Server resource intelligent scheduling method and system based on auxiliary decision making

Abstract

The invention relates to the technical field of server resource scheduling, and discloses a server resource intelligent scheduling method and system based on auxiliary decision making. The method comprises the following steps: setting an execution ending time point of a task in an abnormal execution category in task execution information as a reference time point; analyzing the change trend of the basic parameter set to extract inflection points; dividing the plurality of servers according to a load evaluation value generated by combining the inflection point and the time difference, forming a schedulable server set, selecting a server from the schedulable server set to be matched with the to-be-executed task, and completing task allocation; and calculating an adaptive score of the current task and the current server, and judging whether the current task is migrated to the server in the schedulable server set to be executed or not. According to the method, the server is divided by analyzing the change trend of the execution parameters of the abnormal task, so that the accuracy of server state evaluation is improved; whether migration is executed on the current task is judged through the adaptive score, and stable execution of the task is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of server resource scheduling technology, specifically relating to a server resource intelligent scheduling method and system based on assisted decision-making. Background Technology

[0002] With the development of cloud computing, big data and artificial intelligence technologies, servers, as the core infrastructure for data processing, storage and computing, are playing an increasingly prominent role in the modern information technology system. In order to ensure the stability and efficiency of various online services, enterprise applications and computing tasks, how to achieve reasonable resource scheduling and management of server clusters has become a key technical link to ensure service quality and improve operational efficiency.

[0003] Existing server resource scheduling technologies rely heavily on single, superficial metrics, such as task execution frequency, to assess server load. This approach fails to distinguish between the inherent complexity of tasks and resource consumption, leading to inaccurate scheduling decisions. Furthermore, existing technologies lack guarantees for task continuity at the scheduling execution level. When migrating running tasks from one server to another, seamless transitions in execution status are often impossible, resulting in task interruptions, performance drops, or the need for restarts. This impacts business continuity and user experience. Finally, existing technologies lack adaptability to dynamically changing business demands. End-user-driven business requests are inherently spontaneous and uncertain, making it impossible to match dynamic resource requirements in real time, resulting in resource waste and low overall system utilization.

[0004] In view of this, the present invention proposes a server resource intelligent scheduling method and system based on auxiliary decision-making. Summary of the Invention

[0005] The purpose of this invention is to provide a server resource intelligent scheduling method and system based on assisted decision-making, so as to solve the technical problems of frequent server resource switching, poor continuity of task execution, and decreased operating efficiency in the prior art.

[0006] To achieve the above objectives, embodiments of the present invention provide a server resource intelligent scheduling method based on assisted decision-making, comprising the following steps:

[0007] A server resource intelligent scheduling method based on decision support includes the following steps:

[0008] The load assessment value is dynamically generated for each of the multiple servers to divide the multiple servers into a schedulable server set. The generation of the load assessment value is based on the analysis of the execution parameters of abnormal tasks in the task execution information.

[0009] When there are tasks to be executed, a server is selected from the set of schedulable servers to match the tasks to be executed, so as to complete the task allocation;

[0010] For the current task being executed, calculate the compatibility score between the current task and its current server, and based on the compatibility score, determine whether the current task should continue to be executed on the current server or be migrated to a server in the set of schedulable servers.

[0011] The steps for determining whether the current task should continue execution or be migrated include: if the adaptation score is lower than a preset adaptation score threshold, then perform server migration and migrate the current task to a server in the set of schedulable servers; if the adaptation score is not lower than the adaptation score threshold, then repeatedly calculate the adaptation score at preset time intervals, and if the adaptation score is lower than the adaptation score threshold in any of the preset number of consecutive calculations, then perform server migration; otherwise, determine that the current task should continue to be executed on the current server.

[0012] Preferably, the step of generating a load assessment value for each of the plurality of servers includes: obtaining task execution information and classifying tasks into normal execution categories or abnormal execution categories based on the task completion status; setting the execution completion point of tasks in the abnormal execution category as the baseline time point;

[0013] Based on the total running time, cumulative energy consumption, and frequency of anomalies of abnormal tasks within a preset period before the baseline time point, a set of basic parameters is constructed. Based on the set of basic parameters, the inflection points in the change trends of each parameter in the set of basic parameters within the preset period are extracted, and the time difference between the reference time point and the baseline time point corresponding to the inflection point is calculated. Based on the inflection point and the time difference, a load assessment value is generated.

