A heterogeneous computing power collaborative system and method based on energy efficiency optimization

Abstract

This invention discloses a heterogeneous computing power collaboration system and method based on energy efficiency optimization, relating to the field of heterogeneous computing power collaboration technology. The method includes analyzing the abnormal status of computing power resources in a regional resource pool within the current period; evaluating the migration benefits of other resource pools to the abnormal resource pool; analyzing the energy efficiency gains of migration tasks between the abnormal resource pool and the target resource pool under different data volumes; obtaining the target tasks to be migrated from the abnormal resource pool to the target resource pool based on the task migration data range between the abnormal resource pool and the target resource pool; migrating the target tasks in the abnormal resource pool to the target resource pool; and controlling the target resource pool to collaboratively complete the tasks in the abnormal resource pool. This enables regional resource pools with heterogeneous computing power in different areas of the cloud platform to work effectively together, and also significantly reduces the energy consumed by the regional resource pools in completing tasks, achieving true optimization of energy efficiency.

Description

Technical Field

[0001] This invention relates to the field of heterogeneous computing power collaboration technology, specifically a heterogeneous computing power collaboration system and method based on energy efficiency optimization. Background Technology

[0002] No single computing architecture is universally applicable during the completion of computational tasks. Therefore, it is necessary to combine computing units from different architectures to form powerful integrated computing power. Furthermore, by optimizing energy efficiency and coordinating the management of heterogeneous computing power, electricity costs can be directly reduced, the return on investment in computing power construction can be improved, more services can be supported with less energy, and more powerful heterogeneous computing power can be deployed within the same architectural space and power consumption constraints to handle more complex tasks. This also directly reduces unnecessary energy consumption, lowers carbon emissions, and protects the environment.

[0003] Currently, to ensure data security and optimize costs, cloud platforms typically build resource pools in different regions. These resource pools have different computing power structures and resources. When processing data, these resource pools generally allocate corresponding computing power resources based on the actual situation of the resource pool to complete the task. However, due to the uncertainty of resource status in different regions and the need to migrate data between different resource pools during task completion, it is easy to cause the task migration to fail to reach a matching resource pool. These problems can lead to abnormalities in the heterogeneous computing power collaboration within the resource pool, which not only fails to reduce the energy consumption of the resource pool during task processing but may even affect task processing. Summary of the Invention

[0004] The purpose of this invention is to provide a heterogeneous computing power collaborative system and method based on energy efficiency optimization, so as to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a heterogeneous computing power collaboration method based on energy efficiency optimization, the method comprising: Step S1: Monitor the regional resource pool in the current period, obtain the monitoring record of the regional resource pool, obtain the historical monitoring record of the regional resource pool, analyze the abnormal status of the computing power resources of the regional resource pool in the current period, and obtain the abnormal resource pool. Step S2: Obtain historical migration network records between the abnormal resource pool and other resource pools, and combine them with the monitoring records of other resource pools to evaluate the migration benefits of other resource pools to the abnormal resource pool, and obtain the target resource pool. Step S3: Obtain energy efficiency setting data from the platform and combine it with the monitoring records of the abnormal resource pool to generate the target energy efficiency benefit threshold between the abnormal resource pool and the target resource pool. Obtain historical task migration records between the abnormal resource pool and the target resource pool. Obtain energy efficiency data of the abnormal resource pool and the target resource pool. Analyze the energy efficiency benefit brought by the migration tasks between the abnormal resource pool and the target resource pool under different data volumes. Obtain the task migration data range between the abnormal resource pool and the candidate resource pool. Step S4: Obtain the task data of pending tasks in the abnormal resource pool. Based on the task migration data range between the abnormal resource pool and the target resource pool, obtain the target tasks to be migrated from the abnormal resource pool to the target resource pool. Migrate the target tasks in the abnormal resource pool to the target resource pool and control the target resource pool to collaboratively complete the tasks in the abnormal resource pool.

[0006] Furthermore, step S1 includes: Step S11: Monitor the regional resource pools deployed in different regions within the current period to obtain the monitoring records of the regional resource pools within the current period; Step S12: Obtain data on various monitoring indicators of the regional resource pool from the monitoring records; Obtain all historical monitoring records of the regional resource pool, and extract the mean and standard deviation of the average values ​​of various monitoring indicators in the regional resource pool from each historical monitoring record; Obtain the j-th range Q of the a-th monitoring indicator in the regional resource pool. a j =[μ a -j×σ a ,μ a +j×σ a ], where μ a σ is the mean of the average values ​​of the a-th monitoring indicator in all historical monitoring records. a is the standard deviation of the average value of the a-th monitoring indicator in each historical monitoring record; Obtain the duration of the distance between historical monitoring records and the current period, and record it as the distance duration of the historical monitoring record. Sort the historical monitoring records according to the distance duration, and obtain the maximum distance duration t among all historical monitoring records. max Calculate the characteristic time value of the b-th historical monitoring record in each historical monitoring record. , t b The distance duration of the b-th historical monitoring record; Step S13: Obtain the sum T of the characteristic time values ​​of each historical monitoring record. sum Obtain the average value of the a-th monitoring indicator in the j-th range Q. a jFrom several historical monitoring records, obtain the sum T of characteristic time values ​​from these historical monitoring records. j sum Calculate the j-th range Q a j Feature proportion Set the feature proportion threshold L △ When L j ≥L △ And L j-1 <L △ Then determine the j-th range Q a j The target range for the a-th monitoring indicator in the regional resource pool during the current period; Obtain the target range of various monitoring indicators of the regional resource pool in the current period. If the maximum or minimum value of any monitoring indicator in the monitoring record is not within the target range, it is determined that there is an abnormal computing power resource in the regional resource pool in the current period, and the regional resource pool is recorded as an abnormal resource pool.

