrent neural network model provided by an embodiment of the present invention;

Fig. 4 is a schematic structural diagram of a task scheduling device provided by an embodiment of the present invention.

Detailed ways

The present application will be described in detail below in conjunction with the accompanying drawings, and the specific operation methods in the method embodiments can also be applied to the device embodiments.

Figure 1 exemplarily shows a schematic diagram of a system architecture applicable to an embodiment of the present invention. As shown in Figure 1, a system 100 applicable to an embodiment of the present invention includes a scheduling device 101 and at least one node, such as the node shown in Figure 1 1021 , node 1022 and node 1023 . Wherein, the scheduling device 101 can be connected to each node through a server interface (API Server). For example, as shown in FIG. Connect to node 1023 through API Server.

Further, taking the system 100 as an example of a Kubernetes system, the scheduling device 101 may include a management module 1011 , a resource monitoring module 1012 , an in-depth analysis module 1013 and a resource scheduling module 1014 . Each node may include a Kubelet component and at least one Pod, for example, node 1021 may include a Kubelet component 1211, Pod1212, and Pod1213, node 1022 may include a Kubelet component 1221, Pod1222, and Pod1223, and node 1023 may include a Kubelet component 1231, Pod1232, and Pod1233 .

Specifically, the management module 1011 can be responsible for creating Pods in the nodes. For example, the management module 1011 can create Pod1212 and Pod1213 in the node 1021, can also create Pod1222 and Pod1223 in the node 1022, and can also create Pod1232 and Pod1233 in the node 1023 .

The resource monitoring module 1012 can be used to collect resource usage data in each node, and can aggregate them in Pod units, and provide them to the in-depth analysis module 1013 for use.

The in-depth analysis module 1013 can perform training through a deep learning model according to the resource usage data provided by the resource monitoring module 1012 to predict resource usage data in a period of time in the future.

Further, as shown in FIG. 1 , the depth analysis module 1013 may include a data processing module 1131 , a depth prediction module 1132 and a model update module 1133 . Among them, the data processing module 1131 can be used to preprocess the resource usage data collected by the resource monitoring module 1012; the depth prediction module 1132 can send the processing results of the data processing module 1131 into the deep learning model for learning, and perform future resource usage data Prediction; the model update module 1133 can receive the administrator's model update instruction through the network, and can update the model structure according to the requirement.

The resource scheduling module 1014 can generate a scheduling policy according to the prediction result of the in-depth analysis module 1013, and deliver it to each node.

Further, as shown in FIG. 1 , the resource scheduling module 1014 may include a policy generation module 1141 and a policy delivery module 1142 . Among them, the policy generation module 1141 can generate corresponding scheduling policies according to the processing results of the in-depth analysis module 1013; the policy delivery module 1142 can deliver the policies generated by the policy generation module 1141 to each node, so as to realize reasonable scheduling of system resources .

Based on the system architecture shown in FIG. 1, FIG. 2 exemplarily shows a schematic flowchart corresponding to a task scheduling method provided in an embodiment of the present invention, including the following steps:

Step 201, acquire resource usage data of each node in the current period.

Step 202, using the resource usage data of each node in the current cycle as the input of the bidirectional cyclic neural network model, and predicting and obtaining the resource usage data of each node in the next cycle.

Step 203, according to the resource usage data of each node in the next period, schedule the tasks in each node.

In this way, the embodiment of the present invention uses the resource usage data in the current cycle as the input of the bidirectional cyclic neural network model to predict the resource usage data of each node in the next cycle, and then schedule tasks in each node accordingly. The use of a bidirectional cyclic neural network model can more accurately predict the resource usage data of each node in the next cycle, thereby improving the accuracy of scheduling tasks in each node; further, the embodiments of the present invention can automatically monitor and collect , process data, and conduct learning analysis and make predictions through the two-way cyclic neural network model, the whole process is intelligent, without manual intervention; furthermore, in the embodiment of the present invention, it can The resource usage data within a week flexibly adjusts the thresholds based on scheduling, so that task scheduling is more flexible and reasonable, and tasks can be scheduled in a timely, accurate, and dynamic manner.