[0014] Preferably, the task execution information includes the task's server information, execution duration, runtime power consumption, and task completion status; the total running time is calculated based on the execution duration; the cumulative energy consumption is calculated based on the runtime power consumption; and the anomaly occurrence frequency is calculated based on the frequency of occurrence of anomaly execution categories.

[0015] Preferably, the step of generating the load assessment value further includes: constructing graphical data of the changes of each parameter over time based on the basic parameter set, and calculating the rate of change of the graphical data to determine the trend of change; when the rate of change changes from a positive value to a non-positive value, or from a negative value to a non-negative value, determining the inflection point in the trend of change; using the parameter change amplitude and time difference corresponding to the inflection point as input parameters, and calculating the load assessment value through a preset function.

[0016] Preferably, the step of dividing the multiple servers into a schedulable server set includes: setting an upper limit for load assessment; when the load assessment value of a server is lower than the upper limit for load assessment, classifying the corresponding server as schedulable; when the load assessment value of a server is not lower than the upper limit for load assessment, classifying the corresponding server as under load; and aggregating all servers in the schedulable state to form a schedulable server set.

[0017] Preferably, before aggregating all servers in a schedulable state to form a schedulable server set, the method further includes: obtaining the server temperature change value and power consumption change deviation of the servers in a schedulable state within a preset analysis period; if the server temperature change value exceeds a preset temperature threshold, or the power consumption change deviation exceeds a preset power consumption threshold, then the corresponding server is classified as a resource to be verified; otherwise, the corresponding server is aggregated into the schedulable server set.

[0018] Preferably, for servers classified as the resources to be verified, the method further includes: at a preset verification time point, obtaining the current temperature and current power consumption change deviation of the server; if the current temperature of the server is not higher than the temperature threshold and the current power consumption change deviation does not exceed the power consumption threshold, then the server is added to the set of schedulable servers.

[0019] A server resource intelligent scheduling system based on decision support includes the following modules:

[0020] The load assessment module is used to dynamically generate load assessment values ​​for each of the multiple servers and divide the multiple servers into a schedulable server set based on the load assessment values. The generation of load assessment values ​​is based on the analysis of the execution parameters of abnormal tasks in the task execution information.

[0021] The task allocation module is used to select a server from the set of schedulable servers to match the task when there is a task to be executed;

[0022] The task migration decision module is used to calculate the compatibility score between the currently executing task and the current server, and based on the compatibility score, determine whether the current task should continue to be executed or migrate to a server in the set of schedulable servers.

[0023] Preferably, the task migration decision module determines whether the current task should continue execution or be migrated by: if the adaptation score is lower than a preset adaptation score threshold, then perform server migration and migrate the current task to a server in the set of schedulable servers; if the adaptation score is not lower than the adaptation score threshold, then repeatedly calculate the adaptation score at preset time intervals, and if the adaptation score is lower than the adaptation score threshold in any of the preset number of consecutive calculations, then perform server migration; otherwise, determine that the current task should continue to be executed on the current server.

[0024] Preferably, the step of the load assessment module generating a load assessment value for each of the plurality of servers includes: obtaining task execution information and classifying tasks into normal execution category or abnormal execution category according to the task completion status; setting the execution completion point of tasks in the abnormal execution category as the base time point;

[0025] Based on the total running time, cumulative energy consumption, and frequency of anomalies of abnormal tasks within a preset period before the baseline time point, a set of basic parameters is constructed. Based on the set of basic parameters, the inflection points in the change trends of each parameter in the set of basic parameters within the preset period are extracted, and the time difference between the reference time point and the baseline time point corresponding to the inflection point is calculated. Based on the inflection point and the time difference, a load assessment value is generated.