[0007] Furthermore, step S2 includes: Step S21: Remove abnormal resource pools from the resource pools of each region in the platform, and record the remaining resource pools as the target resource pools; Monitor the network status when migrating tasks between the abnormal resource pool and other resource pools to obtain historical migration network records between the abnormal resource pool and other resource pools. Set the unit duration and obtain the transmission rate set from the historical migration network records; Step S22: Obtain the preset maximum theoretical network transmission rate B between the abnormal resource pool and other resource pools, and calculate the bandwidth utilization F within the d-th unit of time in the historical migration network record. d ; Obtain the mean value F of the average bandwidth utilization in each historical migration network record; Step S23: Obtain the mean μ´ and standard deviation σ´ of the average network transmission rate within each unit time period in the transmission rate set of historical migration network records, and calculate the network comprehensive score β between the abnormal resource pool and other resource pools. If the network comprehensive score β is greater than the preset threshold, other resource pools will be marked. Step S24: Obtain the monitoring records of other marked resource pools, and calculate the abnormal risk value γ of the g-th monitoring indicator in other resource pools. g ; The order m of obtaining abnormal risk values g Calculate the risk value of the g-th monitoring indicator in other resource pools. , of which M sum This represents the total number of all other resource pools. Step S25: Obtain the risk values ​​of various monitoring indicators in other resource pools. When the risk values ​​of various monitoring indicators in other resource pools are all less than the preset risk threshold, it is determined that other resource pools are beneficial for migrating tasks from abnormal resource pools, and the marked other resource pools are recorded as the target resource pools of abnormal resource pools.

[0008] Furthermore, step S3 includes: Step S31: Obtain preset energy efficiency setting data from the platform, and obtain the standard values ​​E for energy efficiency benefit threshold, network latency, and energy efficiency difference from the energy efficiency setting data. △ W △ and U △ ; Calculate the target energy efficiency benefit threshold E between the abnormal resource pool and the target resource pool. ▽ ; Step S32: Obtain historical task migration records between the abnormal resource pool and the target resource pool, and establish an energy consumption linear regression model E. total ; The migration dataset is obtained by aggregating the data volume and total energy consumption of the task migration between the abnormal resource pool and the target resource pool in each historical task migration record. Based on the migration dataset, the least squares method is used to calculate the unit energy consumption k in the energy consumption linear regression model, and the fixed energy consumption E is calculated based on the unit energy consumption k. fixed ; Step S33: Analyze the energy efficiency benefits of task migration between the abnormal resource pool and the target resource pool under different data volumes during the task migration process. The specific analysis process is as follows: Calculate the energy efficiency benefit value E for migrating tasks from the abnormal resource pool to the target resource pool, and ensure that the energy efficiency benefit value E is greater than the target energy efficiency benefit threshold E. ▽ The minimum total amount of data migrated from the abnormal resource pool to the target resource pool is denoted as ζ, which is the minimum total amount of data migrated from the abnormal resource pool to the target resource pool. min ; The energy efficiency benefit value E is greater than the target energy efficiency benefit threshold E. ▽ The minimum total amount of data migrated from the abnormal resource pool to the target resource pool is denoted as ζ, which is the minimum total amount of data migrated from the abnormal resource pool to the target resource pool. min ; Step S34: Obtain the task completion rate of the target resource pool to the abnormal resource pool from the historical task migration record; Get the maximum total processing time T of the preset tasks in the abnormal resource pool. long Calculate the completion time T of the migration task from the target resource pool to the abnormal resource pool. △ ; Get T△ <T long The maximum amount of data that is expected to be migrated from the abnormal resource pool to the target resource pool at that time is denoted as the maximum amount of data that can be migrated from the abnormal resource pool to the target resource pool. max Obtain the task migration data range ζ=[ζ] between the abnormal resource pool and the target resource pool. min， ζ max ]; The above steps analyze the network and energy efficiency between the abnormal resource pool and the target resource pool to dynamically adjust the preset energy efficiency benefit threshold, thereby obtaining the target energy efficiency benefit threshold that adapts to the relationship between the abnormal resource pool and the target resource pool. Furthermore, by using historical task migration records between the abnormal resource pool and the target resource pool, and the target energy efficiency benefit threshold, the minimum total data volume of the migration task between the abnormal resource pool and the target resource pool is obtained. This ensures that no increase in energy consumption occurs during task migration from the abnormal resource pool to the target resource pool, based on the specific value of the migration task volume. Additionally, by comparing the migration time and completion time with the preset maximum total time for task completion in the abnormal resource pool, the maximum total data volume for task migration from the abnormal resource pool to the target pool is obtained. This, in turn, ensures that the task processing time is not excessively long, providing a foundation for subsequent task collaboration.

[0009] Furthermore, step S4 includes: Step S41: Obtain the task data of each pending task in the abnormal resource pool, wherein the task data includes the total amount of data of the pending tasks; Obtain the target resource pools of the abnormal resource pool, and obtain the task migration data range between the abnormal resource pool and each target resource pool; Step S42: Record the unfinished tasks in the abnormal resource pool as pending tasks. According to the task migration range between the abnormal resource pool and the target resource pool, randomly select several pending tasks from the abnormal resource pool as target tasks for task migration from the abnormal resource pool to the target resource pool. The sum of the data volume of the several pending tasks is within the task migration range between the abnormal resource pool and the target resource pool. Migrate target tasks from the abnormal resource pool to the target resource pool, and control the target resource pool to collaboratively complete tasks from the abnormal resource pool.

