Artificial neural network-based cloud data center failure prediction device and method

Abstract

An apparatus and method for predicting failures in a cloud data center are disclosed. An apparatus for predicting failures in a cloud data center according to one disclosed embodiment includes a data acquisition module for acquiring work-related data of a cloud data center, a preprocessing module for preprocessing the acquired work-related data, and a failure prediction module for inputting the preprocessed work-related data into one or more neural network models to predict the end state of each work.

Description

Technology Field

[0001] An embodiment of the present invention relates to a failure prediction technology for a cloud data center. Background Technology

[0003] Cloud data centers play a pivotal role in meeting the continuously increasing demand for computing resources and services. By concentrating servers, networking equipment, and storage systems on a large scale, cloud data centers provide users with stable and scalable services. However, as the complexity of the execution environment and the scale of resources grow, it becomes difficult to guarantee continuous operation and minimize downtime due to the failure of any single component. Nevertheless, the rapid prediction of operational failures is essential for maintaining the reliability and availability of cloud data centers.

[0004] Existing methods for predicting job failure often rely on rule-based approaches or simple statistical models, which has the problem of failing to adequately grasp the complex relationships and time-dependent nature of cloud data centers. Prior art literature

[0006] Korean Patent Publication No. 10-2022-0093948 (July 5, 2022) The problem to be solved

[0007] The disclosed embodiment is intended to provide a new technique for predicting failures in cloud data centers. means of solving the problem

[0009] A failure prediction device for a cloud data center according to a disclosed embodiment comprises one or more processors and a memory for storing one or more programs executed by the one or more processors, and includes: a data acquisition module for acquiring work-related data of the cloud data center; a preprocessing module for preprocessing the acquired work-related data; and a failure prediction module for inputting the preprocessed work-related data into one or more neural network models to predict the end state of each work.

[0010] The above task-related data may include one or more of the event type, timestamp, CPU usage, memory usage, cache memory usage, average disk I / O time, disk usage, task priority, number of task resubmissions, and task scheduling delay for the said task.

[0011] The above event types may represent types of states resulting from the execution of the corresponding task.

[0012] The above preprocessing module can exclude work-related data among the acquired work-related data where the event type indicates an incomplete work life that makes it difficult to determine whether the corresponding work has ended.

[0013] The above preprocessing module can exclude the acquired task-related data where the event type indicates an incomplete task lifespan, and then classify and label the termination status of the remaining task-related data into two categories, success and failure, based on the event type of each task.

[0014] The above failure prediction module includes a teacher model and a student model, inputs a sequence of the preprocessed task-related data into the teacher model so that the teacher model predicts the end state of each task, and when the training of the teacher model is finished, inputs the sequence of the preprocessed task-related data into the teacher model and the student model, respectively, so that the student model predicts the end state of each task.

[0015] The teacher model and the student model are each composed of a Bi-LSTM (Bi-directional Long Short Term Memory) neural network architecture, and the number of LSTM units in the forward LSTM and backward LSTM layers of the student model can be smaller than the number of LSTM units of the teacher model.

[0016] The neural network architecture of the above Bi-LSTM (Bi-directional Long Short Term Memory) may include: an input layer that receives a sequence of the preprocessed task-related data; a forward LSTM layer that generates a forward hidden state by encoding the sequence of the task-related data in the forward direction; a backward LSTM layer that generates a backward hidden state by encoding the sequence of the task-related data in the reverse direction; an output layer that outputs a combined representation of the forward hidden state and the backward hidden state; and a classification layer that predicts whether the end state of each task is success or failure based on the combined representation using a probability distribution.

[0017] The total loss for training the above student model can be expressed as the sum of a knowledge distillation loss that minimizes the difference between the predicted value of the above student model and the predicted value of the above teacher model, and a cross-entropy loss that minimizes the difference between the predicted value of the above student model and a pre-set correct answer value.