[0026] Beneficial effects

[0027] This invention analyzes the changing trends of a set of basic parameters, consisting of total running time, cumulative energy consumption, and frequency of anomalies, within a preset period before the end point of task execution in the abnormal execution category (i.e., the baseline time point). This allows for the extraction of inflection points, and the generation of load assessment values ​​by combining the time difference between the inflection points and the baseline time point. Based on the changing trends of execution parameters before anomalies occur, this invention makes predictions, improves the accuracy of server status assessment, and divides multiple servers based on load assessment values, providing a reliable basis for task allocation.

[0028] After forming a set of schedulable servers based on load assessment values, this invention not only assigns tasks to tasks to be executed, but also calculates the compatibility score between the current task and its current server to decide whether the current task should continue to be executed or be migrated. Through the mechanism of combining task allocation and task migration decision, current tasks with poor compatibility scores are migrated to servers in the set of schedulable servers, avoiding task execution interruption due to changes in server status, and ensuring the stability and success rate of task execution.

[0029] This invention performs a secondary verification on servers that have entered the schedulable state. It verifies whether the deviation between the server's temperature change value and power consumption change value meets the preset temperature threshold and power consumption threshold. Servers that do not meet the conditions are subject to delayed review, forming the final set of schedulable servers. The multi-level filtering mechanism, which combines load assessment based on historical data, secondary verification based on real-time monitoring, and delayed review, can exclude servers with abnormal physical conditions, enhance the reliability of the schedulable server set, avoid assigning tasks to servers with potential faults, and improve the robustness of the entire intelligent server resource scheduling system. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method provided by the present invention;

[0031] Figure 2 This is a flowchart of the method for generating load assessment values ​​according to the present invention;

[0032] Figure 3 This is a flowchart of the method for forming a schedulable server set according to the present invention. Detailed Implementation

[0033] In the following description, certain specific details are set forth in order to provide a thorough understanding of various embodiments of the invention. However, those skilled in the art, upon reading this disclosure, will understand that the invention can be practiced without many of these details.

[0034] Example 1

[0035] Please refer to Figures 1-3 This embodiment provides a server resource intelligent scheduling method based on assisted decision-making, aiming to improve the overall resource utilization efficiency and operational stability of the data center through forward-looking and adaptive intelligent scheduling of server resources. Specifically, it includes the following steps:

[0036] Acquiring task execution information and classifying tasks into normal or abnormal execution categories based on their completion status includes: acquiring task execution information for each running task in real-time or periodically from multiple servers within the data center. This task execution information is structured data records, containing not only information about the server to which the task belongs, task identification parameters, time attributes such as task termination time and execution duration, execution status parameters such as running, suspended, waiting for input / output, and terminated, resource consumption parameters such as runtime power consumption, a list of specific graphics processors or accelerator cards that the task depends on, and logs or error reports from program output when the task terminates. After acquiring this information, each running task is classified into a normal or abnormal execution category based on its completion status. Tasks that complete execution according to predetermined logic and exit normally with a completion status of "normally completed" are classified into the normal execution category. Tasks that terminate unexpectedly due to resource exhaustion, program errors, hardware failures, or other unforeseen reasons with completion statuses such as "timeout," "memory overflow," or "segmentation fault" are classified into the abnormal execution category.

[0037] Furthermore, tasks that are interrupted during execution due to unexpected reasons and are classified as abnormal execution tasks are defined as abnormal tasks. For in-depth source tracing analysis, the associated task name, server number, and characteristic indicators are recorded for each abnormal task. The characteristic indicators are a set of key performance snapshots within a period of time before the abnormality occurred. In this embodiment, the key performance snapshots include CPU utilization, memory usage, disk I / O wait time, and specific error codes recorded before the task terminates. This provides data basis for subsequent identification of potential incompatibility issues between specific tasks and specific server hardware combinations. The baseline time point of the abnormal task, i.e., the task termination time, is set as the baseline time point. The baseline time point is a specific moment that serves as the end point of the time axis during retrospective analysis. The entire time span of the abnormal task on the server from the start of execution to its termination is marked as the abnormal time period.

