Computing power resource scheduling method and device, equipment, storage medium and program product

Abstract

The invention provides a computing power resource scheduling method and device, equipment, a storage medium and a program product, and relates to the technical field of computing network integration resource scheduling. The method comprises the following steps: determining a resource demand corresponding to a computing power task; and scheduling the computing power nodes for the computing power task based on the computing power basic performance data and the computing power real-time performance data of the plurality of computing power nodes and the resource demand. According to the method, the computing power basic performance data and the computing power real-time performance data are combined to schedule computing network resources, so that the matching between the scheduled computing power nodes and the computing power tasks can be improved, the utilization rate of the resources can be improved, and the waste of the resources is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computing and network integrated resource scheduling, and in particular to a computing resource scheduling method and device, equipment, storage medium and program product. BACKGROUND

[0002] The purpose of computing and network integrated resource scheduling is to select nodes with similar performance and sufficient resources to support business operation. If a low-performance node can meet the business demand, selecting a high-performance node will cause resource waste and low resource utilization efficiency. Therefore, computing and network integrated resource scheduling is used to avoid resource waste and improve resource utilization efficiency. The computing power index design relied on by the current computing and network integrated computing power resource scheduling method is relatively single, and there is a problem of low matching between the scheduled computing power nodes and the business demand. SUMMARY

[0003] The embodiments of the present application provide a computing power resource scheduling method, device, equipment, storage medium and program product, which are used to solve the technical problem of low matching between the scheduled computing power nodes and the business demand in the prior art.

[0004] In a first aspect, the embodiments of the present application provide a computing power resource scheduling method, comprising: determining resource demand corresponding to a computing power task; and scheduling a computing power node for the computing power task based on the resource demand and the respective computing power basic performance data and computing power real-time performance data of a plurality of computing power nodes.

[0005] In some embodiments, the resource demand includes computing power basic performance demand and computing power real-time performance demand; and the scheduling of the computing power node for the computing power task based on the resource demand and the respective computing power basic performance data and computing power real-time performance data of the plurality of computing power nodes includes: determining a target computing power node group from a plurality of computing power node groups based on the computing power basic performance demand; wherein the plurality of computing power node groups are obtained based on the respective computing power basic performance data of the plurality of computing power nodes, and each computing power node group includes a plurality of computing power nodes; determining a target computing power node matched with the computing power task based on the respective computing power real-time performance data of the plurality of computing power nodes in the target computing power node group and the computing power real-time performance demand; and scheduling the target computing power node to execute the computing power task.

[0006] In some embodiments, the plurality of computing power node groups are obtained by: determining a basic ability score of each computing power node according to the respective computing power basic performance data of the plurality of computing power nodes; and dividing the plurality of computing power nodes into the plurality of computing power node groups based on the basic ability score.

[0007] As a possible implementation manner, the basic performance data of each computing power node includes respective index values of a plurality of basic performance indexes; and the determining of the basic capability score of each computing power node based on the basic performance data of each computing power node includes: determining a weight value of each basic performance index of the plurality of computing power nodes based on entropy weight method; and determining the basic capability score of each computing power node based on the weight value of each basic performance index and the index value of each basic performance index corresponding to the computing power node.

[0008] As a possible implementation manner, the dividing of the plurality of computing power nodes into the plurality of computing power node groups based on the basic capability scores includes: dividing the plurality of computing power nodes into the plurality of computing power node groups based on a decision tree algorithm with the basic capability scores as characteristic values.

[0009] In some embodiments, the determining of the target computing power node matched with the computing power task based on the real-time performance data of each computing power node in the target computing power node group and the real-time performance requirement includes: determining a plurality of first computing power nodes in the target computing power group that meet the task requirement based on the real-time performance data of each computing power node in the target computing power node group and the real-time performance requirement; determining a distance value between the real-time performance data of each first computing power node and the real-time performance requirement; and determining the first computing power node with the minimum distance value as the target computing power node.

[0010] In a second aspect, an embodiment of the present application provides a computing power resource scheduling device, which includes: a first determining module configured to determine resource requirements corresponding to a computing power task; and a scheduling module configured to schedule computing power nodes for the computing power task based on basic performance data and real-time performance data of a plurality of computing power nodes and the resource requirements.