[0018] A method for predicting failures in a cloud data center according to a disclosed embodiment is a method performed in a computing device having one or more processors and a memory for storing one or more programs executed by said one or more processors, comprising: a step of acquiring task-related data of the cloud data center; a step of preprocessing said acquired task-related data; and a step of inputting said preprocessed task-related data into one or more neural network models to predict the end state of each task. Effects of the invention

[0020] According to the disclosed embodiments, since the success or failure of each task in a cloud data center can be predicted, downtime due to task failure can be minimized and resources can be allocated more efficiently. In addition, by predicting the probability of failure due to task failure in advance, costs associated with system recovery and downtime in the cloud data center can be reduced. Furthermore, through failure prediction, more reliable services can be provided to users. Brief explanation of the drawing

[0022] FIG. 1 is a block diagram showing the configuration of a fault prediction device for a cloud data center according to an embodiment of the present invention. FIG. 2 is a schematic diagram showing the structure of a Bi-LSTM (Bi-directional Long Short Term Memory) according to an embodiment of the present invention. FIG. 3 is a diagram schematically illustrating the state of knowledge distillation between a teacher model and a student model in one embodiment of the present invention. FIG. 4 is a flowchart illustrating a method for predicting failures in a cloud data center according to an embodiment of the present invention. FIG. 5 is a block diagram illustrating a computing environment including a computing device suitable for use in exemplary embodiments. Specific details for implementing the invention

[0023] Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The following detailed description is provided to facilitate a comprehensive understanding of the methods, apparatus, and / or systems described herein. However, this is merely illustrative and the present invention is not limited thereto.

[0024] In describing the embodiments of the present invention, detailed descriptions of known technologies related to the present invention are omitted if it is determined that such detailed descriptions may unnecessarily obscure the essence of the present invention. Furthermore, the terms described below are defined in consideration of their functions within the present invention, and these may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification. Terms used in the detailed description are intended merely to describe the embodiments of the present invention and should not be limiting in any way. Unless explicitly stated otherwise, expressions in the singular form include the meaning of the plural form. In this description, expressions such as "include" or "comprise" are intended to refer to certain characteristics, numbers, steps, actions, elements, parts thereof, or combinations thereof, and should not be interpreted to exclude the existence or possibility of one or more other characteristics, numbers, steps, actions, elements, parts thereof, or combinations thereof other than those described.

[0025] Additionally, terms such as "first," "second," etc., may be used to describe various components, but said components should not be limited by said terms. These terms may be used for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component.

[0027] FIG. 1 is a block diagram showing the configuration of a failure prediction device for a cloud data center according to one embodiment of the present invention.

[0028] Referring to FIG. 1, a failure prediction device (100) of a cloud data center may include a data acquisition module (102), a preprocessing module (104), and a failure prediction module (106).

[0029] The data acquisition module (102) can acquire work-related data from a pre-configured cloud data center. In one embodiment, the data acquisition module (102) can acquire work-related data in units of a pre-configured period, but is not limited thereto. For example, if the cloud data center is Google's data center, the work-related data can use the Google Trace Dataset.

[0030] In this case, a single task-related data set may include the event type, CPU usage, memory usage, cache memory usage, average disk I / O time, disk usage, task priority, number of task resubmissions, and task scheduling delay for the task. Additionally, each task-related data set includes a timestamp (i.e., information about the time of occurrence of the task-related data).

[0031] Here, the event types for a task represent the types of states associated with the execution of the task; for example, event types can be classified as job submission, job schedule, evicted, fail, finished, kill, update while in the queue, update while running, and lost.

[0032] The preprocessing module (104) can preprocess work-related data from a cloud data center. The preprocessing module (104) can preprocess the acquired work-related data into a form that can be learned by neural network models described later.

[0033] Specifically, the preprocessing module (104) can exclude the work-related data if there are missing values ​​for certain items (e.g., CPU usage, memory usage, disk usage, etc.) in the acquired work-related data. Additionally, the preprocessing module (104) can convert the date value, which is a timestamp of the work-related data, into microseconds.

[0034] Additionally, the preprocessing module (104) may exclude some job-related data from the acquired job-related data based on the event type of the corresponding job. In one embodiment, the preprocessing module (104) may exclude job-related data if the event type of the corresponding job indicates an incomplete job life. Here, an incomplete job life may refer to an event type where it is difficult to determine whether the job has been completed. For example, event types indicating an incomplete job life may include job submission, job schedule, update while in the queue, and update while running.

[0035] The preprocessing module (104) can exclude tasks from the acquired task-related data where the task event type indicates an incomplete task lifespan, and then classify the termination status of the remaining task-related data into two categories, success and failure, based on the task event type.