[0010] To better implement the above methods, a heterogeneous computing power collaborative system based on energy efficiency optimization is also proposed. The system includes a computing power resource anomaly analysis module, a resource pool migration benefit evaluation module, a migration benefit analysis module, and a task collaboration module. The computing power resource anomaly analysis module is used to analyze the monitoring records and historical monitoring records in the regional resource pool to identify the abnormal status of the computing power resources in the regional resource pool and obtain the abnormal resource pool. The resource pool migration benefit evaluation module is used to obtain historical migration network records between abnormal resource pools and other resource pools, evaluate the migration benefit of other resource pools to abnormal resource pools, and obtain the target resource pool. The migration benefit analysis module is used to generate the target energy efficiency benefit threshold between the heterogeneous resource pool and the target resource pool, acquire the energy efficiency data of the abnormal resource pool and the target resource pool, analyze the energy efficiency benefit brought by the migration task between the abnormal resource pool and the target resource pool under different data volumes, and obtain the task migration data range between the abnormal resource pool and the candidate resource pool. The task coordination module is used to obtain the target tasks that need to be migrated from the abnormal resource pool to the target resource pool according to the task migration data range, migrate the target tasks in the abnormal resource pool to the target resource pool, and control the target resource pool to coordinate the completion of tasks in the abnormal resource pool.

[0011] Furthermore, the computing resource anomaly analysis module includes an indicator range acquisition unit and a computing resource anomaly analysis unit; The indicator range acquisition unit is used to acquire the target range of various monitoring indicators of the regional resource pool in the current period; The computing power resource anomaly analysis unit is used to analyze the computing power resource anomalies of the regional resource pool in the current period according to the target range of various monitoring indicators, and to obtain the abnormal resource pool.

[0012] Furthermore, the resource pool migration benefit assessment module includes a resource pool marking unit and a resource pool migration benefit assessment unit; The resource pool marking unit is used to calculate the overall network score between abnormal resource pools and other resource pools, and to mark other resource pools based on the overall network score. The resource pool migration benefit assessment unit is used to calculate the risk values ​​of various monitoring indicators in other resource pools. When the risk values ​​of various monitoring indicators in other resource pools are all less than the preset risk threshold, it is determined that other resource pools are beneficial for migrating tasks from abnormal resource pools, and other resource pools are recorded as the target resource pools of abnormal resource pools.

[0013] Furthermore, the migration benefit analysis module includes an energy efficiency benefit threshold generation unit and a migration benefit analysis unit; The energy efficiency benefit threshold generation unit is used to acquire energy efficiency setting data and monitoring records of abnormal resource pools, and generate the target energy efficiency benefit threshold between the abnormal resource pool and the target resource pool. The migration benefit analysis unit is used to analyze the energy efficiency benefits of migration tasks between the abnormal resource pool and the target resource pool under different data volumes, based on the target energy efficiency benefit threshold, and to obtain the task migration data range between the abnormal resource pool and the candidate resource pool.

[0014] Furthermore, the task coordination module includes task coordination units; The task coordination unit is used to obtain the target tasks to be migrated from the abnormal resource pool to the target resource pool based on the task migration data range between the abnormal resource pool and the candidate resource pool, migrate the target tasks in the abnormal resource pool to the target resource pool, and control the target resource pool to coordinate the completion of tasks in the abnormal resource pool.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention achieves effective coordination of regional resource pools with heterogeneous computing power in different areas of a cloud platform based on energy efficiency optimization technology. By acquiring the target resource pool of abnormal resource pools with abnormal computing power resources, a bridge relationship for task migration is established. Furthermore, from the perspective of energy efficiency, it fundamentally avoids the possibility that the cost of data migration may offset the energy efficiency benefits of remote computing. Moreover, it sets an optimal range for the number of task migrations between abnormal resource pools and target resource pools, enabling regional resource pools in different areas of the cloud platform with heterogeneous computing power to work effectively together, greatly reducing the energy consumed by regional resource pools to complete tasks, and achieving true optimization of energy efficiency. Attached Figure Description