[0011] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor and a memory storing a computer program, and the processor implements the above-mentioned computing power resource scheduling method when executing the program.

[0012] In a fourth aspect, an embodiment of the present application provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above-mentioned computing power resource scheduling method.

[0013] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is executed by a processor to implement the above-mentioned computing power resource scheduling method.

[0014] This invention provides a computing power resource scheduling method, apparatus, device, storage medium, and program product, which determines the resource requirements corresponding to a computing power task; based on the basic computing power performance data and real-time computing power performance data of multiple computing power nodes, as well as the resource requirements, it schedules computing power nodes for the computing power task. This invention combines basic computing power performance data and real-time computing power performance data for computing network resource scheduling, which can improve the matching between scheduled computing power nodes and computing power tasks, thereby improving resource utilization and avoiding resource waste. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is one of the flowcharts illustrating the computing resource scheduling method provided in an embodiment of the present invention; Figure 2 This is a second flowchart illustrating the computing resource scheduling method provided in an embodiment of the present invention. Figure 3 The third flowchart illustrates the computing resource scheduling method provided in this embodiment of the invention. Figure 4 The fourth flowchart illustrates the computing resource scheduling method provided in this embodiment of the invention. Figure 5 The fifth flowchart illustrates the computing resource scheduling method provided in this embodiment of the invention. Figure 6 This is a schematic diagram of the computing resource scheduling device provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0018] It should be noted that with the development of cloud computing and cloud-native technologies, the traditional storage, computing, and network separation model in computing power architecture is gradually transforming into a deep integration of storage, computing, and network. Resource scheduling has changed from single performance scheduling to a multi-dimensional computing power scheduling system. Moreover, with the continuous emergence of scenarios such as smart cities, smart transportation, and smart industry, the demand for heterogeneous computing power is gradually increasing, showing a comprehensive development trend of heterogeneous computing power resources such as cloud, edge, and terminal CPUs and GPUs. The computing power resource scheduling method is the key foundation for the efficient operation of computing network convergence scenarios.

[0019] It's important to note that the purpose of computing-network converged resource scheduling is to select nodes with similar performance and sufficient resources to support business operations. If low-performance nodes can already meet business needs, selecting high-performance nodes would result in resource waste and low resource utilization efficiency. Therefore, computing-network converged resource scheduling aims to avoid resource waste and improve resource utilization efficiency. Currently, the computing power resource scheduling methods for computing-network convergence rely on relatively simple computing power metrics, leading to a problem of poor matching between scheduled computing power nodes and business requirements.

[0020] To address the aforementioned issues, embodiments of the present invention provide a computing resource scheduling method, apparatus, device, storage medium, and program product.

[0021] Figure 1 This is one of the flowcharts illustrating the computing resource scheduling method provided in an embodiment of the present invention. Figure 1 As shown, the method may include the following steps.

[0022] Step 101: Determine the resource requirements corresponding to the computing power task.

[0023] In some embodiments, upon receiving a computing task submitted by a business party, the resource requirements corresponding to the computing task are determined based on various characteristic parameters of the computing task, such as computational complexity, data size, task type, and priority. These resource requirements may include the number of CPU (Central Processing Unit) cores, the number of GPUs (Graphics Processing Units), and the amount of memory.

[0024] In some embodiments, the resource requirements corresponding to the computing power task can also be received from the business party. That is, the business party evaluates the relevant characteristic parameters of the computing power task to obtain the resource requirements corresponding to the computing power task, and sends them to the execution device of the computing power resource scheduling method of the present invention.

[0025] Step 102: Based on the basic computing power performance data and real-time computing power performance data of each of the multiple computing power nodes, as well as the resource requirements, schedule computing power nodes for the computing power task.

[0026] In this context, computing nodes refer to high-performance computing devices with powerful processing and storage capabilities, capable of handling large-scale computing tasks. Multiple computing nodes here can refer to multiple computing nodes within a single cluster, or computing nodes across multiple clusters.