[0038] Furthermore, a retrospective preset time period is defined with the baseline time point as the endpoint. The duration of this time period can be set to 1 to 2 hours depending on the business characteristics. The retrospective preset time period is used to collect and analyze behavioral data before the anomaly occurs, capturing the load accumulation and evolution process before the anomaly occurs. Within this retrospective preset time period, for each server with an abnormal task, the following indicators are statistically analyzed: the total running time is calculated by summing the execution times of all tasks on the server; the cumulative energy consumption is calculated by integrating the time based on the runtime power consumption of each task; and the anomaly occurrence frequency is calculated by statistically analyzing the frequency of tasks classified as abnormal execution categories within this time period. The anomaly occurrence frequency is an indicator for measuring the stability of server operation. The total running time, cumulative energy consumption, and anomaly occurrence frequency are collectively constructed into a basic parameter set. The basic parameter set is the core indicator for quantifying the server load status, which can quantify the overall load status of the server before the anomaly occurs from three dimensions: time, energy consumption, and stability.

[0039] Furthermore, to gain insight into the deterioration trend of server status from dynamic changes, a time series data graph is constructed based on the set of basic parameters, with time as the horizontal axis and the values ​​of each basic parameter as the vertical axis. By performing differential calculations on this time series data graph to obtain its first derivative, the rate of change of each basic parameter is determined, and the trend of change is judged based on the rate of change of each basic parameter. When the value of the rate of change changes from a positive value to a non-positive value, or from a negative value to a non-negative value, an inflection point is determined to have occurred in the trend. This inflection point corresponds to a local peak or trough in resource consumption or abnormal frequency in a physical sense, and can serve as a key indicator of a change in server status. The time point corresponding to the inflection point is recorded as a reference time point, which marks the critical moment when a change in server status may occur. The reference time point and the baseline time point are converted into a unified time format. In this embodiment, they are uniformly converted to UNIX timestamps. Then, the time difference between the two is calculated to determine the parameter change amplitude corresponding to the inflection point. The parameter change amplitude and the time difference are processed as inputs to output a load assessment value.

[0040] The functional relationship for the output load assessment value is as follows:

[0041]

[0042] In the formula, This represents the load assessment value, which quantifies the potential risk of the server. The larger the value, the more drastic the load change occurred closer to the point of an anomaly, and the higher the risk. This represents the weighting coefficient, which is a preset constant used to adjust the dimensions and sensitivity of the load assessment value; It represents the magnitude of parameter change, which means the degree of drastic change in basic parameters such as cumulative energy consumption and abnormal frequency at a reference time point. It can be taken as the change near that point or the parameter value itself. This represents the baseline time point, which signifies the termination time of the abnormal task and serves as the endpoint for risk assessment. This indicates a reference time point, which means the time corresponding to the trend inflection point in the basic parameter time series plot.

[0043] Specifically, the function formula for outputting the load assessment value takes the parameter change magnitude as the positive input and the time difference as the negative input. By dividing the parameter change magnitude by the time difference and multiplying it by a preset weighting coefficient, it ensures that severe load shocks occurring near the fault point can produce higher load assessment values. The higher the load assessment value, the higher the risk.

[0044] Furthermore, a load assessment upper limit is set. This limit can be pre-configured by operations personnel based on historical data and acceptable risk levels. After generating the load assessment values ​​for each server, the load assessment upper limit is compared with the load assessment values ​​to determine the following: when the load assessment value of a server is lower than the load assessment upper limit, it indicates that the server has been running stably recently and has low potential risk. At this time, it is safe to accept new task assignments and is classified as schedulable. When the load assessment value is not lower than the load assessment upper limit, it indicates that the server has potential overload or instability risks. At this time, it is not advisable to accept new tasks immediately and is classified as under load.

[0045] Furthermore, before aggregating all servers in a schedulable state into a schedulable server set, a secondary verification is initiated for servers initially classified as schedulable to ensure their physical stability and eliminate potential hardware instability risks. This involves obtaining the server's physical indicators collected within a preset analysis period over the past 24 hours. The returned information is a set of data reflecting the server's physical operating status, specifically including server temperature changes and power consumption deviations. The returned information contains internally set temperature and power consumption thresholds. If the server's detected temperature change, fluctuation amplitude, or rate of change exceeds the preset threshold, the verification will be terminated. If the power consumption deviation or fluctuation detected by the server exceeds the preset power consumption deviation threshold, the server is considered to have potential physical hazards such as poor heat dissipation or unstable power supply. Its status is then updated to "Resources to be Verified." Resources to be verified are intermediate states of servers that need to enter a dynamic observation period to further confirm their stability. Otherwise, if all physical indicators are within the threshold range, the server is considered to be stable and is officially added to the schedulable server set. The schedulable server set is a dynamically updated list that includes all servers that have passed load assessment and secondary verification and have been finally confirmed to be stable and available for allocating new tasks for subsequent task allocation.

