Offline task scheduling optimization method and device

Abstract

The present disclosure provides an offline task scheduling optimization method and device. The offline task scheduling optimization method includes: extracting offline tasks to be scheduled from a list of offline tasks to be scheduled; identifying resource demand constraints of offline tasks; selecting hosts that meet the resource demand constraints from a cluster to generate a host list; using a trained prediction model to predict the CPI index of each host in the host list to evaluate the impact on the online service of each host after the offline task is deployed; selecting the host with the least impact as the target host so as to deploy the offline task to the target host; monitoring the service quality of the target host at a predetermined period; judging whether the service quality is less than a preset threshold; if the service quality is not less than the preset threshold, judging whether the offline task is executed; if the offline task is executed, terminating the scheduling of the offline task.

Description

Technical Field

[0001] The present disclosure relates to the field of cloud computing, and in particular to an offline task scheduling optimization method and device. Background Art

[0002] With the development of cloud computing business, there are more and more types of cloud services, which can be mainly divided into online tasks and offline tasks. Online tasks usually have a long life cycle and are sensitive to latency, such as online web services, which generally reserve more resources to cope with sudden network requests; offline tasks have a short life cycle and are not sensitive to latency, such as big data batch processing jobs, which occupy more resources during operation. Different cloud computing tasks have different resource requirements. In order to further improve resource utilization while meeting user needs, cloud service providers usually adopt a hybrid deployment method to deploy online tasks and offline tasks on the same host. Summary of the invention

[0003] The inventors have found through research that most existing task scheduling strategies aim to minimize task response time, and they mainly have the following problems:

[0004] (1) Different cloud computing tasks consume different host resources. When deploying offline tasks on a host with existing online tasks, the existing scheduling strategy does not consider the heterogeneity of task resources. The evaluation of host performance is overall, lacking differentiated evaluation of offline tasks.

[0005] (2) When the demand for online tasks surges, resource preemption will occur, which can easily lead to task execution failure. At this time, interrupting the scheduled offline tasks may cause the host performance to degrade and cause the online service to crash.

[0006] Based on this, the present disclosure provides an offline task scheduling optimization solution to effectively solve the problems existing in the existing cloud computing task scheduling strategy, while improving the overall resource utilization of the cloud data center, effectively reducing the impact on the online service QoS (Quality of Service) in the hybrid deployment mode.

[0007] According to a first aspect of an embodiment of the present disclosure, a method for optimizing offline task scheduling is provided, comprising: extracting offline tasks to be scheduled from a list of offline tasks to be scheduled; identifying resource demand constraints of the offline tasks; selecting hosts that meet the resource demand constraints from a cluster to generate a host list; using a trained prediction model to predict the CPI index of each host in the host list to evaluate the degree of impact on the online service of each host after the offline task is deployed; selecting the host with the least impact as the target host so as to deploy the offline task to the target host; monitoring the service quality of the target host at a predetermined period; judging whether the service quality is less than a preset threshold; if the service quality is not less than the preset threshold, judging whether the offline task is executed and ended; if the offline task is executed and ended, the scheduling of the offline task is ended.

[0008] In some embodiments, if the offline task has not been completed, the step of monitoring the service quality of the target host at a predetermined period is repeated.

[0009] In some embodiments, if the service quality is less than a preset threshold, the offline task is interrupted; and the offline task is added to a list of offline tasks to be scheduled.

[0010] In some embodiments, monitoring the service quality of the target host includes: determining the service quality q of the target host based on the number s of online service requests successfully executed in the current monitoring period, the total number N of online service requests received in the current monitoring period, the number o of service requests successfully executed but timed out in the current monitoring period, the CPI index in the current monitoring period, and the difference △CPI between the CPI index of the next monitoring period and the CPI index of the current monitoring period.

[0011] In some embodiments, the target host's quality of service q is

[0012]

[0013] Among them, λ0, λ1, and λ2 are weight parameters.

[0014] In some embodiments, evaluating the impact of each host's online service after the offline task is deployed includes: obtaining performance parameters of each host, wherein the performance parameters of each host include the current first CPI indicator; performing data preprocessing on the performance parameters of each host to obtain a data preprocessing result for each host, wherein the data preprocessing result includes the average CPU occupancy c of each host's offline tasks, the average content usage size m of offline tasks, the number of offline tasks plus 1, the CPU usage plus c, and the memory usage size plus m; inputting each data preprocessing result into a trained prediction model to predict the second CPI indicator of each host after the offline task is deployed; and using the difference between the second CPI indicator and the first CPI indicator of each host as the impact of each host's online service after the offline task is deployed.