[0016] Figure 1 This is a method logic diagram of a heterogeneous computing power collaborative method based on energy efficiency optimization according to the present invention; Figure 2 This is a flowchart of a heterogeneous computing power collaborative system based on energy efficiency optimization according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Figures 1-2 As shown, this invention provides a technical solution: a heterogeneous computing power collaboration method based on energy efficiency optimization, the method comprising: Step S1: Monitor the regional resource pool in the current period, obtain the monitoring record of the regional resource pool, obtain the historical monitoring record of the regional resource pool, analyze the abnormal status of the computing power resources of the regional resource pool in the current period, and obtain the abnormal resource pool. Step S1 includes: Step S11: Monitor the regional resource pools deployed in different regions within the current period to obtain the monitoring records of the regional resource pools within the current period; Step S12: Obtain data on various monitoring indicators of the regional resource pool from the monitoring records; For example, the monitoring indicators include CPU utilization, GPU utilization, and memory utilization; Obtain all historical monitoring records of the regional resource pool, and extract the mean and standard deviation of the average values ​​of various monitoring indicators in the regional resource pool from each historical monitoring record; Obtain the j-th range Q of the a-th monitoring indicator in the regional resource pool. a j =[μ a -j×σ a ,μ a +j×σ a ], where μ a σ is the mean of the average values ​​of the a-th monitoring indicator in all historical monitoring records. a is the standard deviation of the average value of the a-th monitoring indicator in each historical monitoring record; Obtain the duration of the distance between historical monitoring records and the current period, and record it as the distance duration of the historical monitoring record. Sort the historical monitoring records according to the distance duration, and obtain the maximum distance duration t among all historical monitoring records. max Calculate the characteristic time value of the b-th historical monitoring record in each historical monitoring record. , t b The distance duration of the b-th historical monitoring record; Step S13: Obtain the sum T of the characteristic time values ​​of each historical monitoring record. sum Obtain the average value of the a-th monitoring indicator in the j-th range Q. a j From several historical monitoring records, obtain the sum T of characteristic time values ​​from these historical monitoring records. j sum Calculate the j-th range Q a j Feature proportion Set the feature proportion threshold L △ When L j ≥L △ And L j-1 <L △ Then determine the j-th range Q a j The target range for the a-th monitoring indicator in the regional resource pool during the current period; Obtain the target range of various monitoring indicators of the regional resource pool in the current period. If the maximum or minimum value of any monitoring indicator in the monitoring record is not within the target range, it is determined that there is an abnormal computing power resource in the regional resource pool in the current period, and the regional resource pool is recorded as an abnormal resource pool. Step S2: Obtain historical migration network records between the abnormal resource pool and other resource pools, and combine them with the monitoring records of other resource pools to evaluate the migration benefits of other resource pools to the abnormal resource pool, and obtain the target resource pool. Step S2 includes: Step S21: Remove abnormal resource pools from the resource pools of each region in the platform, and record the remaining resource pools as the target resource pools; Monitor the network status when migrating tasks between the abnormal resource pool and other resource pools to obtain historical migration network records between the abnormal resource pool and other resource pools. Set the unit duration and obtain the transmission rate set from the historical migration network records; For example, the specific process of obtaining the transmission rate set in historical migration network records is as follows: The average network transmission rate of the abnormal resource pool and other resource pools in each unit of time during the task migration process is obtained from the historical migration network records and aggregated to obtain the transmission rate set in the historical migration network records. Step S22: Obtain the preset maximum theoretical network transmission rate B between the abnormal resource pool and other resource pools, and calculate the bandwidth utilization F within the d-th unit of time in the historical migration network record. d ; For example, bandwidth utilization F d The specific calculation formula is as follows: , where R d It represents the average network transmission rate within the d-th unit of time in the historical migration network record transmission rate set; Obtain the mean value F of the average bandwidth utilization in each historical migration network record; For example, the process of obtaining average bandwidth utilization from historical migration network records is as follows: Obtain the average bandwidth utilization rate within each unit of time in the historical migration network record, and record it as the average bandwidth utilization rate in the historical migration network record; Step S23: Obtain the mean μ´ and standard deviation σ´ of the average network transmission rate within each unit time period in the transmission rate set of historical migration network records, and calculate the network comprehensive score β between the abnormal resource pool and other resource pools. For example, the specific process for calculating the network comprehensive score β between the abnormal resource pool and other resource pools is as follows: Calculate network fluctuation values ​​in historical migration network records Obtain the network fluctuation value C from each historical migration network record; Calculate the network comprehensive score β between the abnormal resource pool and other resource pools: , Where, η c and η F These are the preset bandwidth weighting coefficient and stability coefficient, η. c +η F =1, η c >0, η F >0; If the network comprehensive score β is greater than the preset threshold, other resource pools will be marked. Step S24: Obtain the monitoring records of other marked resource pools, and calculate the abnormal risk value γ of the g-th monitoring indicator in other resource pools. g ; For example, the abnormal risk value γ g The specific calculation formula process is as follows: Obtain the average value H of the g-th monitoring indicator from the monitoring records. g ; Obtain the target range of various monitoring indicators for other resource pools within the current period, and obtain the maximum and minimum values ​​H of the g-th monitoring indicator within the target range. max and H min Calculate the abnormal risk value γ of the g-th monitoring indicator in other resource pools. g : , The order m of obtaining abnormal risk values g Calculate the risk value of the g-th monitoring indicator in other resource pools. , of which M sum This represents the total number of all other resource pools. For example, the order m of the abnormal risk values g The specific acquisition process is as follows: Obtain the abnormal risk value of the g-th monitoring indicator in each of the other resource pools of the platform, sort the abnormal risk values ​​of the g-th monitoring indicator in each of the other resource pools from smallest to largest, and obtain the abnormal risk value γ in the other resource pools. g The order m of the abnormal risk values ​​of the g-th monitoring indicator in each of the other resource pools g ; Step S25: Obtain the risk values ​​of various monitoring indicators in other resource pools. When the risk values ​​of various monitoring indicators in other resource pools are all less than the preset risk threshold, it is determined that other resource pools are beneficial for migrating tasks from abnormal resource pools, and the marked other resource pools are recorded as the target resource pools of abnormal resource pools. Step S3: Obtain energy efficiency setting data from the platform and combine it with the monitoring records of the abnormal resource pool to generate the target energy efficiency benefit threshold between the abnormal resource pool and the target resource pool. Obtain historical task migration records between the abnormal resource pool and the target resource pool. Obtain energy efficiency data of the abnormal resource pool and the target resource pool. Analyze the energy efficiency benefit brought by the migration tasks between the abnormal resource pool and the target resource pool under different data volumes. Obtain the task migration data range between the abnormal resource pool and the candidate resource pool. Step S3 includes: Step S31: Obtain preset energy efficiency setting data from the platform, and obtain the standard values ​​E for energy efficiency benefit threshold, network latency, and energy efficiency difference from the energy efficiency setting data. △ W △ and U △ ; For example, energy efficiency refers to the energy consumed to process a unit of task, specifically how many joules of energy are used to process each GB of data; Calculate the target energy efficiency benefit threshold E between the abnormal resource pool and the target resource pool. ▽ ; For example, the target energy efficiency benefit threshold E ▽ The specific calculation process is as follows: Obtain the average network latency W from the monitoring records of the abnormal resource pool in the current period, obtain the energy efficiency difference U between the abnormal resource pool and the target resource pool from the energy efficiency data, and calculate the target energy efficiency benefit threshold between the abnormal resource pool and the target resource pool. , where λ W and λ U These are the preset network weight coefficients and energy efficiency weight coefficients, respectively; Step S32: Obtain historical task migration records between the abnormal resource pool and the target resource pool, and establish an energy consumption linear regression model E. total ; For example, establishing an energy consumption linear regression model E total Specifically: Extract the data volume and total energy consumption of task migration between the abnormal resource pool and the target resource pool from historical task migration records, and establish an energy consumption linear regression model. , of which E total E represents the total energy consumption of mission migrations in the historical mission migration record. fixed For fixed energy consumption, k is the unit energy consumption, and ζ is the energy consumption per unit. sum This represents the total amount of data related to task migrations in the historical task migration records. For example, unit energy consumption is the energy consumed per GB of data migration, where the unit of energy consumption is kWh; The migration dataset is obtained by aggregating the data volume and total energy consumption of the task migration between the abnormal resource pool and the target resource pool in each historical task migration record. Based on the migration dataset, the least squares method is used to calculate the unit energy consumption k in the energy consumption linear regression model, and the fixed energy consumption E is calculated based on the unit energy consumption k. fixed ; For example, there are three historical task migration records between the abnormal resource pool and the target resource pool. The total energy consumption and total data volume of the task migration in the first historical task migration record are 1.5kWh and 100GB, respectively. The total energy consumption and total data volume of the task migration in the second historical task migration record are 2.2kWh and 200GB, respectively. The total energy consumption and total data volume of the task migration in the third historical task migration record are 2.9kWh and 300GB, respectively. Based on the energy consumption linear regression model To calculate E fixed From k, we can get k = 0.007 kWh / GB, E fixed =0.8kWh; Step S33: Analyze the energy efficiency benefits of task migration between the abnormal resource pool and the target resource pool under different data volumes during the task migration process. The specific analysis process is as follows: Calculate the energy efficiency benefit value E for migrating tasks from the abnormal resource pool to the target resource pool, and ensure that the energy efficiency benefit value E is greater than the target energy efficiency benefit threshold E. ▽ The minimum total amount of data migrated from the abnormal resource pool to the target resource pool is denoted as ζ, which is the minimum total amount of data migrated from the abnormal resource pool to the target resource pool. min ; For example, the calculation process for the energy efficiency benefit value E is as follows: Obtain energy efficiency data for the abnormal resource pool, extract the energy efficiency of the abnormal resource pool from the energy efficiency data, and calculate the energy efficiency benefit value E for migrating tasks from the abnormal resource pool to the target resource pool: , Among them, ζ´ sum This represents the total amount of data migrated from the abnormal resource pool to the target resource pool. The energy efficiency benefit value E is greater than the target energy efficiency benefit threshold E. ▽ The minimum total amount of data migrated from the abnormal resource pool to the target resource pool is denoted as ζ, which is the minimum total amount of data migrated from the abnormal resource pool to the target resource pool. min ; Step S34: Obtain the task completion rate of the target resource pool to the abnormal resource pool from the historical task migration record; For example, the specific process for obtaining the task completion rate of the target resource pool to the abnormal resource pool in the historical task migration record is as follows: The ratio between the total amount of migration data and the migration time when migrating from the abnormal resource pool to the target resource pool is obtained from the historical task migration records, thus obtaining the migration data rate between the abnormal resource pool and the target resource pool in the historical task migration records. Obtain the time consumed by the target resource pool to complete the migration of tasks from the abnormal resource pool from the historical task migration records, and record it as the task completion time; The total amount of data migrated from the abnormal resource pool to the target resource pool in the historical task migration record is divided by the task completion time of the target resource pool to the abnormal resource pool in the historical task migration record to obtain the task completion rate of the target resource pool to the abnormal resource pool in the historical task migration record. Get the maximum total processing time T of the preset tasks in the abnormal resource pool. long Calculate the completion time T of the migration task from the target resource pool to the abnormal resource pool. △ ; For example, completion time T △ The specific acquisition process is as follows: Obtain the average migration data rate V´ when migrating tasks from the abnormal resource pool to the target resource pool in each historical task migration record, and obtain the average task completion rate v of the target resource pool to the abnormal resource pool in each historical task migration record. △ ; Calculate the completion time of migration tasks from the target resource pool to the abnormal resource pool. , where ζ △ This represents the total amount of data expected to migrate from the abnormal resource pool to the target resource pool. Get T △ <T long The maximum amount of data that is expected to be migrated from the abnormal resource pool to the target resource pool at that time is denoted as the maximum amount of data that can be migrated from the abnormal resource pool to the target resource pool. max Obtain the task migration data range ζ=[ζ] between the abnormal resource pool and the target resource pool. min， ζ max ]; Step S4: Obtain the task data of pending tasks in the abnormal resource pool, obtain the target tasks to be migrated from the abnormal resource pool to the target resource pool according to the task migration data range between the abnormal resource pool and the target resource pool, migrate the target tasks in the abnormal resource pool to the target resource pool, and control the target resource pool to collaboratively complete the tasks in the abnormal resource pool. Step S4 includes: Step S41: Obtain the task data of each pending task in the abnormal resource pool, wherein the task data includes the total amount of data of the pending tasks; Obtain the target resource pools of the abnormal resource pool, and obtain the task migration data range between the abnormal resource pool and each target resource pool; Step S42: Record the unfinished tasks in the abnormal resource pool as pending tasks. According to the task migration range between the abnormal resource pool and the target resource pool, randomly select several pending tasks from the abnormal resource pool as target tasks for task migration from the abnormal resource pool to the target resource pool. The sum of the data volume of the several pending tasks is within the task migration range between the abnormal resource pool and the target resource pool. Migrate the target tasks in the abnormal resource pool to the target resource pool, and control the target resource pool to collaboratively complete the tasks in the abnormal resource pool; To better implement the above methods, a heterogeneous computing power collaborative system based on energy efficiency optimization is also proposed. The system includes a computing power resource anomaly analysis module, a resource pool migration benefit evaluation module, a migration benefit analysis module, and a task collaboration module. The computing power resource anomaly analysis module is used to analyze the monitoring records and historical monitoring records in the regional resource pool to identify the abnormal status of the computing power resources in the regional resource pool and obtain the abnormal resource pool. The resource pool migration benefit evaluation module is used to obtain historical migration network records between abnormal resource pools and other resource pools, evaluate the migration benefit of other resource pools to abnormal resource pools, and obtain the target resource pool. The migration benefit analysis module is used to generate the target energy efficiency benefit threshold between the heterogeneous resource pool and the target resource pool, acquire the energy efficiency data of the abnormal resource pool and the target resource pool, analyze the energy efficiency benefit brought by the migration task between the abnormal resource pool and the target resource pool under different data volumes, and obtain the task migration data range between the abnormal resource pool and the candidate resource pool. The task coordination module is used to obtain the target tasks that are to be migrated from the abnormal resource pool to the target resource pool according to the task migration data range, migrate the target tasks in the abnormal resource pool to the target resource pool, and control the target resource pool to complete the task coordination in the abnormal resource pool. The computing resource anomaly analysis module includes an indicator range acquisition unit and a computing resource anomaly analysis unit. The indicator range acquisition unit is used to acquire the target range of various monitoring indicators of the regional resource pool in the current period; The computing power resource anomaly analysis unit is used to analyze the computing power resource anomalies of the regional resource pool in the current period according to the target range of various monitoring indicators, and to obtain the abnormal resource pool. The resource pool migration benefit assessment module includes a resource pool marking unit and a resource pool migration benefit assessment unit. The resource pool marking unit is used to calculate the overall network score between abnormal resource pools and other resource pools, and to mark other resource pools based on the overall network score. The resource pool migration benefit assessment unit is used to calculate the risk values ​​of various monitoring indicators in other resource pools. When the risk values ​​of various monitoring indicators in other resource pools are all less than the preset risk threshold, it is determined that other resource pools are beneficial for migrating tasks from abnormal resource pools, and other resource pools are recorded as the target resource pools of abnormal resource pools. The migration benefit analysis module includes an energy efficiency benefit threshold generation unit and a migration benefit analysis unit. The energy efficiency benefit threshold generation unit is used to acquire energy efficiency setting data and monitoring records of abnormal resource pools, and generate the target energy efficiency benefit threshold between the abnormal resource pool and the target resource pool. The migration benefit analysis unit is used to analyze the energy efficiency benefits of migration tasks between the abnormal resource pool and the target resource pool under different data volumes, based on the target energy efficiency benefit threshold, and to obtain the task migration data range between the abnormal resource pool and the candidate resource pool. The task coordination module includes a task coordination unit. The task coordination unit is used to obtain the target tasks to be migrated from the abnormal resource pool to the target resource pool based on the task migration data range between the abnormal resource pool and the candidate resource pool, migrate the target tasks in the abnormal resource pool to the target resource pool, and control the target resource pool to coordinate the completion of tasks in the abnormal resource pool.