[0046] Specifically, for servers classified as resources to be verified, a dynamic observation period is initiated. During this period, subsequent temperature changes of the server are continuously recorded at a higher frequency to determine whether the risk is instantaneous or persistent. To avoid misjudgments caused by instantaneous fluctuations, the high-frequency collected server temperature readings are used as subsequent temperature change values ​​and substituted into the delay calculation rule. The next inspection time is dynamically adjusted according to the severity of the temperature change to obtain the delay time. The delay calculation rule is set to be proportional to the magnitude of the temperature change value. In this embodiment, the temperature change value is multiplied by a preset time coefficient. The more severe the temperature change, the longer the calculated delay time.

[0047] The relationship of the delayed computation rule function is as follows:

[0048]

[0049] In the formula, This indicates the delay time, which means how long it will take to wait before the next temperature check is performed, starting from the current moment. This time is proportional to the temperature fluctuation, achieving the effect of "the greater the fluctuation, the longer the observation period" for delayed judgment. The time coefficient is a preset constant used to convert temperature changes into time units, which determines the sensitivity to delay. This indicates the subsequent temperature change value, which represents the real-time change in server temperature during the dynamic observation period.

[0050] The current sampling time is added to the calculated delay time to obtain the verification time point. The verification time point serves as the specific time point for final state confirmation. When the verification time point arrives, the current temperature of the server is obtained again. If the current temperature of the server is higher than the temperature threshold, it is determined that the server has a continuous physical risk, and the server should not be assigned any tasks. Its status is updated to unschedulable. If the current temperature of the server is not higher than the temperature threshold, the server is reclassified into a schedulable state and added to the set of schedulable servers.

[0051] Furthermore, when there is a task to be executed, the rated requirement parameters of the task to be executed are obtained. In this embodiment, the rated requirement parameters are preferably the minimum number of CPU cores and memory capacity required for the task to run, network resources such as network bandwidth, and specific hardware dependencies. The task allocation module selects a server from the set of schedulable servers to match the task to be executed in order to complete the task allocation.

[0052] Furthermore, for the currently executing task, its operational status is periodically evaluated. Based on the scoring rules, a compatibility score is calculated between the current task and its current server. This compatibility score reflects the degree of matching between the current task and the server in terms of resources and historical performance. The scoring rules comprehensively consider multiple dimensions, comparing the current task's rated requirements with the server's current CPU idle rate, available memory size, and other real-time available resources to generate a resource matching score. This resource matching score measures the fit between the task's resource requirements and the server's currently available resources. Efficiency data such as the average completion time and historical anomaly rate for similar tasks in the server's history are retrieved to generate a historical performance score. This historical performance score measures the server's historical efficiency and stability when handling similar tasks. Finally, the resource matching score and the historical performance score are weighted and summed to obtain the final compatibility score.

[0053] The functional relationship of the scoring rules is:

[0054]

[0055] In the formula, The matching score represents the final score that comprehensively evaluates the compatibility between the task and the server. The higher the score, the better the compatibility. This represents the weighting coefficients, which are preset constants used to adjust the proportion of resource matching score and historical performance score in the final score, respectively. ; This represents the resource matching score, which is calculated based on the matching between the task's rated requirements and the server's real-time available resources. The score represents historical performance and is calculated based on the server's efficiency, completion time, stability, or anomaly rate data when performing similar tasks in the past.

