A computing power scheduling method and device

Abstract

The application discloses a computing power scheduling method and device, the method comprises: identifying the time delay sensitivity of a to-be-processed job to obtain a time delay sensitivity identification result; in the case where the time delay sensitivity identification result indicates a non-time delay sensitive type, determining a candidate computing power cluster matching the resource demand of the to-be-processed job from a plurality of computing power clusters; the plurality of computing power clusters comprise a first type computing power cluster based on green energy power and a second type computing power cluster based on grid mains; according to the current green energy data and the current cluster load information of each candidate computing power cluster, determining the scheduling score of each candidate computing power cluster, and taking the candidate computing power cluster with the highest scheduling score as a target computing power cluster; distributing the to-be-processed job to the target computing power cluster, so that the target computing power cluster executes the to-be-processed job. The application improves the utilization rate of green energy power and the stability and flexibility of computing power scheduling, and reduces the operation cost of the data center.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a computing power scheduling method and apparatus. Background Technology

[0002] With the rapid development of big data and artificial intelligence technologies, computing power has become a core resource supporting complex computing tasks and intelligent applications. Data centers, as critical computing hubs, integrate large-scale servers, GPU clusters, and high-speed interconnect components. Among these, green data centers based on green energy electricity offer significant advantages in reducing operating costs compared to traditional data centers relying solely on grid power. However, the computing power scheduling methods in related technologies have low utilization rates of green energy electricity, resulting in still relatively high operating costs for data centers, and also exhibiting poor stability and flexibility in computing power scheduling. Summary of the Invention

[0003] To address the problems of existing technologies, this application provides a computing power scheduling method and apparatus. The technical solution is as follows:

[0004] On the one hand, a computing power scheduling method is provided, the method comprising:

[0005] Get pending jobs;

[0006] The latency sensitivity of the task to be processed is identified, and the latency sensitivity identification result of the task to be processed is obtained.

[0007] If the latency sensitivity identification result indicates that the job to be processed is not latency sensitive, a candidate computing power cluster that matches the resource requirements of the job to be processed is determined from multiple computing power clusters; the multiple computing power clusters include a first type of computing power cluster based on green energy power and a second type of computing power cluster based on grid power.

[0008] Based on the current green energy data and current cluster load information of each candidate computing power cluster, the scheduling score of each candidate computing power cluster is determined, and the candidate computing power cluster with the highest scheduling score is selected as the target computing power cluster.

[0009] The pending jobs are distributed to the target computing power cluster so that the target computing power cluster can execute the pending jobs.

[0010] In some exemplary embodiments, the step of performing latency sensitivity identification on the job to be processed to obtain the latency sensitivity identification result of the job to be processed includes:

[0011] Extract the metadata of the job to be processed, and determine the field content of the corresponding latency sensitivity identifier field in the metadata;

[0012] If the content of the field is not empty, the latency sensitivity identification result of the job to be processed is determined based on the content of the latency sensitivity identifier field.

[0013] If the field content is empty, the latency sensitivity identification result of the job to be processed is determined based on the job type and estimated execution time of the job in the metadata.

[0014] In some exemplary embodiments, determining the scheduling score of each candidate computing power cluster based on its current green energy data and current cluster load information includes:

[0015] For each candidate computing power cluster, a green energy availability score is determined based on the current green energy data of the candidate computing power cluster; the current green energy data includes the current power generation and predicted power generation of the green energy corresponding to the candidate computing power cluster; the green energy availability score characterizes the green energy power supply capacity corresponding to the candidate computing power cluster.

[0016] Based on the current cluster load information of the candidate computing power clusters, a cluster load score corresponding to the candidate computing power cluster is determined; the cluster load score represents the idle level of the candidate computing power cluster.

[0017] The scheduling score of the candidate computing power cluster is obtained by weighted summation of the green energy availability score and the cluster load score corresponding to the candidate computing power cluster.

[0018] In some exemplary embodiments, the weighted summation of the green energy availability score and cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster includes:

[0019] The operating cost score of the candidate computing power cluster is determined based on the current output power of the green energy corresponding to the candidate computing power cluster and its proportion in the rated power demand of the candidate computing power cluster; the operating cost score represents the electricity cost of running the job to be processed in the candidate computing power cluster.

[0020] The scheduling score of the candidate computing power cluster is obtained by weighted summation of the green energy availability score, cluster load score and operating cost score corresponding to the candidate computing power cluster.

[0021] In some exemplary embodiments, the weighted summation of the green energy availability score, cluster load score, and operating cost score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster includes:

[0022] The data transmission cost is determined based on the storage location of the computational data required for the task to be processed;

[0023] Based on the data transmission cost, a data affinity score is determined for the candidate computing power cluster; the data affinity score represents the distance between the computational data required by the job to be processed and the candidate computing power cluster.

[0024] The scheduling score of the candidate computing power cluster is obtained by weighted summation of its green energy availability score, cluster load score, operating cost score, and data affinity score.

[0025] In some exemplary embodiments, the weighted summation of the green energy availability score and cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster includes:

[0026] Obtain the set of weight coefficients; the set of weight coefficients includes the weight coefficients corresponding to each item in the weighted summation process.

[0027] Based on the set of weighted coefficients, the green energy availability score and cluster load score corresponding to the candidate computing power cluster are weighted and summed to obtain the scheduling score of the candidate computing power cluster.

[0028] In some exemplary embodiments, the method further includes:

[0029] If the latency sensitivity identification result indicates that the job to be processed is latency sensitive, the first weight coefficient in the set of weight coefficients is increased and the second weight coefficient in the set of weight coefficients is decreased to obtain the adjusted set of weight coefficients; the first weight coefficient is the weight coefficient corresponding to the cluster load score item, and the second weight coefficient is the weight coefficient corresponding to the green energy availability score item.