[0036] Here, the event types of the remaining job-related data, excluding those where the job event type indicates an incomplete job life, include evicted, fail, finished, kill, and lost. At this time, the preprocessing module (104) can classify the termination status of job-related data where the event type is finished as "success," and classify the termination status of job-related data where the event type is evicted, fail, kill, and lost as "failure."

[0037] Additionally, the preprocessing module (104) can label information regarding the termination status for each task-related data. That is, the preprocessing module (104) can label the termination status as "success" for task-related data where the event type is "finished," and label the termination status as "failure" for task-related data where the event type is "evicted," "fail," "kill," and "lost." The preprocessing module (104) can transmit the task-related data with the termination status labeled to the failure prediction module (106).

[0038] The failure prediction module (106) can perform work failure prediction of a cloud data center based on work-related data transmitted from the preprocessing module (104). The failure prediction module (106) can perform work failure prediction of a cloud data center using artificial neural network technology based on knowledge distillation. To this end, the failure prediction module (106) may include a teacher model (106a) and a student model (106b).

[0039] The teacher model (106a) can be trained using task-related data transmitted from the preprocessing module (104) as a training data set. The teacher model (106a) can be trained to receive task-related data as input and predict whether the task will succeed or fail. In one embodiment, the teacher model (106a) may be a model having a Bi-LSTM (Bi-directional Long Short Term Memory) neural network architecture.

[0040] FIG. 2 is a schematic diagram illustrating the structure of a Bi-LSTM (Bi-directional Long Short Term Memory) according to an embodiment of the present invention. Referring to FIG. 2, the input layer receives task-related data. At this time, the input layer comprises a plurality of task-related data having a time series (x1, x2, ↳, x m ) can be entered.

[0041] Forward LSTM can perform forward propagation by processing sequences of task-related data in a forward order (from the beginning to the end). Through forward propagation, Forward LSTM can generate a forward hidden state that encodes temporal dependencies by capturing past and future information at each time step for sequences of task-related data.

[0042] Backward LSTM can perform backpropagation by processing sequences of task-related data in reverse order (from the end to the beginning). Through backpropagation, Backward LSTM can generate a backward hidden state that encodes temporal dependencies by capturing past and future information at each time step for sequences of task-related data.

[0043] The output layer can output a combined representation of the forward and backward hidden states of sequences of task-related data. Then, the classifier classifies the end state of each task as success or failure based on the combined representation of the output layer.

[0044] In this way, by using the Bi-LSTM (Bi-directional Long Short Term Memory) neural network architecture, forward and backward information can be combined to better understand the temporal dependency relationships (precedence relationships) of sequences of task-related data.

[0045] Meanwhile, the teacher model (106a) can transfer its knowledge to the student model (106b) after the learning is completed. The teacher model (106a) can transfer its knowledge to the student model (106b) through a knowledge distillation technique during the student model (106b)'s learning.

[0046] The teacher model (106a) may refer to a neural network model that has already been trained on the task that the student model (106b) intends to perform, and may refer to a model with a larger neural network size compared to the student model (106b). Here, a larger neural network size may mean that there are more neural network layers, kernels, nodes, or parameters. The student model (106b) may refer to a neural network model that is trained by utilizing knowledge acquired from the previously trained teacher model (106a), and may refer to a model with a smaller neural network size compared to the teacher model (106a).

[0047] FIG. 3 is a schematic diagram illustrating the state of knowledge distillation between a teacher model (106a) and a student model (106b) in one embodiment of the present invention. Referring to FIG. 3, the teacher model (106a) and the student model (106b) may be composed of the same neural network structure. That is, both the teacher model (106a) and the student model (106b) may be composed of a Bi-LSTM (Bi-directional Long Short Term Memory) neural network architecture.

[0048] However, the number of LSTM units in the forward LSTM and backward LSTM layers of the student model (106b) may be smaller than that of the teacher model (106a). For example, if the number of LSTM units in the forward LSTM and backward LSTM layers of the teacher model (106a) is 128, the number of LSTM units in the corresponding layers of the student model (106b) may be 64.

[0049] During the training of the student model (106b), a sequence of task-related data can be input to the teacher model (106a) and the student model (106b), respectively. The teacher model (106a) and the student model (106b) can each predict whether the end state of each task is successful or unsuccessful based on the input sequence of task-related data. At this time, the parameters of the teacher model (106a) are not learned but have fixed values ​​in a state where learning is complete.