[0019] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A heterogeneous computing power collaborative method based on energy efficiency optimization, characterized in that, The method includes: Step S1: Monitor the regional resource pool in the current period, obtain the monitoring record of the regional resource pool, obtain the historical monitoring record of the regional resource pool, analyze the abnormal status of the computing power resources of the regional resource pool in the current period, and obtain the abnormal resource pool. Step S2: Obtain historical migration network records between the abnormal resource pool and other resource pools, and combine them with the monitoring records of other resource pools to evaluate the migration benefits of other resource pools to the abnormal resource pool, and obtain the target resource pool. Step S3: Obtain energy efficiency setting data from the platform and combine it with the monitoring records of the abnormal resource pool to generate the target energy efficiency benefit threshold between the abnormal resource pool and the target resource pool. Obtain historical task migration records between the abnormal resource pool and the target resource pool. Obtain energy efficiency data of the abnormal resource pool and the target resource pool. Analyze the energy efficiency benefit brought by the migration tasks between the abnormal resource pool and the target resource pool under different data volumes. Obtain the task migration data range between the abnormal resource pool and the candidate resource pool. Step S4: Obtain the task data of pending tasks in the abnormal resource pool. Based on the task migration data range between the abnormal resource pool and the target resource pool, obtain the target tasks to be migrated from the abnormal resource pool to the target resource pool. Migrate the target tasks in the abnormal resource pool to the target resource pool and control the target resource pool to collaboratively complete the tasks in the abnormal resource pool.