[0015] In some embodiments, a neural network model is trained to obtain a prediction model, wherein: all hybrid deployed host performance parameters in multiple continuous time windows in a cluster are collected; all hybrid deployed host performance parameters are processed to construct an input matrix and an output CPI indicator; and the GRU neural network is trained using the input matrix and the output CPI indicator to obtain a prediction model.

[0016] According to a second aspect of an embodiment of the present disclosure, an offline task scheduling optimization device is provided, comprising: a first processing module, extracting offline tasks to be scheduled from a list of offline tasks to be scheduled, identifying resource demand constraints of the offline tasks, selecting hosts that meet the resource demand constraints from cluster hosts to generate a host list, using a trained prediction model to predict the CPI index of each host in the host list in turn to evaluate the impact on the online service of each host after the offline task is deployed, and selecting the host with the least impact as the target host so as to deploy the offline task to the target host; a second processing module, configured to monitor the service quality of the target host at a predetermined period to determine whether the service quality is less than a preset threshold, if the service quality is not less than the preset threshold, determine whether the offline task is executed and terminate the scheduling of the offline task if the offline task is executed and terminate the scheduling of the offline task.

[0017] In some embodiments, the second processing module is configured to repeatedly monitor the service quality of the target host at a predetermined period if the offline task has not been completed.

[0018] In some embodiments, the second processing module is configured to interrupt the offline task and add the offline task to a list of offline tasks to be scheduled if the service quality is less than a preset threshold.

[0019] In some embodiments, the second processing module is configured to determine the service quality q of the target host based on the number s of online service requests successfully executed in the current monitoring cycle, the total number N of online service requests received in the current monitoring cycle, the number o of service requests successfully executed but timed out in the current monitoring cycle, the CPI index in the current monitoring cycle, and the difference △CPI between the CPI index of the next monitoring cycle and the CPI index of the current monitoring cycle.

[0020] In some embodiments, the target host's quality of service q is

[0021]

[0022] Among them, λ0, λ1, and λ2 are weight parameters.

[0023] In some embodiments, the first processing module is configured to obtain performance parameters of each host, wherein the performance parameters of each host include the current first CPI indicator, perform data preprocessing on the performance parameters of each host to obtain data preprocessing results for each host, wherein the data preprocessing results include the average CPU occupancy c of offline tasks of each host, the average content usage size m of offline tasks, the number of offline tasks plus 1, the CPU usage plus c, and the memory usage size plus m, input each data preprocessing result into a trained prediction model to predict the second CPI indicator of each host after the offline task is deployed, and use the difference between the second CPI indicator of each host and the first CPI indicator as the degree of impact on the online service of each host after the offline task is deployed.

[0024] In some embodiments, the above-mentioned device also includes: a third processing module, configured to train the neural network model to obtain a prediction model, wherein all hybrid deployed host performance parameters in multiple continuous time windows in the cluster are collected, and all hybrid deployed host performance parameters are processed to construct an input matrix and an output CPI indicator, and the input matrix and the output CPI indicator are used to train the GRU neural network to obtain a prediction model.

[0025] According to a third aspect of an embodiment of the present disclosure, an offline task scheduling optimization device is provided, comprising: a memory configured to store instructions; a processor coupled to the memory, the processor being configured to execute a method as described in any of the above embodiments based on the instructions stored in the memory.

[0026] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method involved in any of the above embodiments is implemented.

[0027] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings, which constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0029] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0030] Figure 1 is a flowchart of an offline task scheduling optimization method according to an embodiment of the present disclosure;

[0031] Figure 2 is a flowchart of an offline task scheduling optimization method according to another embodiment of the present disclosure;

[0032] Figure 3 is a flowchart of an offline task scheduling optimization method according to yet another embodiment of the present disclosure;

[0033] Figure 4 is a structural diagram of an offline task scheduling optimization device according to an embodiment of the present disclosure;

[0034] Figure 5 is a structural diagram of an offline task scheduling optimization device according to another embodiment of the present disclosure;

[0035] Figure 6 It is a structural diagram of an offline task scheduling optimization device according to another embodiment of the present disclosure.