[0030] Based on the adjusted set of weight coefficients, the green energy availability score and cluster load score corresponding to the candidate computing power cluster are weighted and summed to obtain the scheduling score of the candidate computing power cluster. The candidate computing power cluster with the highest scheduling score is selected as the target computing power cluster.

[0031] In some exemplary embodiments, after distributing the job to be processed to the target computing power cluster, the method further includes:

[0032] The green energy data and cluster load information of each computing cluster are detected;

[0033] Based on the detection results, update the current green energy data and current cluster load information of each computing cluster.

[0034] On the other hand, a computing power scheduling device is provided, the device comprising:

[0035] The job acquisition module is used to acquire jobs to be processed.

[0036] A latency sensitivity identification module is used to identify the latency sensitivity of the job to be processed and obtain the latency sensitivity identification result of the job to be processed.

[0037] The candidate computing power cluster determination module is used to determine, when the latency sensitivity identification result indicates that the job to be processed is non-latency sensitive, a candidate computing power cluster that matches the resource requirements of the job to be processed from multiple computing power clusters; the multiple computing power clusters include a first type of computing power cluster based on green energy power and a second type of computing power cluster based on grid power.

[0038] The cluster scheduling score determination module is used to determine the scheduling score of each candidate computing power cluster based on the current green energy data and current cluster load information of each candidate computing power cluster, and to select the candidate computing power cluster with the highest scheduling score as the target computing power cluster.

[0039] The job distribution module is used to distribute the job to be processed to the target computing power cluster, so that the target computing power cluster can execute the job to be processed.

[0040] In some exemplary embodiments, the delay sensitivity identification module includes:

[0041] The field content determination module is used to extract the metadata of the job to be processed and determine the field content of the corresponding latency sensitivity identifier field in the metadata;

[0042] The first determining module is used to determine the latency sensitivity identification result of the job to be processed based on the field content of the latency sensitivity identifier field when the field content is not empty.

[0043] The second determining module is used to determine the latency sensitivity identification result of the job to be processed based on the job type and estimated execution time of the job in the metadata when the field content is empty.

[0044] In some exemplary embodiments, the cluster scheduling score determination module includes:

[0045] The green energy availability score determination module is used to determine the green energy availability score corresponding to each candidate computing power cluster based on the current green energy data of the candidate computing power cluster; the current green energy data includes the current power generation and predicted power generation of the green energy corresponding to the candidate computing power cluster; the green energy availability score characterizes the green energy power supply capacity corresponding to the candidate computing power cluster.

[0046] The cluster load score determination module is used to determine the cluster load score corresponding to the candidate computing power cluster based on the current cluster load information of the candidate computing power cluster; the cluster load score represents the idle degree of the candidate computing power cluster.

[0047] The weighted processing module is used to perform weighted summation processing based on the green energy availability score and cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0048] In some exemplary embodiments, the weighting processing module includes:

[0049] The operating cost score determination module is used to determine the operating cost score of the candidate computing power cluster based on the proportion of the current output power of the green energy corresponding to the candidate computing power cluster in the rated power demand of the candidate computing power cluster; the operating cost score represents the electricity cost of running the job to be processed in the candidate computing power cluster.

[0050] The first submodule of weighted processing is used to perform weighted summation processing based on the green energy availability score, cluster load score and operating cost score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0051] In some exemplary embodiments, the weighted processing first submodule includes:

[0052] The data transmission cost determination module is used to determine the data transmission cost based on the storage location of the computational data required for the task to be processed.

[0053] The data affinity score determination module is used to determine the data affinity score corresponding to the candidate computing power cluster based on the data transmission cost; the data affinity score represents the distance between the computational data required by the job to be processed and the candidate computing power cluster.

[0054] The second submodule of weighted processing is used to perform weighted summation processing based on the green energy availability score, cluster load score, operating cost score and data affinity score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0055] In some exemplary embodiments, the weighted processing module is specifically used to: obtain a set of weight coefficients; the set of weight coefficients includes the weight coefficients corresponding to each item in the weighted summation process; based on the set of weight coefficients, perform weighted summation processing on the green energy availability score and cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster, and take the candidate computing power cluster with the highest scheduling score as the target computing power cluster.

[0056] In some exemplary embodiments, the apparatus further includes:

[0057] The weight coefficient adjustment module is used to increase the first weight coefficient in the weight coefficient set and decrease the second weight coefficient in the weight coefficient set when the latency sensitivity identification result indicates that the job to be processed is latency sensitive, so as to obtain an adjusted weight coefficient set; the first weight coefficient is the weight coefficient corresponding to the cluster load score item, and the second weight coefficient is the weight coefficient corresponding to the green energy availability score item.

[0058] The summation processing module is used to perform weighted summation processing on the green energy availability score and cluster load score corresponding to the candidate computing power cluster based on the adjusted weight coefficient set, to obtain the scheduling score of the candidate computing power cluster, and to select the candidate computing power cluster with the highest scheduling score as the target computing power cluster.

[0059] In some exemplary embodiments, the apparatus further includes:

[0060] The detection module is used to detect the green energy data and cluster load information of each computing cluster.

[0061] The update module is used to update the current green energy data and current cluster load information of each computing power cluster based on the detection results.

[0062] On the other hand, an electronic device is provided, including a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the computing power scheduling method of any of the above aspects.

[0063] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, wherein the at least one instruction or the at least one program is loaded and executed by a processor to implement the computing power scheduling method of any of the above aspects.

[0064] On the other hand, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the computing power scheduling methods described above.