Specifically, the resource usage data may include at least one resource usage amount and resource usage status.

Wherein, the resource usage may be memory usage, CPU usage, disk usage, IO throughput, etc., and is not specifically limited. Furthermore, according to the content shown in Figure 1, it can be seen that there may be multiple Pods in each node. When obtaining the resource usage of each node, the resource usage of each Pod in each node can be obtained first, and then according to the The resource usage of determines the resource usage of the node.

As shown in Table 1, it is an example of resource usage in the embodiment of the present invention. Among them, Node A includes Pod A-1 and Pod A-2, the memory usage of Pod A-1 is 20%, the CPU usage is 50%, the disk usage is 10%, and the IO throughput is 30%; Pod A The memory usage of -2 is 25%, the CPU usage is 30%, the disk usage is 80%, and the IO throughput is 30%, so it can be seen that the memory usage of node A is 45%, and the CPU usage is 80% %, disk usage is 90%, and IO throughput is 60%. Node B includes Pod B-1 and Pod B-2. The memory usage of Pod B-1 is 35%, the CPU usage is 55%, the disk usage is 20%, and the IO throughput is 30%; Pod B-2 The memory usage is 25%, the CPU usage is 45%, the disk usage is 55%, and the IO throughput is 60%. It can be seen that the memory usage of Node B is 60%, and the CPU usage is 100%. Disk usage is 75%, IO throughput is 90%.

Table 1: An example of resource usage in the embodiment of the present invention

It should be noted that the calculation method for the resource usage of a node shown above is only an example. In other possible implementations, the resource usage of a node can also be based on the resource usage of each Pod and the The weight is calculated, which will not be described in detail.

In this embodiment of the present invention, the resource usage status may be determined according to at least one resource usage amount. Specifically, taking the resource usage of node A shown in Table 1 as an example, the resource usage status of node A can be determined according to the memory usage, CPU usage, disk usage and IO throughput of node A. For example, when calculating the resource usage status of node A, it can be determined whether the number of resource usage in node A whose resource usage is greater than the usage threshold is greater than the number threshold. The usage status is a high load status, otherwise, it can be determined that the resource usage status of node A is a low load status. For another example, when calculating the resource usage status of node A, the load of node A can be determined according to the memory usage, CPU usage, disk usage and IO throughput of node A, and then by judging node A Whether the medium load is greater than the load threshold, if it is greater, it can be determined that the resource usage state of node A is a high load state, otherwise, it can be determined that the resource usage state of node A is a low load state.

Before step 202 is executed, the embodiment of the present invention may preprocess the resource usage data of each node. Wherein, the preprocessing may include two steps of format conversion and dimension reconstruction. Format conversion refers to converting the obtained resource usage data into a format that the neural network model can recognize. Depending on the type of data used by the resource, the format conversion method is also different. For example, the data field of the Boolean type can be converted into a binary value format and used as the input data format; Bag of Word (BOW) method for format conversion, as the input data format; or, you can also keep the original type of the data field of the numeric type, as the input data format.

Furthermore, dimensionality reconstruction can be performed on the format-converted data, so as to construct a dimension that meets the input requirements of the deep learning model.

In step 202, as shown in FIG. 3 , it is a schematic structural diagram of a bidirectional recurrent neural network model provided by an embodiment of the present invention. Among them, the bidirectional cyclic neural network model can include 1 layer of input layer (Input Layer), 3 layers of forward neural network layer (Forward Layer), 3 layers of reverse neural network layer (backward layer), 2 layers of fully connected layer (FullyConnected Layer) and 1 layer output layer (Output Layer). When the bidirectional cyclic neural network model performs calculations, the processed data is first input to the input layer. After the input layer performs calculation processing, the calculation results are simultaneously input to the forward neural network layer and the reverse neural network layer. After the calculation of the neural network layer and the reverse neural network layer, the results are combined and input to the fully connected layer. After the fully connected layer processes the data, the output layer predicts the output.

Specifically, the bidirectional cyclic neural network model can be obtained by training according to the resource usage data of each node in a historical period, where the historical period can be a period before the current period. Further, the bidirectional cyclic neural network model is obtained by training according to at least one resource usage and resource usage status of each node in the historical period.