2. The heterogeneous computing power collaboration method based on energy efficiency optimization according to claim 1, characterized in that, Step S1 includes: Step S11: Monitor the regional resource pools deployed in different regions within the current period to obtain the monitoring records of the regional resource pools within the current period; Step S12: Obtain data on various monitoring indicators of the regional resource pool from the monitoring records; Obtain all historical monitoring records of the regional resource pool, and extract the mean and standard deviation of the average values ​​of various monitoring indicators in the regional resource pool from each historical monitoring record; Obtain the j-th range Q of the a-th monitoring indicator in the regional resource pool. a j =[μ a -j×σ a ,μ a +j×σ a ], where μ a σ is the mean of the average values ​​of the a-th monitoring indicator in all historical monitoring records. a is the standard deviation of the average value of the a-th monitoring indicator in each historical monitoring record; Obtain the duration of the distance between historical monitoring records and the current period, and record it as the distance duration of the historical monitoring record. Sort the historical monitoring records according to the distance duration, and obtain the maximum distance duration t among all historical monitoring records. max Calculate the characteristic time value of the b-th historical monitoring record in each historical monitoring record. , t b The distance duration of the b-th historical monitoring record; Step S13: Obtain the sum T of the characteristic time values ​​of each historical monitoring record. sum Obtain the average value of the a-th monitoring indicator in the j-th range Q. a j From several historical monitoring records, obtain the sum T of characteristic time values ​​from these historical monitoring records. j sum Calculate the j-th range Q a j Feature proportion Set the feature proportion threshold L △ When L j ≥L △ And L j-1 <L △ Then determine the j-th range Q a j The target range for the a-th monitoring indicator in the regional resource pool during the current period; Obtain the target range of various monitoring indicators of the regional resource pool in the current period. If the maximum or minimum value of any monitoring indicator in the monitoring record is not within the target range, it is determined that there is an abnormal computing power resource in the regional resource pool in the current period, and the regional resource pool is recorded as an abnormal resource pool.