[0065] This application embodiment identifies the latency sensitivity of the job to be processed. When the latency sensitivity identification result indicates that it is not latency sensitive, it determines the candidate computing power clusters that match the resource requirements of the job to be processed from multiple computing power clusters. These multiple computing power clusters include a first type of computing power cluster based on green energy power and a second type of computing power cluster based on grid power. Then, based on the current green energy data and current cluster load information of each candidate computing power cluster, it determines the scheduling score of each candidate computing power cluster, and selects the candidate computing power cluster with the highest scheduling score as the target computing power cluster. The job to be processed is then distributed to the target computing power cluster so that the target computing power cluster can execute the job. This effectively transfers a large amount of computing load to computing power clusters with abundant green energy power, improves the utilization rate of green energy power and the local consumption rate of green energy, effectively reduces the operating cost of data centers, and can adapt to the dynamic time-varying nature of green energy power, improving the stability and flexibility of computing power scheduling. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a system architecture diagram of joint scheduling provided in an embodiment of this application;

[0068] Figure 2 This is a flowchart illustrating a computing power scheduling method provided in an embodiment of this application;

[0069] Figure 3 This is a flowchart illustrating another computing power scheduling method provided in an embodiment of this application;

[0070] Figure 4 This is a flowchart illustrating another computing power scheduling method provided in an embodiment of this application;

[0071] Figure 5 This is a structural block diagram of a computing power scheduling device provided in an embodiment of this application;

[0072] Figure 6 This is a hardware structure block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0073] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0074] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0075] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0076] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0077] Please see Figure 1The diagram illustrates a system architecture for joint scheduling provided in this application embodiment. It includes a joint scheduling node 110 and multiple data centers 120 distributed across different geographical locations (e.g., region a, region b, region c, and region d). The joint scheduling node 110 and each data center 120 can communicate via a network connection. The multiple data centers 120 include green data centers 121 and traditional data centers 122. The green data centers 121 are powered by green energy sources (e.g., wind power, hydropower, and photovoltaic power) and grid power, while the traditional data centers 122 are powered solely by grid power. Figure 1 As shown, green data centers 121 are deployed in regions a and b, while traditional data centers 122 are deployed in regions c and d.

[0078] Both the green data center 121 and the traditional data center 122 can include multiple computing power clusters. These clusters can be of types including, but not limited to, Kubernetes clusters and Slurm clusters. The multiple computing power clusters can be heterogeneous computing power clusters, and the heterogeneous computing power within these clusters can include, but is not limited to, graphics processing units (GPUs), neural network processing units (NPUs), tensor processing units (TPUs), and FPGA (Field Programmable Gate Array) hardware. It should be noted that the computing power clusters in this embodiment include a first type of computing power cluster and a second type of computing power cluster. The power source for the first type of computing power cluster includes green energy electricity, while the power source for the second type of computing power cluster is solely grid power. It is understood that the power source for the first type of computing power cluster can include grid power in addition to green energy electricity.

[0079] Specifically, in this embodiment of the application, an energy monitoring agent is deployed in each data center, including a green data center 121 and a traditional data center 122. The energy monitoring agent is used to detect the green energy data of each computing cluster in its data center at preset time intervals, and send the detected green energy data of each computing cluster to the joint scheduling node 110, so that the joint scheduling node 110 can obtain the current green energy data of each computing cluster and provide a data basis for scheduling decisions. The time interval for green energy data detection, i.e., the preset time interval, can be set according to actual needs. Generally, the shorter the preset time interval, the stronger the real-time detection, and the more real-time green energy data detection can be achieved. In this embodiment of the application, the green energy data of the computing power cluster can characterize the green energy supply status of the computing power cluster. Specifically, it can include the current power generation, i.e. the current output power (kW) of green energy, and the predicted power generation. The predicted power generation is the power generation curve for a future period of time (e.g., the next 6 hours) predicted based on current environmental information (e.g., light intensity, wind speed) and historical power generation data model. The historical power generation data model is a trained neural network model used to predict the power generation curve for a future period of time. In practical applications, the current environmental information and the power generation data for a recent period of time (e.g., the last week) can be input into the historical power generation data model to predict the power generation curve for a future period of time.

[0080] In this embodiment, a cluster load monitoring agent is deployed in each computing power cluster. For example, Prometheus Node Exporter and kube-state-metrics are deployed in a Kubernetes cluster, and SlurmExporter for Prometheus is deployed in a Slurm cluster. The cluster load monitoring agent detects the cluster load information of each computing power cluster within its data center at preset time intervals and sends the detected cluster load information to the joint scheduling node 110. This allows the joint scheduling node 110 to obtain the current cluster load information of each computing power cluster, providing a data basis for scheduling decisions. The preset time interval can be set based on actual needs. Generally, the shorter the preset time interval, the stronger the real-time detection, enabling real-time cluster load information detection. In this embodiment, the cluster load information of the computing power cluster represents the load status of the computing power cluster. Specifically, it can include resource utilization, such as the real-time utilization of CPU, memory, and GPU; resource allocation rate, i.e., the ratio of resources allocated to existing jobs to the total resources; job queue status, such as the number of jobs waiting to be executed and the total waiting time; and node status, such as whether each computing node is available.

[0081] The joint scheduling node 110 stores the current green energy data and current cluster load information of each computing power cluster. Understandably, this current green energy data and current cluster load information will be updated based on the detection results of each energy monitoring agent and each cluster load monitoring agent.

[0082] It should be noted that the nodes / servers involved in the embodiments of this application can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The terminal devices involved in the embodiments of this application include, but are not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, and in-vehicle terminals.

[0083] In one exemplary implementation, the joint scheduling node 110 and each computing power cluster can be node devices in the blockchain system, capable of sharing acquired and generated information with other node devices in the blockchain system, thus realizing information sharing among multiple node devices. Multiple node devices in the blockchain system can be configured with the same blockchain, which consists of multiple blocks, and adjacent blocks are related, ensuring that any data tampering in any block can be detected by the next block, thereby preventing data tampering in the blockchain and guaranteeing the security and reliability of the data in the blockchain.