Further, when training the bidirectional cyclic neural network model, the resource usage data can be preprocessed to form a time series as a training sample. Specifically, at least one resource usage and Resource usage state; then, use at least one resource usage amount and resource usage state of each node in the first historical period as the input parameters of the training sample, and use at least one resource usage amount and resource usage state of each node in the second historical cycle as The output parameters of the training samples, wherein the first historical period is the previous period of the second historical period; finally, the training samples can be used to train the bidirectional recurrent neural network model, so as to obtain the bidirectional recurrent neural network model.

In step 203, according to the resource usage data of each node in the next period, a corresponding scheduling strategy can be generated, and then the Pods in each node can be scheduled according to the scheduling strategy, so as to realize reasonable resource scheduling.

Considering that the accuracy of the bidirectional cyclic neural network model may decrease as the amount of data continues to expand, in the embodiment of the present invention, the accuracy of the bidirectional cyclic neural network model may also be improved by updating the model. Specifically, firstly, a user's model update instruction can be received, and then the number of forward neural network layers and/or reverse neural network layers in the bidirectional recurrent neural network model can be modified according to the model update instruction. In this way, when the accuracy of the bidirectional cyclic neural network model decreases, the choice of the optimizer can be changed to reduce the learning rate by increasing the number of model layers.

Based on the same idea, an embodiment of the present invention provides a task scheduling device, as shown in FIG. 4 , the device includes a collection unit 401, a processing unit 402, and a scheduling unit 403; wherein,

A receiving unit 401, configured to acquire resource usage data of each node in the current cycle;

The processing unit 402 is configured to use the resource usage data of each node in the current cycle as the input of the bidirectional cyclic neural network model, and predict the resource usage data of each node in the next cycle; the bidirectional cyclic neural network model is based on historical The resource usage data of each node in the cycle is obtained by training, and the historical cycle is a cycle before the current cycle;

The scheduling unit 403 is configured to schedule the tasks in each node according to the resource usage data of each node in the next cycle.

In a possible implementation manner, the resource usage data includes at least one resource usage amount and a resource usage status, and the resource usage status is determined according to the at least one resource usage amount;

The bidirectional cyclic neural network model is obtained by training according to at least one resource usage and resource usage status of each node in the historical period.

In a possible implementation manner, the at least one resource usage includes memory usage, CPU usage, disk usage, and IO throughput.

In a possible implementation manner, the processing unit 402 is specifically configured to:

Obtaining at least one resource usage and resource usage status of each node in any historical period;

Using at least one resource usage and resource usage status of each node in the first historical period as an input parameter of the training sample, and using at least one resource usage and resource usage status of each node in the second historical period as the The output parameters of the training samples; the first historical period is the previous period of the second historical period;

Using the training samples to train the bidirectional cyclic neural network model to obtain the bidirectional cyclic neural network model.

In a possible implementation, the bidirectional cyclic neural network model includes a forward neural network layer and a reverse neural network layer;

The receiving unit 401 is also configured to receive a user's model update instruction;

The processing unit 402 is further configured to modify the number of layers of the forward neural network layer and/or the reverse neural network layer in the bidirectional recurrent neural network model according to the model update instruction.

Embodiments of the present application also provide an apparatus, which has a function of implementing the task scheduling method described above. This function can be implemented by hardware executing corresponding software. In a possible design, the device includes: a processor, a transceiver, and a memory; the memory is used to store computer-executed instructions, and the transceiver is used to realize the communication between the device and other The entity communicates, the processor and the memory are connected through the bus, and when the device is running, the processor executes the computer-executable instructions stored in the memory, so that the device executes the task scheduling method described above.

An embodiment of the present invention also provides a computer storage medium, where a software program is stored in the storage medium, and when the software program is read and executed by one or more processors, the tasks described in the above-mentioned various possible implementation modes are realized Scheduling method.

An embodiment of the present invention also provides a computer program product including instructions, which, when run on a computer, cause the computer to execute the task scheduling method described in the foregoing various possible implementation manners.