[0027] In some embodiments, the basic computing performance data of each computing node is used to characterize the inherent performance level of the computing resources of that computing node. As an example, the basic computing performance data of a computing node may include CPU MIPS (Million Instructions Per Second), GPU video memory, AI computing power metric GOPS (GigaOperations Per Second), storage size, network bandwidth, etc. The real-time computing performance data of a computing node is used to characterize the real-time carrying capacity of the computing resources of that computing node. As an example, the real-time computing performance data of a computing node may include the CPU idle rate, GPU idle rate, remaining storage, network throughput, etc.

[0028] It is understandable that the basic computing power performance data and real-time computing power performance data of computing power nodes can characterize the computing power of computing power nodes from different aspects. At the same time, the resource requirements corresponding to computing power tasks also include both basic computing power performance requirements and real-time computing power performance requirements. In order to more accurately determine the matching degree between computing power nodes and computing power tasks and improve the adaptability of computing power nodes and computing power tasks, computing power nodes can be scheduled by combining the basic computing power performance data and real-time computing power performance data of computing power nodes.

[0029] In some embodiments, the process of scheduling computing nodes for a computing task based on the basic computing power performance data and real-time computing power performance data of each of the multiple computing power nodes, as well as the resource requirements, may include: comparing the resource requirements corresponding to the computing task with the basic computing power performance data and real-time computing power performance data of each computing power node, and determining multiple first computing power nodes that can meet the resource requirements from the multiple computing power nodes; determining the difference between the computing power capability and the resource requirements of each first computing power node based on the basic computing power performance data and real-time computing power performance data of each of the multiple first computing power nodes, and taking the first computing power node with the smallest difference as the target computing power node; and scheduling the target computing power node to execute the computing task.

[0030] As one possible implementation, the basic computing power performance data of a computing power node can include the values ​​of multiple basic computing power performance indicators, and the real-time computing power performance data of a computing power node can include the values ​​of multiple real-time computing power performance indicators. Based on the resource requirements corresponding to the computing power task, the required values ​​of multiple basic computing power performance indicators and multiple real-time computing power performance indicators corresponding to the computing power task can also be determined. Based on the basic computing power performance data and real-time computing power performance data of multiple computing power nodes, as well as resource requirements, the process of scheduling computing power nodes for computing power tasks may include: for each computing power node, based on the index values ​​of multiple basic computing power performance indicators corresponding to that computing power node, and the index values ​​of multiple real-time computing power performance indicators that may be included in the computing power performance data of the computing power node, a weighted calculation is performed according to a preset weight value to obtain the computing power capability value of each computing power node; based on the requirement values ​​of multiple basic computing power performance indicators corresponding to the computing power task, and the requirement values ​​of multiple real-time computing power performance indicators, the computing power requirement value corresponding to the computing power task is determined; the difference between the computing power requirement value and the computing power capability value of each computing power node is determined, and the computing power node with the smallest difference is determined as the target computing power node.

[0031] Since the computing power resource scheduling method of this invention combines the basic computing power performance data and real-time computing power performance data of computing power nodes, it can more comprehensively determine the matching degree between computing power nodes and computing power tasks, so as to allocate more matching computing power nodes to computing power tasks, thereby avoiding resource waste and improving the utilization rate of computing network resources.

[0032] According to an embodiment of the present invention, a computing power resource scheduling method is provided, which determines the resource requirements corresponding to a computing power task; and schedules computing power nodes for a computing power task based on the basic computing power performance data and real-time computing power performance data of multiple computing power nodes, as well as the resource requirements. This invention combines basic computing power performance data and real-time computing power performance data for computing network resource scheduling, which can improve the matching between scheduled computing power nodes and computing power tasks, thereby improving resource utilization and avoiding resource waste.

[0033] Figure 2 This is the second flowchart illustrating the computing resource scheduling method provided in this embodiment of the invention. In this embodiment, the resource requirements corresponding to a computing task can include basic computing performance requirements and real-time computing performance requirements. Basic computing performance requirements refer to the inherent performance level requirements of the computing task for the computing resources, such as the requirement for CPU MIPS, storage space, and GPU memory. Real-time computing performance requirements refer to the requirements of the computing task for the real-time computing capabilities of the computing resources, such as the requirement for the number of CPU cores and network throughput. Figure 2 As shown, based on the above embodiments,Figure 1 The implementation process of step 102 may include the following steps.