[0084] Please see Figure 2 The diagram shown is a flowchart of a computing power scheduling method provided in an embodiment of this application. This method can be applied to... Figure 1 The joint scheduling node in the process. It should be noted that this specification provides the operational steps of the methods described in the embodiments or flowcharts, but based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one possible execution order among many, and does not represent the only execution order. In actual system or product execution, the methods shown in the embodiments or figures can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment). Specifically, as shown... Figure 2 As shown, the computing power scheduling method in this application embodiment may include:

[0085] S201, Get jobs to be processed.

[0086] Specifically, a job configuration interface can be displayed to the user through a terminal device. The terminal device can respond to the job submission command triggered by the job configuration interface, generate the corresponding scheduling job based on the job parameters configured in the job configuration interface, and send the scheduling job to the joint scheduling node. Thus, the joint scheduling node obtains the scheduling job submitted by the user and treats the scheduling job as a job to be processed.

[0087] The pending job contains metadata generated based on user configuration. This metadata can include the content of multiple preset fields, such as cluster type, resource requirements, runtime estimation, latency sensitivity flag, and job type label. The content of each preset field can be configured by the user through the job configuration interface. Specifically, the cluster type field specifies the cluster type of the job, such as a Kubernetes cluster job or a Slurm cluster job; the resource requirements field configures resource requirements, such as the number of CPU cores, memory size, and GPU type and quantity; the runtime estimation field configures the estimated execution time of the job; the latency sensitivity flag configures whether the job is latency-sensitive, typically a boolean label (true / false) explicitly set by the user; and the job type label field configures the job type, such as model training, online inference, batch processing, or interactive application.

[0088] S203, perform latency sensitivity identification on the job to be processed, and obtain the latency sensitivity identification result of the job to be processed.

[0089] The latency sensitivity identification result of the pending job indicates whether the pending job is latency-sensitive. Latency-sensitive pending jobs are usually highly sensitive to runtime latency and must be executed immediately. For example, online inference and interactive applications are "latency-sensitive" jobs that must be executed immediately. On the other hand, non-latency-sensitive pending jobs are less sensitive to runtime latency and can be postponed, paused, or migrated. For example, AI model training, data batch processing, and scientific simulation are usually "non-latency-sensitive" jobs.

[0090] In some exemplary embodiments, step S203 above, when performing latency sensitivity identification on the job to be processed and obtaining the latency sensitivity identification result of the job to be processed, may include: extracting the metadata of the job to be processed and determining the field content of the corresponding latency sensitivity identifier field in the metadata; if the field content is not empty, determining the latency sensitivity identification result of the job to be processed based on the field content of the latency sensitivity identifier field; if the field content is empty, determining the latency sensitivity identification result of the job to be processed based on the job type and estimated job execution time in the metadata.

[0091] Specifically, if the content of the latency sensitivity identifier field is not empty, it indicates that the user has set whether the scheduled job is latency sensitive. In this case, explicit identification can be performed based on the content of this field. That is, the latency sensitivity identification result of the scheduled job can be determined according to the content of this field set by the user. For example, if the user sets the content of the latency sensitivity identifier field to false, the latency sensitivity identification result of the scheduled job can be determined to be non-latency sensitive. Conversely, if the user sets the content of the latency sensitivity identifier field to true, the latency sensitivity identification result of the scheduled job can be determined to be latency sensitive.

[0092] If the latency sensitivity identifier field is empty, it indicates that the user has not set whether the scheduled job is latency sensitive. In this case, implicit identification can be performed based on the job type and estimated execution time in the metadata. Specifically, when the job type is a preset first job type, or the estimated execution time is less than a preset execution time threshold, the latency sensitivity identification result of the job to be scheduled is determined to be latency sensitive. The preset first job type can be set based on practical experience, such as online inference or interactive applications. The preset execution time threshold can also be set based on practical experience, such as 30 minutes. When the job type is a preset second job type, or the estimated execution time is greater than the preset execution time threshold, the latency sensitivity identification result of the job to be scheduled is determined to be non-latency sensitive. The preset second job type can be set based on practical experience, such as model training or batch processing.

[0093] The above implementation combines explicit and implicit identification of latency sensitivity based on the metadata of the job to be processed, thereby accurately identifying the sensitivity of the job to runtime latency and providing a basis for subsequent finer-grained computing power scheduling based on the latency sensitivity identification results.

[0094] S205, if the latency sensitivity identification result indicates that the job to be processed is not latency sensitive, a candidate computing power cluster that matches the resource requirements of the job to be processed is determined from multiple computing power clusters.

[0095] The multiple computing power clusters include a first type of computing power cluster based on green energy electricity and a second type of computing power cluster based on grid electricity.

[0096] Specifically, the resource requirements of a pending job can be obtained from the metadata of that job. The joint scheduling center can select computing power clusters that meet the resource requirements of the pending job as candidate computing power clusters based on the current cluster load information of each computing power cluster. It can be understood that candidate computing power clusters may include first-type computing power clusters and / or second-type computing power clusters.

[0097] S207, Based on the current green energy data and current cluster load information of each candidate computing power cluster, determine the scheduling score of each candidate computing power cluster, and select the candidate computing power cluster with the highest scheduling score as the target computing power cluster.

[0098] Specifically, the joint scheduling node can traverse each candidate computing power cluster. During the traversal, a scheduling score is calculated for each candidate computing power cluster based on its current green energy data and current cluster load information. This scheduling score allows candidate computing power clusters with abundant green energy power and low cluster load to obtain high scores. After the traversal, the candidate computing power cluster with the highest scheduling score is selected as the target computing power cluster. This allows "non-latency sensitive" scheduling jobs to be scheduled to the green computing power cluster with the highest current scheduling score, thereby maximizing the utilization rate of green energy.