3. The heterogeneous computing power collaboration method based on energy efficiency optimization according to claim 1, characterized in that, Step S2 includes: Step S21: Remove abnormal resource pools from the resource pools of each region in the platform, and record the remaining resource pools as the target resource pools; Monitor the network status when migrating tasks between the abnormal resource pool and other resource pools to obtain historical migration network records between the abnormal resource pool and other resource pools. Set the unit duration and obtain the transmission rate set from the historical migration network records; Step S22: Obtain the preset maximum theoretical network transmission rate B between the abnormal resource pool and other resource pools, and calculate the bandwidth utilization rate F in the d-th unit of time in the historical migration network record. d ; Obtain the mean value F of the average bandwidth utilization in each historical migration network record; Step S23: Obtain the mean μ´ and standard deviation σ´ of the average network transmission rate within each unit time period in the transmission rate set of the historical migration network records, and calculate the network comprehensive score β between the abnormal resource pool and other resource pools; If the network comprehensive score β is greater than the preset threshold, then the other resource pools are marked. Step S24: Obtain the monitoring records of the marked other resource pools, and calculate the abnormal risk value γ of the g-th monitoring indicator in the other resource pools. g ; The order m of obtaining abnormal risk values g Calculate the risk value of the g-th monitoring indicator in the other resource pools. , of which M sum This represents the total number of all other resource pools. Step S25: Obtain the risk values ​​of each monitoring indicator in the other resource pools. When the risk values ​​of each monitoring indicator in the other resource pools are all less than the preset risk threshold, it is determined that the other resource pools are beneficial for migrating tasks from the abnormal resource pools, and the marked other resource pools are recorded as the target resource pools of the abnormal resource pools.

4. The heterogeneous computing power collaboration method based on energy efficiency optimization according to claim 3, characterized in that, Step S3 includes: Step S31: Obtain preset energy efficiency setting data from the platform, and obtain the standard values ​​E for energy efficiency benefit threshold, network latency, and energy efficiency difference from the energy efficiency setting data. △ W △ and U △ ; Calculate the target energy efficiency benefit threshold E between the abnormal resource pool and the target resource pool. ▽ ; Step S32: Obtain historical task migration records between the abnormal resource pool and the target resource pool, and establish an energy consumption linear regression model E. total ; The migration dataset is obtained by aggregating the data volume and total energy consumption of the task migration between the abnormal resource pool and the target resource pool in each historical task migration record. Based on the migration dataset, the least squares method is used to calculate the unit energy consumption k in the energy consumption linear regression model, and the fixed energy consumption E is calculated based on the unit energy consumption k. fixed ; Step S33: Analyze the energy efficiency benefits of task migration between the abnormal resource pool and the target resource pool under different data volumes during the task migration process. The specific analysis process is as follows: Calculate the energy efficiency benefit value E for migrating tasks from the abnormal resource pool to the target resource pool, and ensure that the energy efficiency benefit value E is greater than the target energy efficiency benefit threshold E. ▽ The minimum total amount of data migrated from the abnormal resource pool to the target resource pool is denoted as ζ, which is the minimum total amount of data migrated from the abnormal resource pool to the target resource pool. min ; The energy efficiency benefit value E is greater than the target energy efficiency benefit threshold E. ▽ The minimum total amount of data migrated from the abnormal resource pool to the target resource pool is denoted as ζ, which is the minimum total amount of data migrated from the abnormal resource pool to the target resource pool. min ; Step S34: Obtain the task completion rate of the target resource pool to the abnormal resource pool from the historical task migration record; Get the maximum total processing time T of the preset tasks in the abnormal resource pool. long Calculate the completion time T of the migration task from the target resource pool to the abnormal resource pool. △ ; Get T △ <T long The maximum amount of data that is expected to be migrated from the abnormal resource pool to the target resource pool at that time is denoted as the maximum amount of data that can be migrated from the abnormal resource pool to the target resource pool. max Obtain the task migration data range ζ=[ζ] between the abnormal resource pool and the target resource pool. min， ζ max ].