[0056] Furthermore, an adaptation score threshold is set. If the calculated adaptation score is lower than the threshold, a server migration operation is performed, dynamically and without interruption migrating the current task to another server with a better adaptation score in the set of schedulable servers. If the adaptation score is not lower than the threshold, the adaptation score is repeatedly calculated at preset time intervals. If the adaptation score is not lower than the threshold for a preset number of consecutive calculations, the current task is determined to continue to be executed on the current server. This decision-making mechanism with a lag effect effectively avoids frequent task migrations caused by small fluctuations in the score around the threshold. If, during the preset number of consecutive calculations, the adaptation score is lower than the threshold at any time, server migration is performed immediately.

[0057] This embodiment forms a closed-loop intelligent server resource scheduling method by collecting the status of running tasks in real time, dynamically evaluating server load and physical status, and making scheduling decisions based on adaptation scores. This method effectively reduces the risk of server overload caused by untimely monitoring, avoids performance fluctuations and efficiency degradation of running tasks during unnecessary migration processes, and significantly improves the reuse rate and intelligent scheduling efficiency of server resources.

[0058] This embodiment also provides a server resource intelligent scheduling terminal based on auxiliary decision-making. The intelligent scheduling terminal includes a processor and a memory connected to the processor. The memory stores a computer program. When the computer program is executed by the processor, it implements the server resource intelligent scheduling method based on auxiliary decision-making.

[0059] Example 2

[0060] This embodiment provides a server resource intelligent scheduling system based on assisted decision-making. This system analyzes abnormal patterns in historical task execution information to assess server load health and, combined with real-time adaptability assessments of currently executing tasks, achieves dynamic and intelligent scheduling of server resources. This improves the operational stability and resource utilization efficiency of the entire server cluster. Specifically, this system can be deployed on data center management nodes, cloud platform control servers, or any terminal device capable of centrally monitoring and managing multiple servers. The system includes the following modules:

[0061] The load assessment module is configured to dynamically generate load assessment values ​​for each of multiple servers and divide the servers into a schedulable server set based on these values. This module is crucial for proactive risk warning and resource selection. The load assessment module first acquires task execution information, including the server to which the task belongs, execution duration, runtime power consumption, and task completion status. Based on the completion status, historical tasks are categorized into normal or abnormal execution categories. For tasks in the abnormal execution category, their termination time is set as a baseline time point. A basic parameter set is constructed based on the total runtime, cumulative energy consumption, and frequency of abnormal occurrences of abnormal tasks within a preset time period prior to the baseline time point. The total runtime is calculated based on the execution duration; the cumulative energy consumption is calculated based on runtime power consumption; and the frequency of abnormal occurrences is calculated based on the frequency of occurrences of the abnormal execution category.

[0062] To uncover early signs of server performance deterioration from historical data, the load assessment module constructs graphical data of how each parameter changes over time based on a set of fundamental parameters. It calculates the rate of change of this data to determine its trend. When the rate of change changes from positive to negative, or from negative to non-negative, an inflection point in this trend is identified. This inflection point signifies a shift in server load-related parameters from a stable or linearly increasing state to a state of accelerated deterioration or leveling off, serving as a key indicator of potential problems. The load assessment module records the reference time point corresponding to the inflection point and calculates the time difference between it and the baseline time point. Using the parameter change magnitude and time difference corresponding to the inflection point as input parameters, the module calculates the load assessment value. This value comprehensively reflects the characteristics of server load changes and the early warning amount before task execution anomalies occur.

[0063] After generating the load assessment value, the load assessment module is further responsible for forming a set of schedulable servers. By setting an upper limit for the load assessment, when the load assessment value of a server is lower than the upper limit, the server is classified as schedulable; otherwise, it is classified as under load. To further ensure the reliability of resources, the load assessment module can also obtain the server temperature change value and power consumption change deviation of the servers in the schedulable state within a preset analysis period. If the server temperature change value exceeds the preset temperature threshold, or the power consumption change deviation exceeds the preset power consumption threshold, the corresponding server is classified as a resource to be verified and is temporarily not included in the set of schedulable servers. For servers classified as resources to be verified, the module will re-obtain their current temperature and current power consumption change deviation at a preset verification time. If the indicators return to normal, the server is added to the set of schedulable servers. All servers in the schedulable state that have passed the second verification are gathered to form the final set of schedulable servers.