[0099] In some exemplary implementations, for each candidate computing power cluster, step S207 may include the following:

[0100] Based on the current green energy data of the candidate computing power cluster, determine the green energy availability score corresponding to the candidate computing power cluster; based on the current cluster load information of the candidate computing power cluster, determine the cluster load score corresponding to the candidate computing power cluster; perform weighted summation processing on the green energy availability score and cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0101] The current green energy data includes the current and predicted power generation of the green energy corresponding to the candidate computing power cluster; the green energy availability score represents the green energy power supply capacity corresponding to the candidate computing power cluster; and the cluster load score represents the idle level of the candidate computing power cluster.

[0102] In specific implementation, the green energy availability score of the candidate computing power cluster can be obtained using the following formula (1):

[0103] (1)

[0104] in, This represents the green energy availability score of the candidate computing power cluster; This represents the current power generation of the green energy source corresponding to the candidate computing power cluster, i.e., the currently available green energy power. It represents the average green energy power during the estimated execution time of the pending tasks, based on the corresponding predicted power generation. The total power consumption of the candidate computing power cluster when it is running at full load can be obtained by summing the rated power consumption of each computing device in the candidate computing power cluster; This represents a decay factor (0 to 1) used to adjust the level of confidence in future predictions.

[0105] The cluster load score of the candidate computing power cluster can be obtained using the following formula (2):

[0106] (2)

[0107] in, This represents the cluster load score of the candidate computing power cluster; This indicates the amount of resources allocated to the candidate computing power cluster (such as CPU utilization, GPU utilization, memory occupancy, and network bandwidth utilization). This represents the total resources of the candidate computing power cluster.

[0108] When performing weighted summation, weight coefficients can be set for the green energy availability score and the cluster load score respectively. Then, the green energy availability score and the cluster load score are weighted and summed based on their respective weight coefficients to obtain the scheduling score of the corresponding candidate computing power cluster. This scheduling score can enable candidate computing power clusters with abundant green energy power and low cluster load to obtain high scores, guide scheduling jobs to the candidate computing power clusters, and avoid sending jobs to computing power clusters that are about to be saturated, thereby improving job start-up speed and operational stability.

[0109] Understandably, when the supply of green energy electricity is insufficient, Approaching 0, at this point The score is extremely low, which effectively prevents new scheduling jobs from being assigned to the candidate computing power cluster, enabling dynamic adjustment of cluster load and green energy supply.

[0110] In some exemplary embodiments, when obtaining the scheduling score of the candidate computing power cluster by performing a weighted summation process based on the green energy availability score and cluster load score corresponding to the candidate computing power cluster, the following steps may be included: determining the operating cost score of the candidate computing power cluster based on the proportion of the current output power of the green energy corresponding to the candidate computing power cluster in the rated demand power of the candidate computing power cluster; and obtaining the scheduling score of the candidate computing power cluster by performing a weighted summation process based on the green energy availability score, cluster load score, and operating cost score corresponding to the candidate computing power cluster.

[0111] The operating cost score represents the electricity cost of running the job to be processed in the candidate computing power cluster.

[0112] In specific implementation, the operating cost score of the candidate computing power cluster can be calculated using the following formula (3):

[0113] (3)

[0114] in, This represents the operating cost score of the candidate computing power cluster; This indicates the unit cost of green energy electricity (usually very low or zero). This indicates the current unit price of mains electricity; This indicates the proportion of green energy electricity in the candidate computing power clusters. = , This represents the current power generation of the green energy source corresponding to the candidate computing power cluster, i.e., the currently available green energy power. This indicates the rated power requirement of the candidate computing cluster; This represents the highest grid electricity price among all computing power clusters, used for normalization.

[0115] When performing weighted summation, weight coefficients can be set for the green energy availability score, cluster load score, and operating cost score, respectively. Then, the green energy availability score, cluster load score, and operating cost score are weighted and summed based on their respective weight coefficients to obtain the scheduling score of the corresponding candidate computing power cluster. This scheduling score can enable candidate computing power clusters with abundant green energy power, low cluster load, and low operating costs to obtain high scores, thereby improving the absorption rate of green energy in the data center and effectively reducing the operating cost of the data center.

[0116] In some exemplary embodiments, the process of obtaining the scheduling score of the candidate computing power cluster by performing a weighted summation based on the green energy availability score, cluster load score, and operating cost score corresponding to the candidate computing power cluster may include: determining the data transmission cost based on the storage location of the computational data required by the job to be processed; determining the data affinity score corresponding to the candidate computing power cluster based on the data transmission cost; and performing a weighted summation based on the green energy availability score, cluster load score, operating cost score, and data affinity score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0117] The data affinity score represents the distance between the computational data required for the job to be processed and the candidate computing power cluster.

[0118] In specific implementation, the data affinity score of the candidate computing power cluster can be calculated using the following formula (4):

[0119] (4)

[0120] in, This represents the data affinity score of the candidate computing power cluster; This represents the cost of data transmission, which is proportional to the transmission time or transmission cost.

[0121] Specifically, if the large dataset required for the task (such as AI training data) already exists in the storage of a candidate computing cluster, then The value is 0, indicating that the candidate computing power cluster... A score of 1 indicates that if data needs to be transferred across clusters or regions, then due to... Proportional to transmission time or transmission cost, making The decay is due to transmission time or transmission cost.

[0122] When performing weighted summation, weight coefficients can be set for the green energy availability score, cluster load score, operating cost score, and data affinity score. Then, based on their respective weight coefficients, the green energy availability score, cluster load score, operating cost score, and data affinity score are weighted and summed to obtain the scheduling score of the corresponding candidate computing power cluster. This scheduling score allows candidate computing power clusters with abundant green energy power, low cluster load, low operating cost, and close proximity to obtain high scores. This not only improves the absorption rate of green energy in the data center but also effectively reduces the operating cost of the data center. At the same time, it can also reduce the latency and network costs caused by data transmission and improve scheduling efficiency.