[0034] Step 201: Based on the basic computing power performance requirements, determine the target computing power node group from multiple computing power node groups; wherein, multiple computing power node groups are obtained by dividing the basic computing power performance data of multiple computing power nodes, and each computing power node group contains multiple computing power nodes.

[0035] In other words, based on the basic performance data of each computing node, multiple computing nodes are divided into multiple computing node groups in advance. Each computing node group contains multiple computing nodes. When scheduling computing nodes for computing tasks, a target computing node group that meets the basic performance requirements is first determined from the multiple computing node groups based on the basic performance requirements. Then, a target computing node that matches the real-time performance requirements of the computing power is determined from the target computing node group. This can greatly reduce the resource search range in the computing resource scheduling process and improve scheduling efficiency.

[0036] In some embodiments, multiple computing nodes within each computing node group possess similar basic computing power performance. As an example, the number of computing node groups and the conditions that the basic computing power performance data of the computing nodes within each computing node group must meet can be preset. Based on the basic computing power performance data of each computing node and the basic computing power performance data conditions corresponding to each computing node group, multiple computing nodes are divided into multiple computing node groups. Based on the basic computing power performance data conditions of the computing nodes within each computing node group, a target computing node group containing computing nodes that meet the basic computing power performance requirements can be determined from the multiple computing node groups.

[0037] In other embodiments, the basic computing power performance of each computing node can be characterized by a basic capability score. This basic capability score can be determined using the basic computing power performance data of each computing node, and computing node groups can be divided based on these scores. For example, the basic computing power performance data of a computing node can include the values ​​of multiple basic computing power performance indicators, with a pre-defined weight value for each indicator. Based on the basic computing power performance data and the weight values ​​of each indicator, the basic capability score of each computing node can be determined. Based on the basic capability score range corresponding to each computing node group, each computing node is assigned to a computing node group with a matching basic capability score. The basic computing power performance requirements can also include the requirement values ​​of multiple basic computing power performance indicators. A weighted calculation combining the weight values ​​of each indicator yields the basic capability requirement score for the computing power task. The basic capability score range to which the basic capability requirement score falls is determined, and the corresponding computing node group is designated as the target computing node group.

[0038] Step 202: Based on the real-time computing power performance data of each computing power node in the target computing power node group and the real-time computing power performance requirements, determine the target computing power node that matches the computing power task.

[0039] In other words, based on the real-time performance data of each of the multiple computing nodes within the target computing node group, the target computing node that matches the real-time performance requirements is determined.

[0040] In some embodiments, the real-time performance data of a computing node may include the values ​​of multiple real-time performance indicators, such as CPU idle rate, GPU idle rate, remaining storage, and network throughput. The real-time performance requirements may also include the required values ​​of multiple real-time performance requirement indicators, such as the required number of CPU cores, required storage space, and required number of GPUs. The process of determining the target computing node matching the computing task based on the real-time performance data of multiple computing nodes within the target computing node group and the real-time performance requirements may include: normalizing the real-time performance indicators of multiple computing nodes within the target computing node group and the indicators in the basic computing performance requirements to unify their dimensions; calculating the real-time capability value of each computing node within the target computing node group and the real-time capability requirement value of the computing task based on the normalized indicators and preset indicator weight values; determining the difference between the real-time capability value and the real-time capability requirement value of each computing node within the target computing node group; and determining the computing node with the smallest difference as the target computing node.

[0041] Step 203: Schedule the target computing node to execute the computing task.

[0042] According to the computing power resource scheduling method of this invention, a target computing power node group is determined from multiple computing power node groups based on the basic computing power performance requirements. The multiple computing power node groups are obtained by dividing the data based on the basic computing power performance data of each computing power node, and each computing power node group contains multiple computing power nodes. Based on the real-time computing power performance data of each computing power node within the target computing power node group, and the real-time computing power performance requirements, a target computing power node matching the computing power task is determined. The target computing power node is then scheduled to execute the computing power task. This invention divides multiple computing power nodes into multiple computing power node groups based on basic computing power performance data. During computing power resource scheduling calculations, calculations are performed only on the computing power nodes within the target computing power node group matching the computing power task, which can significantly reduce the computational load and improve the efficiency of resource scheduling.