[0036] It should be understood that the size of each part shown in the accompanying drawings is not drawn according to the actual proportional relationship. In addition, the same or similar reference numerals represent the same or similar components. DETAILED DESCRIPTION

[0037] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. The description of the exemplary embodiments is merely illustrative and is in no way intended to limit the present disclosure and its application or use. The present disclosure can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to make the present disclosure thorough and complete and to fully express the scope of the present disclosure to those skilled in the art. It should be noted that unless otherwise specifically stated, the relative arrangement of the components and steps, the composition of the materials, and the numerical values ​​set forth in these embodiments should be interpreted as being merely exemplary and not as limiting.

[0038] The words “include” or “comprising” and the like used in the present disclosure mean that the elements preceding the words include the elements listed after the words, and do not exclude the possibility of also including other elements.

[0039] All terms (including technical terms or scientific terms) used in the present disclosure have the same meanings as those understood by ordinary technicians in the field to which the present disclosure belongs, unless otherwise specifically defined. It should also be understood that terms defined in general dictionaries, such as general dictionaries, should be interpreted as having meanings consistent with their meanings in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined herein.

[0040] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0041] Figure 1 1 is a flow chart of an offline task scheduling optimization method according to an embodiment of the present disclosure. In some embodiments, the following offline task scheduling optimization method is executed by an offline task scheduling optimization device.

[0042] In step 101, an offline task to be scheduled is extracted from a list of offline tasks to be scheduled.

[0043] In step 102, resource requirement constraints of the offline task are identified.

[0044] In step 103, hosts that meet resource demand constraints are selected from the cluster to generate a host list.

[0045] In step 104, the trained prediction model is used to predict the CPI (Cycleper Instruction) index of each host in the host list to evaluate the impact on the online service of each host after the offline task is deployed.

[0046] In some embodiments, a prediction model is obtained by training a neural network model.

[0047] For example, the performance parameters of all hybrid deployed hosts in multiple continuous time windows in the cluster are collected. The host performance parameters include CPU (Central Processing Unit), memory, GPU (Graphics Processing Unit), CPU frequency, the number of current online tasks, the number of offline tasks, CPI, CPU usage, memory usage, network bandwidth, I / O throughput, etc., as well as the CPU usage and memory usage of each task in the most recent time window.

[0048] Next, all the performance parameters of the hybrid deployed hosts are processed to construct the input matrix and output CPI index. The GRU (Gate Recurrent Unit) neural network is trained using the input matrix and output CPI index to obtain a prediction model.

[0049] In some embodiments, the step of evaluating the impact on each host's online service after deploying the offline task is as follows: Figure 2 shown.

[0050] In step 201, the performance parameters of each host are obtained, wherein the performance parameters of each host include the current first CPI indicator CPI0, and also include CPU, memory, GPU, CPU frequency, the current number of online tasks, the number of offline tasks, CPU usage, memory usage size, network bandwidth, I / O throughput, etc., as well as the CPU usage and memory usage size of each task in the most recent time window.

[0051] In step 202, data preprocessing is performed on the performance parameters of each host to obtain data preprocessing results for each host, wherein the data preprocessing results include the average CPU occupancy c of offline tasks of each host, the average content usage size m of offline tasks, the number of offline tasks plus 1, the CPU usage plus c, the memory usage size plus m, etc.

[0052] In step 203, each data preprocessing result is input into a trained prediction model to predict a second CPI indicator CPI1 of each host after the offline task is deployed.

[0053] In step 204, the difference between the second CPI index CPI1 and the first CPI index CPI0 of each host is used as the impact on the online service of each host after the offline task is deployed. For example, the smaller the difference between CPI1 and CPI0, the smaller the impact.

[0054] return Figure 1 In step 105, the host with the least impact is selected as the target host, so as to deploy the offline task to the target host.

[0055] In step 106, the service quality of the target host is monitored at a predetermined period.

[0056] In some embodiments, the service quality q of the target host is determined based on the number s of online service requests successfully executed in the current monitoring period, the total number N of online service requests received in the current monitoring period, the number o of service requests successfully executed but timed out in the current monitoring period, the CPI index in the current monitoring period, and the difference △CPI between the CPI index of the next monitoring period and the CPI index of the current monitoring period.

[0057] For example, the target host's quality of service q is

[0058]

[0059] Among them, λ0, λ1, and λ2 are weight parameters.

[0060] return Figure 1 In step 107, it is determined whether the service quality is less than a preset threshold.

[0061] In step 108, if the service quality is not less than the preset threshold, it is determined whether the offline task is completed.