Those skilled in the art should understand that the embodiments of the present invention may be provided as methods, systems, or computer program products. Accordingly, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

While preferred embodiments of the invention have been described, additional changes and modifications to these embodiments can be made by those skilled in the art once the basic inventive concept is appreciated. Therefore, it is intended that the appended claims be construed to cover the preferred embodiment as well as all changes and modifications which fall within the scope of the invention.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims (8)

1. A method of task scheduling, the method comprising:

acquiring resource utilization data of each node in a current period, wherein the resource utilization data comprises at least one resource utilization amount and a resource utilization state, the resource utilization state is determined according to the at least one resource utilization amount, and the resource utilization state is determined according to the relation between the number of the resource utilization amounts, of which the resource utilization amount is larger than a utilization amount threshold value, in the node and a number threshold value; the at least one resource usage includes memory usage, CPU usage, disk usage, and IO throughput;

taking the resource use data of each node in the current period as the input of a bidirectional cyclic neural network model, and predicting to obtain the resource use data of each node in the next period; the bidirectional cyclic neural network model is obtained by training according to resource use data of each node in a history period, wherein the history period is a period before the current period;

and scheduling the tasks in each node according to the resource use data of each node in the next period.

2. The method of claim 1, wherein the bi-directional recurrent neural network model is trained from resource usage data for each node over the historical period, comprising:

acquiring at least one resource usage amount and resource usage state of each node in any history period;

taking at least one resource usage amount and resource usage state of each node in a first history period as input parameters of a training sample, and taking at least one resource usage amount and resource usage state of each node in a second history period as output parameters of the training sample; the first history period is the last period of the second history period;

and training the bidirectional circulating neural network model by using the training sample to obtain the bidirectional circulating neural network model.

3. The method according to claim 1 or 2, wherein the two-way recurrent neural network model comprises a forward neural network layer and a reverse neural network layer;

the method further comprises the steps of:

receiving a model updating instruction of a user;

and modifying the layer number of the forward neural network layer and/or the reverse neural network layer in the bidirectional circulating neural network model according to the model updating instruction.

4. A task scheduling device, the device comprising:

a receiving unit, configured to obtain resource usage data of each node in a current period, where the resource usage data includes at least one resource usage amount and a resource usage state, where the resource usage state is determined according to the at least one resource usage amount, and the resource usage state is determined according to a relationship between a number of resource usage amounts in the node where the resource usage amount is greater than a usage amount threshold and a number threshold; the at least one resource usage includes memory usage, CPU usage, disk usage, and IO throughput;

the processing unit is used for taking the resource use data of each node in the current period as the input of a bidirectional cyclic neural network model, and predicting to obtain the resource use data of each node in the next period; the bidirectional cyclic neural network model is obtained by training according to resource use data of each node in a history period, wherein the history period is a period before the current period;

and the scheduling unit is used for scheduling the tasks in each node according to the resource use data of each node in the next period.

5. The apparatus of claim 4, wherein the processing unit is specifically configured to:

acquiring at least one resource usage amount and resource usage state of each node in any history period;

taking at least one resource usage amount and resource usage state of each node in a first history period as input parameters of a training sample, and taking at least one resource usage amount and resource usage state of each node in a second history period as output parameters of the training sample; the first history period is the last period of the second history period;

and training the bidirectional circulating neural network model by using the training sample to obtain the bidirectional circulating neural network model.

6. The apparatus of claim 4 or 5, wherein the two-way recurrent neural network model comprises a forward neural network layer and a reverse neural network layer;

the receiving unit is also used for receiving a model updating instruction of a user;

the processing unit is further configured to modify the number of layers of the forward neural network layer and/or the reverse neural network layer in the bidirectional recurrent neural network model according to the model update instruction.

7. A computer readable storage medium storing instructions which, when run on a computer, cause the computer to carry out the method of any one of claims 1 to 3.

8. A computer device, comprising:

a memory for storing program instructions;

a processor for invoking program instructions stored in said memory and for performing the method according to any of claims 1 to 3 in accordance with the obtained program.

Images