[0050] The teacher model (106a) and the student model (106b) each output a representation combining the forward hidden state and the backward hidden state of sequences of task-related data in the output layer. Then, the teacher model (106a) and the student model (106b) can each output a softmax distribution as a predicted value through a softmax function in the classification layer. That is, the predicted value regarding whether the task's completion state is success or failure can be represented as a softmax distribution.

[0051] The student model (106b) can be trained such that the difference between the predicted value of the student model (106b) and the predicted value of the teacher model (106a) is minimized. That is, the student model (106b) can be trained to mimic the learned knowledge of the teacher model (106a). Through this process, the learned knowledge of the teacher model (106a) can be transferred to the student model (106b). The student model (106b) can be trained through a knowledge distillation loss that minimizes the difference between the predicted value of the student model (106b) and the predicted value of the teacher model (106a). Knowledge distillation loss (L kd ) can be expressed as shown in the following mathematical formula 1.

[0052] (Mathematical Formula 1)

[0053]

[0054] τ: A pre-set constant for scaling the softmax distribution

[0055] x : Task-related data

[0056] D x : Training dataset containing task-related data

[0057] C: Number of classes

[0058] h i T (x; τ): Softmax distribution of the i-th logit in the teacher model

[0059] h i S (x; τ): Softmax distribution of the i-th logit in the student model

[0060] Additionally, the student model (106b) can be trained to minimize the difference between the predicted value of the student model (106b) (a value regarding whether the end status of the task classified by the student model (106b) is successful or failed) and the actual ground truth value (a value regarding whether the end status of the actual task is successful or failed). The student model (106b) can be trained through a cross-entropy loss that minimizes the difference between the predicted value of the student model (106b) and the ground truth value.

[0061] The total loss for training the student model (106b) can be expressed as the sum of the knowledge distillation loss and the cross-entropy loss. The total loss (L) for training the student model (106b) can be expressed by the following mathematical formula 2.

[0062] (Mathematical Formula 2)

[0063]

[0064] L ce : Cross-entropy loss

[0065] L kd : Knowledge distillation loss

[0066] λ1, λ2: Pre-set weights

[0067] Here, λ1+λ2= 1. Also, λ2 can be set to a value greater than λ1. For example, λ2= 0.8 and λ1= 0.2, but is not limited thereto.

[0068] According to the disclosed embodiments, since the success or failure of each task in a cloud data center can be predicted, downtime due to task failure can be minimized and resources can be allocated more efficiently. In addition, by predicting the probability of failure due to task failure in advance, costs associated with system recovery and downtime in the cloud data center can be reduced. Furthermore, through failure prediction, more reliable services can be provided to users.

[0069] In this specification, the term "module" may refer to a functional and structural combination of hardware for carrying out the technical concept of the present invention and software for driving said hardware. For example, the "module" may refer to a logical unit of a specific code and a hardware resource for executing said code, and does not necessarily refer to physically connected code or a single type of hardware.

[0071] FIG. 4 is a flowchart illustrating a method for predicting failures in a cloud data center according to an embodiment of the present invention. In the illustrated flowchart, the method is described by dividing it into a plurality of steps, but at least some of the steps may be performed in a different order, combined with other steps and performed together, omitted, divided into detailed steps, or performed with one or more steps not illustrated added.

[0072] Referring to FIG. 4, the failure prediction device (100) can acquire work-related data from a cloud data center (S 101). Next, the failure prediction device (100) can preprocess the work-related data from the cloud data center into a form that can be learned by the teacher model (106a) and the student model (106b) (S 103).

[0073] Next, the failure prediction device (100) can input a sequence of task-related data into a teacher model (106a) to train the teacher model (106a) to predict the end state for each task (S 105). Next, the failure prediction device (100) can train a student model (106b) after the training of the teacher model (106a) is finished (S 107).

[0075] FIG. 5 is a block diagram illustrating a computing environment (10) including a computing device suitable for use in exemplary embodiments. In the illustrated embodiments, each component may have different functions and capabilities in addition to those described below, and may include additional components in addition to those described below.

[0076] The illustrated computing environment (10) includes a computing device (12). In one embodiment, the computing device (12) may be a failure prediction device (100) of a cloud data center.