5. The heterogeneous computing power collaboration method based on energy efficiency optimization according to claim 4, characterized in that, Step S4 includes: Step S41: Obtain the task data of each pending task in the abnormal resource pool, wherein the task data includes the total amount of data of the pending tasks; Obtain the target resource pools of the abnormal resource pool, and obtain the task migration data range between the abnormal resource pool and each target resource pool; Step S42: Record the unfinished tasks in the abnormal resource pool as pending tasks. According to the task migration range between the abnormal resource pool and the target resource pool, randomly select several pending tasks from the abnormal resource pool as target tasks for task migration from the abnormal resource pool to the target resource pool. The sum of the data volume of the several pending tasks is within the task migration range between the abnormal resource pool and the target resource pool. Migrate target tasks from the abnormal resource pool to the target resource pool, and control the target resource pool to collaboratively complete tasks from the abnormal resource pool.

6. A heterogeneous computing power collaborative system based on energy efficiency optimization, used to execute the heterogeneous computing power collaborative method based on energy efficiency optimization as described in any one of claims 1-5, characterized in that, The system includes a computing resource anomaly analysis module, a resource pool migration benefit evaluation module, a migration revenue analysis module, and a task collaboration module; The computing power resource anomaly analysis module is used to analyze the monitoring records and historical monitoring records in the regional resource pool to determine the abnormal status of the computing power resources in the regional resource pool and obtain the abnormal resource pool. The resource pool migration benefit evaluation module is used to acquire historical migration network records between abnormal resource pools and other resource pools, evaluate the migration benefit of other resource pools to abnormal resource pools, and obtain the target resource pool. The migration benefit analysis module is used to generate the target energy efficiency benefit threshold between the heterogeneous resource pool and the target resource pool, acquire the energy efficiency data of the abnormal resource pool and the target resource pool, analyze the energy efficiency benefit brought by the migration task between the abnormal resource pool and the target resource pool under different data volumes, and obtain the task migration data range between the abnormal resource pool and the candidate resource pool. The task coordination module is used to acquire the target tasks that are to be migrated from the abnormal resource pool to the target resource pool according to the task migration data range, migrate the target tasks in the abnormal resource pool to the target resource pool, and control the target resource pool to coordinate the completion of tasks in the abnormal resource pool.

7. A heterogeneous computing power collaborative system based on energy efficiency optimization according to claim 6, characterized in that, The computing resource anomaly analysis module includes an indicator range acquisition unit and a computing resource anomaly analysis unit. The indicator range acquisition unit is used to acquire the target range of various monitoring indicators of the regional resource pool in the current period; The computing power resource anomaly analysis unit is used to analyze the computing power resource anomalies of the regional resource pool in the current period according to the target range of various monitoring indicators, and obtain the abnormal resource pool.

8. A heterogeneous computing power collaborative system based on energy efficiency optimization according to claim 6, characterized in that, The resource pool migration benefit assessment module includes a resource pool marking unit and a resource pool migration benefit assessment unit; The resource pool marking unit is used to calculate the comprehensive network score between abnormal resource pools and other resource pools, and to mark other resource pools based on the comprehensive network score. The resource pool migration benefit assessment unit is used to calculate the risk values ​​of various monitoring indicators in the other resource pools. When the risk values ​​of various monitoring indicators in the other resource pools are all less than the preset risk threshold, it is determined that the other resource pools are beneficial for migrating the abnormal resource pool tasks, and the other resource pools are recorded as the target resource pools of the abnormal resource pools.

9. A heterogeneous computing power collaborative system based on energy efficiency optimization according to claim 6, characterized in that, The migration benefit analysis module includes an energy efficiency benefit threshold generation unit and a migration benefit analysis unit. The energy efficiency benefit threshold generation unit is used to acquire energy efficiency setting data and monitoring records of abnormal resource pools, and generate a target energy efficiency benefit threshold between the abnormal resource pool and the target resource pool. The migration benefit analysis unit is used to analyze the energy efficiency benefits of migration tasks between the abnormal resource pool and the target resource pool under different data volumes, based on the target energy efficiency benefit threshold, and to obtain the task migration data range between the abnormal resource pool and the candidate resource pool.

10. A heterogeneous computing power collaborative system based on energy efficiency optimization according to claim 6, characterized in that, The task coordination module includes a task coordination unit; The task coordination unit is used to obtain the target task to be migrated from the abnormal resource pool to the target resource pool based on the task migration data range between the abnormal resource pool and the candidate resource pool, migrate the target task in the abnormal resource pool to the target resource pool, and control the target resource pool to coordinate the completion of the task in the abnormal resource pool.

Images