[0123] For example, obtaining the scheduling score of the candidate computing power cluster by performing a weighted summation process based on the green energy availability score and cluster load score corresponding to the candidate computing power cluster may include: obtaining a set of weight coefficients; the set of weight coefficients includes the weight coefficients corresponding to each item in the weighted summation process; and performing a weighted summation process on the green energy availability score and cluster load score corresponding to the candidate computing power cluster based on the set of weight coefficients to obtain the scheduling score of the candidate computing power cluster.

[0124] Specifically, the scheduling score of the candidate computing power cluster can be represented by the following formula (5):

[0125] (5)

[0126] in, This indicates the weighting coefficients for each item, which can be set by the administrator according to actual needs (such as cost priority). .

[0127] S209, the job to be processed is distributed to the target computing power cluster so that the target computing power cluster can execute the job to be processed.

[0128] Specifically, if the target computing power cluster is a Kubernetes cluster, the joint scheduling node can convert the description file of the job to be processed into the job format of the Kubernetes API Server and then submit it to the target computing power cluster. If the target computing power cluster is a Slurm cluster, the joint scheduling node can submit the job script of the job to be processed to the controller of the target computing power cluster through the sbatch command.

[0129] The technical solution of this application embodiment identifies non-latency-sensitive jobs and schedules them based on the calculated scheduling score. This effectively transfers massive computing loads to computing power clusters with abundant green energy power, improving the utilization rate of green energy power and the local consumption rate of green energy. In addition, maximizing the use of green energy with zero marginal cost reduces the electricity expenditure of data centers. At the same time, for every additional kilowatt-hour of green electricity used, the dependence on fossil fuel power is reduced accordingly, thereby significantly reducing the operating costs of data centers.

[0130] In some exemplary implementations, such as Figure 3 As shown, the method may further include:

[0131] S211, if the latency sensitivity identification result indicates that the job to be processed is latency sensitive, increase the first weight coefficient in the set of weight coefficients and decrease the second weight coefficient in the set of weight coefficients to obtain the adjusted set of weight coefficients.

[0132] Wherein, the first weighting coefficient is the weighting coefficient corresponding to the cluster load score item, and the second weighting coefficient is the weighting coefficient corresponding to the green energy availability score item.

[0133] S213, based on the adjusted set of weight coefficients, perform weighted summation on the green energy availability score and cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster, and take the candidate computing power cluster with the highest scheduling score as the target computing power cluster.

[0134] Specifically, the set of weight coefficients includes the weight coefficients corresponding to each sub-score item used to calculate the scheduling score. The sum of the weight coefficients corresponding to each sub-score item is 1. Each sub-score item includes the green energy availability score and the cluster load score. It may also include the operating cost score and the data affinity score.

[0135] In the initial set of weighting coefficients, the first weighting coefficient is less than the second weighting coefficient. This prioritizes scheduling "non-latency-sensitive" jobs to the green computing power cluster with the highest current scheduling score, maximizing the utilization of green energy. When a job is latency-sensitive, increasing the first weighting coefficient and decreasing the second weighting coefficient ensures that the first weighting coefficient is greater than the second. This prioritizes selecting the most idle and fastest-responding computing power cluster, regardless of whether it is a green computing power cluster, achieving the goal of "execution as quickly as possible" for latency-sensitive jobs. By identifying and scheduling latency-sensitive jobs, we can ensure that high-priority latency-sensitive jobs receive timely responses while pursuing "green" goals, achieving the dual objectives of "green" and "performance".

[0136] In some exemplary implementations, such as Figure 4 As shown, the method may further include:

[0137] S401, detect the green energy data and cluster load information of each computing cluster.

[0138] S403, based on the detection results, update the current green energy data and current cluster load information of each computing power cluster.

[0139] Specifically, after submitting the jobs to be processed, the joint scheduling node can detect the green energy data and cluster load information of the corresponding computing power clusters through the energy monitoring agent and cluster load monitoring agent of each computing power cluster. Based on the detection results, it updates the current green energy data and current cluster load information of each computing power cluster, thereby achieving dynamic feedback and continuously monitoring the green energy status and computing power cluster load information. When the green energy power supply of a green computing power cluster decreases, its scheduling score will automatically decrease, and the joint scheduling node will automatically stop distributing new jobs to it. This achieves the coupling of computing power load and green energy power supply, enabling computing power scheduling to adapt to changes in green energy power and computing power cluster load, continuously optimizing and matching computing power demand with green energy power supply, and improving the automation and stability of computing power scheduling.

[0140] Corresponding to the computing power scheduling methods provided in the above embodiments, this application also provides a computing power scheduling device. Since the computing power scheduling device provided in this application corresponds to the computing power scheduling methods provided in the above embodiments, the implementation methods of the aforementioned computing power scheduling methods are also applicable to the computing power scheduling device provided in this embodiment, and will not be described in detail in this embodiment.

[0141] Please see Figure 5 The diagram shows a structural schematic of a computing power scheduling device provided in an embodiment of this application. This device has the function of implementing the computing power scheduling method in the above-described method embodiments. This function can be implemented in hardware or by hardware executing corresponding software. Figure 5 As shown, the computing power scheduling device 500 may include:

[0142] The job acquisition module 510 is used to acquire jobs to be processed.

[0143] The latency sensitivity identification module 520 is used to identify the latency sensitivity of the job to be processed and obtain the latency sensitivity identification result of the job to be processed.

[0144] The candidate computing power cluster determination module 530 is used to determine, from multiple computing power clusters, a candidate computing power cluster that matches the resource requirements of the job to be processed when the latency sensitivity identification result indicates that the job to be processed is non-latency sensitive; the multiple computing power clusters include a first type of computing power cluster based on green energy power and a second type of computing power cluster based on grid power.

[0145] The cluster scheduling score determination module 540 is used to determine the scheduling score of each candidate computing power cluster based on the current green energy data and current cluster load information of each candidate computing power cluster, and to select the candidate computing power cluster with the highest scheduling score as the target computing power cluster.