[0062] In step 109, if the offline task execution is completed, the scheduling of the offline task is terminated.

[0063] Figure 3 1 is a flow chart of an offline task scheduling optimization method according to another embodiment of the present disclosure. In some embodiments, the following offline task scheduling optimization method is executed by an offline task scheduling optimization device.

[0064] In step 301, an offline task to be scheduled is extracted from a list of offline tasks to be scheduled.

[0065] In step 302, resource requirement constraints of the offline task are identified.

[0066] In step 303, hosts that meet resource demand constraints are selected from the cluster to generate a host list.

[0067] In step 304, the trained prediction model is used to predict the CPI index of each host in the host list to evaluate the impact of deploying the offline task on the online service of each host.

[0068] In some embodiments, using Figure 2 The illustrated embodiment evaluates the impact on the online services of each host after the offline task is deployed.

[0069] In step 305, the host with the least impact is selected as the target host, so as to deploy the offline task to the target host.

[0070] In step 306, the service quality of the target host is monitored at a predetermined period.

[0071] In some embodiments, the service quality q of the target host is determined based on the number s of online service requests successfully executed in the current monitoring period, the total number N of online service requests received in the current monitoring period, the number o of service requests successfully executed but timed out in the current monitoring period, the CPI index in the current monitoring period, and the difference △CPI between the CPI index of the next monitoring period and the CPI index of the current monitoring period.

[0072] For example, the target host's quality of service q is

[0073]

[0074] Among them, λ0, λ1, and λ2 are weight parameters.

[0075] In step 307, it is determined whether the service quality is less than a preset threshold.

[0076] If the service quality is not less than the preset threshold, step 308 is executed; if the service quality is less than the preset threshold, step 310 is executed.

[0077] In step 308, it is determined whether the offline task is completed.

[0078] If the offline task is completed, execute step 309; if the offline task is not completed, return to step 306.

[0079] In step 309, the scheduling of the offline task is ended.

[0080] In step 310, the offline task is interrupted.

[0081] In step 311, the offline task is added to the list of offline tasks to be scheduled. Then step 309 is executed.

[0082] Figure 4 FIG. 1 is a schematic diagram of the structure of an offline task scheduling optimization device according to an embodiment of the present disclosure. Figure 4 As shown, the offline task scheduling optimization device includes a first processing module 41 and a second processing module 42 .

[0083] The first processing module 41 is configured to extract offline tasks to be scheduled from the offline task list to be scheduled, identify the resource demand constraints of the offline tasks, select hosts that meet the resource demand constraints from the cluster hosts to generate a host list, and use the trained prediction model to predict the CPI index of each host in the host list in turn to evaluate the impact on the online services of each host after the offline task is deployed, and select the host with the least impact as the target host so as to deploy the offline task to the target host.

[0084] The second processing module 42 is configured to monitor the service quality of the target host at a predetermined period, determine whether the service quality is less than a preset threshold, and if the service quality is not less than the preset threshold, determine whether the offline task is completed, and if the offline task is completed, terminate the scheduling of the offline task.

[0085] In some embodiments, the second processing module 42 is configured to repeatedly monitor the service quality of the target host at a predetermined period if the offline task has not been completed.

[0086] In some embodiments, if the service quality is less than a preset threshold, the second processing module 42 interrupts the offline task and adds the offline task to a list of offline tasks to be scheduled.

[0087] In some embodiments, the second processing module 42 determines the service quality q of the target host based on the number s of online service requests successfully executed in the current monitoring cycle, the total number N of online service requests received in the current monitoring cycle, the number o of service requests successfully executed but timed out in the current monitoring cycle, the CPI index in the current monitoring cycle, and the difference △CPI between the CPI index of the next monitoring cycle and the CPI index of the current monitoring cycle.

[0088] For example, the target host's quality of service q is

[0089]

[0090] Among them, λ0, λ1, and λ2 are weight parameters.

[0091] In some embodiments, the first processing module 41 is configured to obtain performance parameters of each host, wherein the performance parameters of each host include the current first CPI indicator, perform data preprocessing on the performance parameters of each host to obtain data preprocessing results for each host, wherein the data preprocessing results include the average CPU occupancy c of offline tasks of each host, the average content usage size m of offline tasks, the number of offline tasks plus 1, the CPU usage plus c, and the memory usage size plus m, input each data preprocessing result into a trained prediction model to predict the second CPI indicator of each host after the offline task is deployed, and use the difference between the second CPI indicator of each host and the first CPI indicator as the degree of impact on the online service of each host after the offline task is deployed.