[0064] The task allocation module is configured to select a server from the set of schedulable servers to match the task when there is a task to be executed, so as to complete the task allocation. When a new task arrives in the system, the set of schedulable servers, which is updated in real time by the load assessment module, is queried, and one or more of the most suitable servers are selected from the set of schedulable servers. The selection criteria may include, but are not limited to, the current remaining capacity of the server, the degree of matching between the server hardware configuration and the rated requirement parameters of the task, or the implementation of quota policies. After the selection is completed, the set of schedulable servers sends the scheduling list to the target server to complete the initial deployment of the task.

[0065] The task migration decision module is configured to continuously monitor and dynamically adjust the currently executing task. Its core function is to calculate the compatibility score between the current task and its current server, and decide whether to continue the task or perform server migration based on the score. The compatibility score is a comprehensive indicator that takes into account factors such as the task's resource consumption pattern, execution efficiency, and the server's current real-time load status. The set of schedulable servers periodically calculates the compatibility score for each current task. If the compatibility score is lower than a preset compatibility score threshold, it indicates that the task is not well matched with the current server, and there may be resource contention or performance bottlenecks. At this time, server migration will be triggered, and the current task will be migrated to another more suitable server in the set of schedulable servers maintained by the load assessment module.

[0066] To avoid frequent task migrations caused by instantaneous fluctuations, the task migration decision module also includes a decision mechanism with a lag effect. If the adaptation score is not lower than the adaptation score threshold, it will not be immediately determined to be in good condition. Instead, the adaptation score will be repeatedly calculated at a preset time interval. Only when the adaptation score is not lower than the adaptation score threshold in a preset number of consecutive calculations will it be finally determined that the current task will continue to be executed on the current server. Conversely, if the adaptation score is lower than the adaptation score threshold in any calculation during this continuous observation period, the server migration will also be performed.

[0067] Through the collaborative work of the load assessment module, task allocation module, and task migration decision module, this embodiment can construct a closed loop for full lifecycle server resource scheduling. This not only avoids assigning tasks to servers with potential risks, but also dynamically adjusts current tasks to ensure that tasks are always executed in the optimal environment. This improves the overall service quality and operational stability of the data center or server cluster and is suitable for handling computing tasks with high concurrency, long cycles, or large fluctuations in resource requirements.

[0068] Finally, it should be noted that the above examples are merely some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments and many variations are possible. All variations that can be directly derived or conceived by those skilled in the art from the disclosure of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A server resource intelligent scheduling method based on assisted decision-making, characterized in that, Includes the following steps: The load assessment value is dynamically generated for each of the multiple servers to divide the multiple servers into a schedulable server set. The generation of the load assessment value is based on the analysis of the execution parameters of abnormal tasks in the task execution information. When there are tasks to be executed, a server is selected from the set of schedulable servers to match the tasks to be executed, so as to complete the task allocation; For the current task being executed, calculate the compatibility score between the current task and its current server, and based on the compatibility score, determine whether the current task should continue to be executed on the current server or be migrated to a server in the set of schedulable servers. The steps for determining whether the current task should continue execution or be migrated include: if the adaptation score is lower than a preset adaptation score threshold, then perform server migration and migrate the current task to a server in the set of schedulable servers; if the adaptation score is not lower than the adaptation score threshold, then repeatedly calculate the adaptation score at preset time intervals, and if the adaptation score is lower than the adaptation score threshold in any of the preset number of consecutive calculations, then perform server migration; otherwise, determine that the current task should continue to be executed on the current server.

2. The intelligent server resource scheduling method based on assisted decision-making according to claim 1, characterized in that, The step of generating a load assessment value for each of the plurality of servers includes: obtaining task execution information and classifying tasks into normal execution category or abnormal execution category according to the task completion status; setting the execution completion point of tasks in the abnormal execution category as the base time point; Based on the total running time, cumulative energy consumption, and frequency of anomalies of abnormal tasks within a preset period before the baseline time point, a set of basic parameters is constructed. Based on the set of basic parameters, the inflection points in the change trends of each parameter in the set of basic parameters within the preset period are extracted, and the time difference between the reference time point and the baseline time point corresponding to the inflection point is calculated. Based on the inflection point and the time difference, a load assessment value is generated.