[0146] The job distribution module 550 is used to distribute the job to be processed to the target computing power cluster, so that the target computing power cluster can execute the job to be processed.

[0147] In some exemplary embodiments, the delay sensitivity identification module 520 includes:

[0148] The field content determination module is used to extract the metadata of the job to be processed and determine the field content of the corresponding latency sensitivity identifier field in the metadata;

[0149] The first determining module is used to determine the latency sensitivity identification result of the job to be processed based on the field content of the latency sensitivity identifier field when the field content is not empty.

[0150] The second determining module is used to determine the latency sensitivity identification result of the job to be processed based on the job type and estimated execution time of the job in the metadata when the field content is empty.

[0151] In some exemplary embodiments, the cluster scheduling score determination module 540 includes:

[0152] The green energy availability score determination module is used to determine the green energy availability score corresponding to each candidate computing power cluster based on the current green energy data of the candidate computing power cluster; the current green energy data includes the current power generation and predicted power generation of the green energy corresponding to the candidate computing power cluster; the green energy availability score characterizes the green energy power supply capacity corresponding to the candidate computing power cluster.

[0153] The cluster load score determination module is used to determine the cluster load score corresponding to the candidate computing power cluster based on the current cluster load information of the candidate computing power cluster; the cluster load score represents the idle degree of the candidate computing power cluster.

[0154] The weighted processing module is used to perform weighted summation processing based on the green energy availability score and cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0155] In some exemplary embodiments, the weighting processing module includes:

[0156] The operating cost score determination module is used to determine the operating cost score of the candidate computing power cluster based on the proportion of the current output power of the green energy corresponding to the candidate computing power cluster in the rated power demand of the candidate computing power cluster; the operating cost score represents the electricity cost of running the job to be processed in the candidate computing power cluster.

[0157] The first submodule of weighted processing is used to perform weighted summation processing based on the green energy availability score, cluster load score and operating cost score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0158] In some exemplary embodiments, the weighted processing first submodule includes:

[0159] The data transmission cost determination module is used to determine the data transmission cost based on the storage location of the computational data required for the task to be processed.

[0160] The data affinity score determination module is used to determine the data affinity score corresponding to the candidate computing power cluster based on the data transmission cost; the data affinity score represents the distance between the computational data required by the job to be processed and the candidate computing power cluster.

[0161] The second submodule of weighted processing is used to perform weighted summation processing based on the green energy availability score, cluster load score, operating cost score and data affinity score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0162] In some exemplary embodiments, the weighted processing module is specifically used to: obtain a set of weight coefficients; the set of weight coefficients includes the weight coefficients corresponding to each item in the weighted summation process; based on the set of weight coefficients, perform weighted summation processing on the green energy availability score and cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster.

[0163] In some exemplary embodiments, the device 500 further includes:

[0164] The weight coefficient adjustment module is used to increase the first weight coefficient in the weight coefficient set and decrease the second weight coefficient in the weight coefficient set when the latency sensitivity identification result indicates that the job to be processed is latency sensitive, so as to obtain an adjusted weight coefficient set; the first weight coefficient is the weight coefficient corresponding to the cluster load score item, and the second weight coefficient is the weight coefficient corresponding to the green energy availability score item.

[0165] The summation processing module is used to perform weighted summation processing on the green energy availability score and cluster load score corresponding to the candidate computing power cluster based on the adjusted set of weight coefficients, so as to obtain the scheduling score of the candidate computing power cluster.

[0166] In some exemplary embodiments, the device 500 further includes:

[0167] The detection module is used to detect the green energy data and cluster load information of each computing cluster.

[0168] The update module is used to update the current green energy data and current cluster load information of each computing power cluster based on the detection results.

[0169] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0170] This application provides an electronic device, which includes a processor and a memory. The memory stores at least one instruction or at least one program. The at least one instruction or at least one program is loaded and executed by the processor to implement any of the computing power scheduling methods provided in this application.

[0171] Memory can be used to store software programs and modules. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory. Memory can primarily include a program storage area and a data storage area. The program storage area can store the operating system, application programs required for the functions, etc.; the data storage area can store data created based on the use of the device, etc. Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, memory can also include a memory controller to provide the processor with access to the memory.

[0172] The methods and embodiments provided in this application can be executed in a computer terminal, server, or similar computing device; that is, the aforementioned electronic device may include a computer terminal, server, or similar computing device. Taking running on a server as an example... Figure 6 This is a hardware structure block diagram of a server running a computing power scheduling method according to an embodiment of the present invention, such as... Figure 6As shown, the server 600 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 610 (CPUs 610 may include, but are not limited to, microprocessors (MCUs) or programmable logic devices (FPGAs), a memory 630 for storing data, and one or more storage media 620 (e.g., one or more mass storage devices) for storing application programs 623 or data 622. The memory 630 and storage media 620 may be temporary or persistent storage. The program stored in the storage media 620 may include one or more modules, each module may include a series of instruction operations on the server. Furthermore, the CPU 610 may be configured to communicate with the storage media 620 and execute the series of instruction operations stored in the storage media 620 on the server 600. Server 600 may also include one or more power supplies 660, one or more wired or wireless network interfaces 650, one or more input / output interfaces 640, and / or one or more operating systems 621, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0173] The input / output interface 640 can be used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of server 600. In one example, input / output interface 640 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, input / output interface 640 may be a radio frequency (RF) module used for wireless communication with the Internet.

[0174] Those skilled in the art will understand that Figure 6 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, server 600 may also include... Figure 6 The more or fewer components shown, or having the same Figure 6 The different configurations shown.

[0175] Embodiments of this application also provide a computer-readable storage medium, which can be disposed in an electronic device to store at least one instruction or at least one program related to implementing a computing power scheduling method. The at least one instruction or the at least one program is loaded and executed by the processor to implement any of the computing power scheduling methods provided in the above-described method embodiments.