[0092] Figure 5 It is a structural diagram of an offline task scheduling optimization device according to another embodiment of the present disclosure. Figure 5 and Figure 4 The difference is that in Figure 5 In the illustrated embodiment, the offline task scheduling optimization device further includes a third processing module 43 .

[0093] The third processing module 43 is configured to train the neural network model to obtain a prediction model, wherein all hybrid deployed host performance parameters in multiple continuous time windows in the cluster are collected, all hybrid deployed host performance parameters are processed to construct an input matrix and an output CPI indicator, and the input matrix and the output CPI indicator are used to train the GRU neural network to obtain a prediction model.

[0094] Figure 6 FIG. 1 is a schematic diagram of the structure of an offline task scheduling optimization device according to another embodiment of the present disclosure. Figure 6 As shown, the offline task scheduling optimization device includes a memory 61 and a processor 62 .

[0095] The memory 61 is used to store instructions. The processor 62 is coupled to the memory 61. The processor 62 is configured to execute the instructions stored in the memory to implement the following. Figure 1-3 The method of any one of the embodiments.

[0096] like Figure 6 As shown, the offline task scheduling optimization device also includes a communication interface 63 for information exchange with other devices. At the same time, the offline task scheduling optimization device also includes a bus 64, through which the processor 62, the communication interface 63, and the memory 61 communicate with each other.

[0097] The memory 61 may include a high-speed RAM (Random Access Memory) or a NVM (Non-Volatile Memory). For example, at least one disk memory. The memory 61 may also be a memory array. The memory 61 may also be divided into blocks, and the blocks may be combined into virtual volumes according to certain rules.

[0098] In addition, the processor 62 may be a central processing unit, or may be an ASIC (Application Specific Integrated Circuit), or may be configured to implement one or more integrated circuits of the embodiments of the present disclosure.

[0099] The present disclosure also provides a computer-readable storage medium. The computer-readable storage medium stores computer instructions, which are executed by a processor to implement the following Figure 1-3 The method of any one of the embodiments.

[0100] By implementing the above solution of the present disclosure, the following beneficial effects can be obtained:

[0101] 1. In the hybrid deployment mode, the present disclosure considers the heterogeneity of resource requirements of offline tasks when selecting a scheduling host, and performs differentiated evaluation of the host performance for offline tasks, so as to minimize the impact on the host's online service quality after the offline task is deployed.

[0102] 2. The present disclosure can realize the prediction of host performance. By setting a suitable time window and a small amount of historical data, the host performance can be quickly predicted.

[0103] 3. The present invention predicts whether it is necessary to interrupt the batch processing job by regularly monitoring the online service quality of the host, rather than stopping the offline task when the resources are about to run out. Compared with the traditional monitoring method, it can execute the interrupt command and recycle resources at least one monitoring cycle in advance, further ensuring the quality of online services.

[0104] In some embodiments, the above-mentioned functional modules can be implemented as a general-purpose processor, a programmable logic controller (PLC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components or any appropriate combination thereof for performing the functions described in the present disclosure.

[0105] So far, the embodiments of the present disclosure have been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solutions disclosed herein.

[0106] Although some specific embodiments of the present disclosure have been described in detail by way of examples, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. It should be understood by those skilled in the art that the above embodiments may be modified or some technical features may be replaced by equivalents without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. An offline task scheduling optimization method, comprising: Extracting offline tasks to be scheduled from the offline task list to be scheduled; Identifying resource requirement constraints of the offline task; Selecting hosts that meet the resource requirement constraint from the cluster to generate a host list; Using the trained prediction model, predict the CPI index of each host in the host list to evaluate the impact of deploying the offline task on the online service of each host; Selecting the least affected host as the target host, so as to deploy the offline task on the target host; Monitoring the service quality of the target host at a predetermined period; Determining whether the service quality is less than a preset threshold; If the service quality is not less than a preset threshold, determining whether the offline task is completed; If the offline task is completed, the scheduling of the offline task is terminated; Wherein, monitoring the service quality of the target host includes: Determine the service quality q of the target host according to the number s of online service requests successfully executed in the current monitoring period, the total number N of online service requests received in the current monitoring period, the number o of service requests successfully executed but timed out in the current monitoring period, the CPI index in the current monitoring period, and the difference ΔCPI between the CPI index of the next monitoring period and the CPI index of the current monitoring period; The service quality q of the target host is Among them, λ0, λ1, and λ2 are weight parameters.