[0043] Next, we will provide a detailed introduction to the division method of computing node groups.

[0044] Figure 3This is the third flowchart illustrating the computing resource scheduling method provided in this embodiment of the invention. Figure 3 As shown, multiple computing node groups are obtained through the following steps.

[0045] Step 301: Determine the basic capability score of each computing power node based on the basic computing power performance data of each of the multiple computing power nodes.

[0046] The basic capability score of a computing node is used to characterize its basic computing power performance level. The higher the basic capability score, the higher the basic computing power performance level of the computing node.

[0047] In some embodiments, the basic computing power performance data of a computing power node may include the values ​​of multiple basic computing power performance indicators, each of which affects the basic computing power performance of the computing power node. A weight value for the impact of each basic computing power performance indicator on the basic computing power performance is pre-defined. For each computing power node, a weighted sum is performed based on the values ​​of the multiple basic computing power performance indicators and the weight value of each indicator to obtain the basic capability score of the computing power node. The weight value of each basic computing power performance indicator can also be determined based on the entropy weight method.

[0048] In other embodiments, the values ​​of multiple basic computing power performance indicators can be normalized to a unified dimension. For each computing power node, the normalized indicators corresponding to that computing power node are averaged, and the result is used as the basic capability score of that computing power node.

[0049] Step 302: Based on the basic capability score, divide the multiple computing power nodes into multiple computing power node groups.

[0050] In some embodiments, the number of computing node groups and the basic capability score range corresponding to each computing node group can be preset. Based on the basic capability score of each computing node and the basic capability score range corresponding to each computing node group, multiple computing nodes are divided into multiple computing node groups. The basic capability score range corresponding to each computing node group can also be determined based on a decision tree algorithm.

[0051] In other embodiments, clustering can also be performed based on the basic capability score of each computing node to group computing nodes with similar basic capability scores into the same computing node group, and multiple computing node groups can be obtained based on the clustering results.

[0052] According to the computing power resource scheduling method of this invention, a basic capability score is determined for each computing power node based on its respective basic computing power performance data. Based on the basic capability score, the multiple computing power nodes are divided into multiple computing power node groups. A target computing power node group is then determined from within these groups based on basic computing power performance requirements. Finally, a target computing power node is determined from among the multiple computing power nodes within the target computing power node group based on real-time computing power performance requirements. This invention divides multiple computing power nodes into multiple computing power node groups based on basic computing power performance data. During computing power resource scheduling calculations, calculations are performed only on the computing power nodes within the target computing power node group that match the computing power task, which can significantly reduce the computational load and improve the efficiency of resource scheduling.

[0053] Figure 4 This is the fourth flowchart illustrating the computing resource scheduling method provided in this embodiment of the invention. In some embodiments of the invention, the basic computing performance data of a computing node may include the values ​​of multiple basic computing performance indicators, which may include the CPU's MIPS, GPU memory, storage size, network bandwidth, GOPS, etc. Figure 4 As shown, multiple computing node groups are obtained through the following steps.

[0054] Step 401: Analyze the basic computing power performance data of each of the multiple computing power nodes based on the entropy weight method, and determine the weight value of each basic computing power performance index.

[0055] Among them, the entropy weight method is a weight determination method based on the principle of information entropy. It assigns weights according to the degree of difference of each indicator, thereby obtaining the corresponding weights of each indicator and the comprehensive evaluation score. It fully considers the objective information of each indicator and can more accurately reflect the contribution of each indicator to the whole. At the same time, the calculation results of the entropy weight method have good stability and reliability, and can effectively avoid errors in weight determination caused by subjective factors.

[0056] In some embodiments, it can be assumed that the number of computing nodes is m and the number of basic computing performance indicators is n. Therefore, the basic computing performance data of each of the multiple computing nodes can be represented by the matrix of the following equation (1).

[0057] (1) in, This represents the value of the j-th basic performance index of the i-th computing power node. For a certain basic performance index, the greater the difference between different computing power nodes, the greater the role that basic performance index plays in the entire evaluation process. When the values ​​of the same basic performance index are equal among different computing power nodes, that index has no effect and its value is low.