[0077] The computing device (12) includes at least one processor (14), a computer-readable storage medium (16), and a communication bus (18). The processor (14) can cause the computing device (12) to operate according to the exemplary embodiment described above. For example, the processor (14) can execute one or more programs stored in the computer-readable storage medium (16). The one or more programs may include one or more computer-executable instructions, and the computer-executable instructions may be configured to cause the computing device (12) to perform operations according to the exemplary embodiment when executed by the processor (14).

[0078] A computer-readable storage medium (16) is configured to store computer-executable instructions or program code, program data and / or other suitable forms of information. A program (20) stored in the computer-readable storage medium (16) includes a set of instructions executable by a processor (14). In one embodiment, the computer-readable storage medium (16) may be memory (volatile memory such as random access memory, non-volatile memory, or a suitable combination thereof), one or more magnetic disk storage devices, optical disk storage devices, flash memory devices, other forms of storage media that are accessed by a computing device (12) and capable of storing desired information, or a suitable combination thereof.

[0079] The communication bus (18) interconnects various other components of the computing device (12), including the processor (14) and the computer-readable storage medium (16).

[0080] The computing device (12) may also include one or more input / output interfaces (22) and one or more network communication interfaces (26) that provide interfaces for one or more input / output devices (24). The input / output interfaces (22) and network communication interfaces (26) are connected to a communication bus (18). The input / output devices (24) may be connected to other components of the computing device (12) through the input / output interfaces (22). An exemplary input / output device (24) may include an input device such as a pointing device (such as a mouse or trackpad), a keyboard, a touch input device (such as a touchpad or touchscreen), a voice or sound input device, various types of sensor devices and / or imaging devices, and / or an output device such as a display device, a printer, a speaker and / or a network card. An exemplary input / output device (24) may be included inside the computing device (12) as a component constituting the computing device (12), or it may be connected to the computing device (12) as a separate device distinct from the computing device (12).

[0082] Although representative embodiments of the present invention have been described in detail above, those skilled in the art will understand that various modifications can be made to the above-described embodiments without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be defined by the claims set forth below as well as equivalents thereof. Explanation of the symbols

[0084] 10: Computing Environment 12: Computing device 14 : Processor 16: Computer-readable storage media 18: Communication bus 20 : Program 22 : Input / Output Interface 24 : Input / Output Devices 26: Network communication interface 100: Fault Prediction Device 102: Data Acquisition Module 104 : Preprocessing Module 106: Fault Prediction Module 106a : Teacher model 106b : Student model