[0176] Embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the computing power scheduling methods provided in the above-described method embodiments.

[0177] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0178] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0179] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0180] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0181] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A computing power scheduling method, characterized in that, The method includes: Get pending jobs; Extract the metadata of the job to be processed, and determine the field content of the corresponding latency sensitivity identifier field in the metadata; if the field content is not empty, determine the latency sensitivity identification result of the job to be processed based on the field content of the latency sensitivity identifier field; if the field content is empty, determine the latency sensitivity identification result of the job to be processed based on the job type and estimated execution time of the job in the metadata. If the latency sensitivity identification result indicates that the job to be processed is not latency sensitive, a candidate computing power cluster that matches the resource requirements of the job to be processed is determined from multiple computing power clusters; the multiple computing power clusters include a first type of computing power cluster based on green energy power and a second type of computing power cluster based on grid power. For each candidate computing power cluster, a green energy availability score is determined based on the current green energy data of the candidate computing power cluster; the current green energy data includes the current power generation and predicted power generation of the green energy corresponding to the candidate computing power cluster; the green energy availability score represents the green energy power supply capacity of the candidate computing power cluster; a cluster load score is determined based on the current cluster load information of the candidate computing power cluster; the cluster load score represents the idle level of the candidate computing power cluster; a weighted summation is performed on the green energy availability score and the cluster load score of the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster, and the candidate computing power cluster with the highest scheduling score is selected as the target computing power cluster; The pending jobs are distributed to the target computing power cluster so that the target computing power cluster can execute the pending jobs.

2. The method according to claim 1, characterized in that, The weighted summation of the green energy availability score and cluster load score corresponding to the candidate computing power cluster yields the scheduling score of the candidate computing power cluster, which includes: The operating cost score of the candidate computing power cluster is determined based on the current output power of the green energy corresponding to the candidate computing power cluster and its proportion in the rated power demand of the candidate computing power cluster; the operating cost score represents the electricity cost of running the job to be processed in the candidate computing power cluster. The scheduling score of the candidate computing power cluster is obtained by weighted summation of the green energy availability score, cluster load score and operating cost score corresponding to the candidate computing power cluster.

3. The method according to claim 2, characterized in that, The scheduling score of the candidate computing power cluster is obtained by weighted summation of its green energy availability score, cluster load score, and operating cost score. The data transmission cost is determined based on the storage location of the computational data required for the task to be processed; Based on the data transmission cost, a data affinity score is determined for the candidate computing power cluster; the data affinity score represents the distance between the computational data required by the job to be processed and the candidate computing power cluster. The scheduling score of the candidate computing power cluster is obtained by weighted summation of its green energy availability score, cluster load score, operating cost score, and data affinity score.

4. The method according to claim 1, characterized in that, The weighted summation of the green energy availability score and cluster load score corresponding to the candidate computing power cluster yields the scheduling score of the candidate computing power cluster, which includes: Obtain a set of weight coefficients; the set of weight coefficients includes the weight coefficients corresponding to each item in the weighted summation process. Based on the set of weight coefficients, the green energy availability score and cluster load score corresponding to the candidate computing power cluster are weighted and summed to obtain the scheduling score of the candidate computing power cluster.

5. The method according to claim 4, characterized in that, The method further includes: If the latency sensitivity identification result indicates that the job to be processed is latency sensitive, the first weight coefficient in the set of weight coefficients is increased and the second weight coefficient in the set of weight coefficients is decreased to obtain the adjusted set of weight coefficients; the first weight coefficient is the weight coefficient corresponding to the cluster load score item, and the second weight coefficient is the weight coefficient corresponding to the green energy availability score item. Based on the adjusted set of weight coefficients, the green energy availability score and cluster load score corresponding to the candidate computing power cluster are weighted and summed to obtain the scheduling score of the candidate computing power cluster. The candidate computing power cluster with the highest scheduling score is selected as the target computing power cluster.

6. The method according to claim 4, characterized in that, After distributing the jobs to be processed to the target computing power cluster, the method further includes: The green energy data and cluster load information of each computing cluster are detected; Based on the detection results, update the current green energy data and current cluster load information of each computing power cluster.

7. A computing power scheduling device, characterized in that, The device includes: The job acquisition module is used to acquire jobs to be processed. The latency sensitivity identification module is used to extract the metadata of the job to be processed, determine the field content of the corresponding latency sensitivity identifier field in the metadata; if the field content is not empty, determine the latency sensitivity identification result of the job to be processed based on the field content of the latency sensitivity identifier field; if the field content is empty, determine the latency sensitivity identification result of the job to be processed based on the job type and estimated execution time of the job in the metadata. The candidate computing power cluster determination module is used to determine, when the latency sensitivity identification result indicates that the job to be processed is non-latency sensitive, a candidate computing power cluster that matches the resource requirements of the job to be processed from multiple computing power clusters; the multiple computing power clusters include a first type of computing power cluster based on green energy power and a second type of computing power cluster based on grid power. The cluster scheduling score determination module is used to determine the green energy availability score corresponding to each candidate computing power cluster based on the current green energy data of the candidate computing power cluster; the current green energy data includes the current power generation and predicted power generation of the green energy corresponding to the candidate computing power cluster; the green energy availability score represents the green energy power supply capacity corresponding to the candidate computing power cluster; based on the current cluster load information of the candidate computing power cluster, the module determines the cluster load score corresponding to the candidate computing power cluster; the cluster load score represents the idle degree of the candidate computing power cluster; and a weighted summation is performed on the green energy availability score and the cluster load score corresponding to the candidate computing power cluster to obtain the scheduling score of the candidate computing power cluster, and the candidate computing power cluster with the highest scheduling score is selected as the target computing power cluster. The job distribution module is used to distribute the job to be processed to the target computing power cluster, so that the target computing power cluster can execute the job to be processed.

Images