[0058] It is understandable that the values ​​of the above-mentioned basic computing power performance indicators are all positive numbers and the larger the better. However, because the two performance indicators are different and have different value ranges, in order to make the calculation reasonable and scientific, the basic computing power performance data can be normalized. For example, the index value of each basic computing power performance indicator can be mapped to the [0,1] interval by using the linear normalization method. The normalization calculation formula is shown in the following formula (2).

[0059] (2) in, for The normalized result is the normalized value of the j-th basic performance index of the i-th computing power node. It represents the minimum value of the basic performance index of the j-th computing power among multiple computing power nodes; Let be the maximum value of the j-th basic performance index of computing power among multiple computing power nodes. After normalization, the data matrix obtained by equation (1) above is shown in equation (3) below.

[0060] (3) The weight of the i-th computing node is calculated based on the following formula (4) for the j-th basic computing power performance index. ,and The value range of is [0,1], and the weight matrix corresponding to matrix (3) is shown in equation (5) below.

[0061] (4) (5) The entropy value of each computing power basic performance index is calculated based on matrix (5), and the calculation process is shown in equation (6) below.

[0062] (6) in, Let be the entropy value of the j-th basic performance index of computing power. Based on the entropy value of each basic performance index of computing power, the degree of difference of each basic performance index of computing power is calculated based on the following formula (7).

[0063] (7) in, Let represent the degree of difference in the j-th basic computing power performance index. The entropy weight method can calculate the weight value of each index based on the degree of difference in the index values ​​of each index, as shown in the following formula (8).

[0064] (8) in, Let be the weight value of the j-th basic computing power performance index. In other words, based on the above calculation process, the weight value of each basic computing power performance index is obtained.

[0065] Step 402: For each computing power node, determine the basic capability score of the computing power node based on the weight value of each basic computing power performance indicator and the index value of each basic computing power performance indicator corresponding to the computing power node.

[0066] In other words, the basic capability score of each computing power node is obtained by weighting and summing the individual index values ​​and weight values ​​of multiple basic computing power performance indicators.

[0067] In some embodiments, the process of determining the basic capability score of a computing power node based on the weight value of each basic computing power performance index and the index value of each basic computing power performance index corresponding to the computing power node can be shown in the following formula (9).

[0068] (9) in, Let be the basic capability score of the i-th computing node.

[0069] Step 403: Based on the decision tree algorithm, the basic capability score is used as the feature value to divide the multiple computing power nodes into multiple computing power node groups.

[0070] In some embodiments, the grouping of computing power nodes can be based on the CART tree algorithm. The CART tree algorithm is a classic decision tree algorithm that can be used for both classification and regression tasks. In classification tasks, CART trees can not only perform binary classification but also achieve multi-class classification by constructing multiple binary trees. CART uses the Gini index as the basis for splitting. The calculation of the Gini index only requires four arithmetic operations, eliminating the need for logarithmic calculations, which significantly improves computational efficiency. CART trees automatically divide the basic capabilities of computing power nodes into intervals by calculating the Gini index. Compared to subjective interval division, CART trees can automatically divide based on data features, ensuring the objectivity and reliability of the division results.

[0071] In some embodiments, the number of computing node groups K can be preset. Based on the CART algorithm, the basic capability scores of multiple computing nodes are divided into K intervals according to their respective basic capability scores. Each computing node group corresponds to a basic capability score interval, thereby dividing multiple computing nodes into K computing node groups.

[0072] In some embodiments, the number of multi-computing node groups is K. The basic capability scores of multiple computing nodes are arranged in ascending order. The multiple computing nodes are gradually split based on the CART tree algorithm. Each split is performed by selecting the splitting method with the smallest Gini index until the number of groups after splitting reaches K. The formula for calculating the Gini index corresponding to each splitting method is shown in the following formula (10).

[0073] (10) in, The Gini index corresponds to a certain type of splitting. It is a set determined by multiple computing nodes; This represents the sample subset corresponding to the k-th computing node group; K is the number of computing node groups.