Claims

Claim 1 A fault prediction device for a cloud data center comprising one or more processors and a memory for storing one or more programs executed by said one or more processors, the device comprising: a data acquisition module for acquiring work-related data of the cloud data center; a preprocessing module for preprocessing said acquired work-related data; and a fault prediction module for inputting said preprocessed work-related data into one or more neural network models to predict the end state of each work, wherein the fault prediction module comprises a teacher model and a student model, inputting a sequence of said preprocessed work-related data into said teacher model to enable said teacher model to predict the end state of each work, and, when said teacher model has finished learning, inputting said preprocessed work-related data sequence into said teacher model and said student model, respectively to enable said student model to predict the end state of each work. Claim 2 A failure prediction device for a cloud data center according to claim 1, wherein the task-related data comprises one or more of an event type, timestamp, CPU utilization, memory utilization, cache memory utilization, average disk I / O time, disk usage, task priority, number of task resubmissions, and task scheduling delay for the said task. Claim 3 A cloud data center failure prediction device according to claim 2, wherein the event type represents types of states according to the performance of the corresponding operation. Claim 4 A failure prediction device for a cloud data center according to claim 3, wherein the preprocessing module excludes work-related data among the acquired work-related data in which the event type indicates an incomplete work life in which it is difficult to determine whether the corresponding work has been terminated. Claim 5 A failure prediction device for a cloud data center according to claim 4, wherein the preprocessing module excludes the acquired work-related data in which the event type indicates an incomplete work life, and then classifies and labels the termination status of the remaining work-related data into two categories, success and failure, based on the event type of each work. Claim 6 delete Claim 7 A fault prediction device for a cloud data center according to claim 1, wherein the teacher model and the student model are each composed of a Bi-LSTM (Bi-directional Long Short Term Memory) neural network architecture, and the number of LSTM units in the forward LSTM and backward LSTM layers of the student model is smaller than the number of LSTM units of the teacher model. Claim 8 A fault prediction device for a cloud data center according to claim 7, wherein the neural network architecture of the Bi-LSTM (Bi-directional Long Short Term Memory) comprises: an input layer receiving a sequence of preprocessed task-related data; a forward LSTM layer generating a forward hidden state by forward encoding the sequence of task-related data; a backward LSTM layer generating a backward hidden state by backward encoding the sequence of task-related data; an output layer outputting a combined representation of the forward hidden state and the backward hidden state; and a classification layer predicting whether the termination state of each task is success or failure based on the combined representation using a probability distribution. Claim 9 A failure prediction device for a cloud data center according to claim 8, wherein the total loss for training the student model is expressed as the sum of a knowledge distillation loss that minimizes the difference between the predicted value of the student model and the predicted value of the teacher model, and a cross-entropy loss that minimizes the difference between the predicted value of the student model and a preset correct answer value. Claim 10 A method for predicting failures in a cloud data center, performed on a computing device having one or more processors and a memory for storing one or more programs executed by said one or more processors, comprising: a step of acquiring task-related data of a cloud data center; a step of preprocessing said acquired task-related data; and a step of inputting said preprocessed task-related data into one or more neural network models to predict the end state of each task, wherein said neural network models include a teacher model and a student model, and the step of predicting the end state of said task comprises: a step of inputting a sequence of said preprocessed task-related data into said teacher model to cause said teacher model to predict the end state of each task; and a step of, when said teacher model has finished learning, inputting said preprocessed task-related data sequences into said teacher model and said student model, respectively, to cause said student model to predict the end state of each task. Claim 11 A method for predicting failure in a cloud data center according to claim 10, wherein the task-related data includes one or more of an event type, timestamp, CPU utilization, memory utilization, cache memory utilization, average disk I / O time, disk usage, task priority, number of task resubmissions, and task scheduling delay for the said task. Claim 12 A method for predicting failures in a cloud data center according to claim 11, wherein the event type represents types of states according to the performance of the corresponding operation. Claim 13 A method for predicting failures in a cloud data center according to claim 12, wherein the preprocessing step excludes work-related data among the acquired work-related data in which the event type indicates an incomplete work life where it is difficult to determine whether the corresponding work has ended. Claim 14 A method for predicting failures in a cloud data center according to claim 13, wherein the preprocessing step excludes the acquired work-related data in which the event type indicates an incomplete work life, and then labels the termination status of the remaining work-related data by classifying it into two categories, success and failure, based on the event type of each work. Claim 15 delete Claim 16 A method for predicting failures in a cloud data center according to claim 10, wherein the teacher model and the student model are each composed of a Bi-LSTM (Bi-directional Long Short Term Memory) neural network architecture, and the number of LSTM units in the forward LSTM and backward LSTM layers of the student model is smaller than the number of LSTM units of the teacher model. Claim 17 A method for predicting failure in a cloud data center according to claim 16, wherein the neural network architecture of the Bi-LSTM (Bi-directional Long Short Term Memory) comprises: an input layer receiving a sequence of preprocessed task-related data; a forward LSTM layer generating a forward hidden state by forward encoding the sequence of task-related data; a backward LSTM layer generating a backward hidden state by backward encoding the sequence of task-related data; an output layer outputting a combined representation of the forward hidden state and the backward hidden state; and a classification layer predicting whether the termination state of each task is success or failure based on the combined representation using a probability distribution. Claim 18 A method for predicting failures in a cloud data center according to claim 17, wherein the total loss for training the student model is expressed as the sum of a knowledge distillation loss that minimizes the difference between the predicted value of the student model and the predicted value of the teacher model, and a cross-entropy loss that minimizes the difference between the predicted value of the student model and a preset correct answer value. Claim 19 A computer program stored on a non-transitory computer-readable storage medium, wherein the computer program comprises one or more instructions, and when the instructions are executed by a computing device having one or more processors, the computing device performs the steps of: acquiring job-related data from a cloud data center; preprocessing the acquired job-related data; and inputting the preprocessed job-related data into one or more neural network models to predict the end state of each job, wherein the neural network models include a teacher model and a student model, and the step of predicting the end state of the job includes the step of inputting a sequence of the preprocessed job-related data into the teacher model to cause the teacher model to predict the end state of each job; and, when the training of the teacher model is finished, inputting the sequence of the preprocessed job-related data into the teacher model and the student model, respectively, to cause the student model to predict the end state of each job.

Images