3. The intelligent server resource scheduling method based on assisted decision-making according to claim 2, characterized in that, The task execution information includes the task's owner server information, execution duration, runtime power consumption, and task completion status; Total runtime is calculated based on execution duration; cumulative energy consumption is calculated based on runtime power consumption; and the frequency of exception occurrences is calculated based on the frequency of occurrence of exception execution categories.

4. The intelligent server resource scheduling method based on assisted decision-making according to claim 2, characterized in that, The step of generating the load assessment value further includes: constructing graphical data of the changes of each parameter over time based on the basic parameter set, and calculating the rate of change of the graphical data to determine the trend of change; when the rate of change changes from a positive value to a non-positive value, or from a negative value to a non-negative value, determining the inflection point in the trend of change; using the parameter change amplitude and time difference corresponding to the inflection point as input parameters, and calculating the load assessment value through a preset function.

5. The intelligent server resource scheduling method based on assisted decision-making according to claim 1, characterized in that, The steps of dividing the multiple servers and forming a schedulable server set include: setting an upper limit for load assessment; when the load assessment value of a server is lower than the upper limit for load assessment, classifying the corresponding server as schedulable; when the load assessment value of a server is not lower than the upper limit for load assessment, classifying the corresponding server as under load; and aggregating all servers in the schedulable state to form a schedulable server set.

6. The intelligent server resource scheduling method based on assisted decision-making according to claim 5, characterized in that, Before aggregating all servers in a schedulable state into a schedulable server set, the method further includes: obtaining the server temperature change value and power consumption change deviation of the servers in a schedulable state within a preset analysis period; if the server temperature change value exceeds a preset temperature threshold, or the power consumption change deviation exceeds a preset power consumption threshold, then the corresponding server is classified as a resource to be verified; otherwise, the corresponding server is aggregated into the schedulable server set.

7. The intelligent server resource scheduling method based on assisted decision-making according to claim 6, characterized in that, For servers classified as the resources to be verified, the method further includes: at a preset verification time point, obtaining the current temperature and current power consumption change deviation of the server; if the current temperature of the server is not higher than the temperature threshold and the current power consumption change deviation does not exceed the power consumption threshold, then the server is added to the set of schedulable servers.

8. A server resource intelligent scheduling system based on assisted decision-making, characterized in that, Includes the following modules: The load assessment module is used to dynamically generate load assessment values ​​for each of the multiple servers and divide the multiple servers into a schedulable server set based on the load assessment values. The generation of load assessment values ​​is based on the analysis of the execution parameters of abnormal tasks in the task execution information. The task allocation module is used to select a server from the set of schedulable servers to match the task when there is a task to be executed; The task migration decision module is used to calculate the compatibility score between the currently executing task and the current server, and based on the compatibility score, determine whether the current task should continue to be executed or migrate to a server in the set of schedulable servers.

9. The intelligent server resource scheduling system based on assisted decision-making according to claim 8, characterized in that, The steps of the task migration decision module to determine whether the current task should continue to be executed or migrate include: if the adaptation score is lower than a preset adaptation score threshold, then perform server migration and migrate the current task to a server in the set of schedulable servers; if the adaptation score is not lower than the adaptation score threshold, then repeatedly calculate the adaptation score at preset time intervals, and if the adaptation score is lower than the adaptation score threshold in any of the preset number of consecutive calculations, then perform server migration; otherwise, determine that the current task should continue to be executed on the current server.

10. The server resource intelligent scheduling system based on assisted decision-making according to claim 8, characterized in that, The steps of the load assessment module to generate a load assessment value for each of the plurality of servers include: obtaining task execution information and classifying tasks into normal execution category or abnormal execution category according to the task completion status; setting the execution completion point of tasks in the abnormal execution category as the base time point; Based on the total running time, cumulative energy consumption, and frequency of anomalies of abnormal tasks within a preset period before the baseline time point, a set of basic parameters is constructed. Based on the set of basic parameters, the inflection points in the change trends of each parameter in the set of basic parameters within the preset period are extracted, and the time difference between the reference time point and the baseline time point corresponding to the inflection point is calculated. Based on the inflection point and the time difference, a load assessment value is generated.

Images