2. The method according to claim 1, further comprising: If the offline task is not completed, the step of monitoring the service quality of the target host at a predetermined period is repeated.

3. The method according to claim 2, further comprising: If the service quality is less than a preset threshold, interrupting the offline task; The offline task is added to the list of offline tasks to be scheduled.

4. The method according to claim 1, wherein: The evaluation of the impact on each host's online services after deploying the offline task includes: Obtaining performance parameters of each host, wherein the performance parameters of each host include a current first CPI indicator; Perform data preprocessing on the performance parameters of each host to obtain the data preprocessing results of each host, where the data preprocessing results include the average CPU occupancy rate c of offline tasks, the average content usage size m of offline tasks, the number of offline tasks plus 1, the CPU usage rate plus c, and the memory usage size plus m of each host; Input each data preprocessing result into the trained prediction model to predict the second CPI index of each host after the offline task is deployed; The difference between the second CPI index and the first CPI index of each host is used as the impact degree on the online service of each host after the offline task is deployed.

5. The method according to any one of claims 1 to 4, further comprising: The neural network model is trained to obtain a prediction model, where: Collect the performance parameters of all hybrid deployed hosts in the cluster within multiple consecutive time windows; All hybrid deployed host performance parameters are processed to construct input matrices and output CPI indicators; The GRU neural network is trained using the input matrix and the output CPI indicator to obtain a prediction model.

6. An offline task scheduling optimization device, comprising: The first processing module is configured to extract the offline tasks to be scheduled from the offline task list to be scheduled, identify the resource demand constraint conditions of the offline tasks, select the host that meets the resource demand constraint conditions from the cluster hosts to generate a host list, use the trained prediction model to predict the CPI index of each host in the host list in turn, so as to evaluate the impact of the deployment of the offline tasks on the online services of each host, and select the host with the least impact as the target host, so as to deploy the offline task to the target host; The second processing module is configured to monitor the service quality of the target host at a predetermined period, determine whether the service quality is less than a preset threshold, and if the service quality is not less than the preset threshold, determine whether the offline task is completed, and if the offline task is completed, terminate the scheduling of the offline task, wherein the second processing module is configured to determine the service quality q of the target host according to the number s of online service requests successfully executed in the current monitoring period, the total number N of online service requests received in the current monitoring period, the number o of service requests successfully executed but timed out in the current monitoring period, the CPI index in the current monitoring period, and the difference ΔCPI between the CPI index of the next monitoring period and the CPI index of the current monitoring period, wherein the service quality q of the target host is Among them, λ0, λ1, and λ2 are weight parameters.

7. The device according to claim 6, wherein: The second processing module is configured to repeatedly perform the operation of monitoring the service quality of the target host at a predetermined period if the offline task has not been completed.

8. The device according to claim 6, wherein: The second processing module is configured to interrupt the offline task and add the offline task to a list of offline tasks to be scheduled if the service quality is less than a preset threshold.

9. The device according to claim 6, wherein: The first processing module is configured to obtain performance parameters of each host, wherein the performance parameters of each host include the current first CPI indicator, perform data preprocessing on the performance parameters of each host to obtain the data preprocessing result of each host, wherein the data preprocessing result includes the average CPU occupancy c of offline tasks of each host, the average content usage size m of offline tasks, the number of offline tasks plus 1, the CPU usage plus c, and the memory usage size plus m, input each data preprocessing result into a trained prediction model, predict the second CPI indicator of each host after the offline task is deployed, and use the difference between the second CPI indicator of each host and the first CPI indicator as the degree of impact on the online service of each host after the offline task is deployed.

10. The device according to any one of claims 6 to 9, further comprising: The third processing module is configured to train the neural network model to obtain a prediction model, wherein the performance parameters of all hybrid deployed hosts in multiple continuous time windows in the cluster are collected, and all hybrid deployed host performance parameters are processed to construct an input matrix and an output CPI indicator, and the input matrix and the output CPI indicator are used to train the GRU neural network to obtain a prediction model.

11. An offline task scheduling optimization device, comprising: a memory configured to store instructions; A processor is coupled to the memory, and the processor is configured to execute the method according to any one of claims 1 to 5 based on instructions stored in the memory.

12. A computer-readable storage medium, wherein: The computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images