[0074] Taking the first split as an example, the basic capability score is used as the feature value. Based on the range of the basic capability scores of the computing nodes in D, the computing node set D can be divided into two parts, D1 and D2. The sets D1 and D2 can be represented by the following equation (11). Based on the obtained D1 and D2, the split continues until the number of subsets is consistent with the number of computing node groups. Each subset is equivalent to a computing node group, and each computing node group corresponds to a basic capability score range.

[0075] (11) in, for Any computing node in the system, i.e. Elements in; The basic capability score for computing node t; The corresponding basic ability score range is , The corresponding basic ability score range is .

[0076] In some embodiments, based on the described embodiments, Figure 2 Step 201, which involves determining the target computing power node group from multiple computing power node groups based on the basic computing power performance requirements, may include: determining the required values ​​of multiple basic computing power performance indicators based on the basic computing power performance requirements; performing a weighted summation based on the weight values ​​of each basic computing power performance indicator determined in step 401 and the required values ​​of multiple basic computing power performance indicators to determine the basic capability requirement score corresponding to the computing power task; and determining the computing power node group into which the basic capability requirement score falls based on the basic capability score range determined in step 403, and using it as the target computing power node group.

[0077] According to the computing power resource scheduling method of this invention, the basic computing power performance data of multiple computing power nodes are analyzed based on the entropy weight method to determine the weight value of each basic computing power performance indicator. For each computing power node, based on the weight value of each basic computing power performance indicator and the indicator value of each basic computing power performance indicator corresponding to the computing power node, the basic capability score of the computing power node is determined. Based on the decision tree algorithm, the multiple computing power nodes are divided into multiple computing power node groups using the basic capability score as the feature value. This invention introduces the entropy weight method to make the obtained weight value of each basic computing power performance indicator more objective, thereby making the obtained basic capability score of each computing power node more representative of the basic computing power performance level of the computing power node. In addition, the division of computing power node groups based on the decision tree algorithm can make the division of computing power node groups more accurate and improve scheduling efficiency.

[0078] Figure 5 This is the fifth flowchart illustrating the computing resource scheduling method provided in this embodiment of the invention. Figure 5 As shown, based on the above embodiments, Figure 2 The implementation process of step 202 may include the following steps.

[0079] Step 501: Based on the real-time performance data of each computing node in the target computing power node group and the real-time computing power requirements, determine the first computing power nodes in the target computing power group that meet the task requirements.

[0080] In some embodiments, the real-time computing power requirement corresponding to the computing power task may include the required values ​​of different real-time computing power performance indicators. The real-time computing power performance data of each computing power node may include the indicator values ​​of different real-time computing power performance indicators. The indicator values ​​of the real-time computing power performance indicators of each of the multiple computing power nodes in the target computing power node group are compared with the required values ​​of the real-time computing power performance indicators. The computing power node whose indicator values ​​are all greater than the required values ​​is regarded as the first computing power node that meets the task requirements.

[0081] In some embodiments, since different indicators have different units of measurement, it is necessary to first normalize the indicators to unify their dimensions, and then perform the above comparison process based on the normalized values.

[0082] Step 502: Determine the distance between the real-time performance data of each first computing node and the real-time computing requirement.

[0083] In some embodiments, for each first computing power node, the distance between the real-time computing power performance data of the first computing power node and the real-time computing power requirement can be determined by the difference between the index values ​​of each of the multiple real-time computing power performance indicators of the first computing power node and the requirement values ​​of each of the multiple real-time computing power performance indicators in the real-time computing power performance requirement.

[0084] As one possible implementation, for each first computing power node, the difference between the index value and the demand value of each computing power real-time performance index is accumulated to obtain the distance value between the computing power real-time performance data and the computing power real-time demand of the first computing power node. Specifically, it can be calculated based on the following formula (12).

[0085] (12); in, This represents the distance between the first computing node t and the real-time computing performance requirements of computing task s. This is the normalized result corresponding to the real-time performance index value of the i-th computing power node t; is the normalized result corresponding to the required value of the i-th real-time performance index in the real-time performance requirements of computing power task s; N is the number of real-time performance indexes.

[0086] Step 503: The first computing power node with the smallest distance value is determined as the target computing power node.

[0087] It is understandable that the distance value can characterize the difference between the performance of the computing node and the requirements of the computing task. Therefore, the performance of the first computing node with the smallest distance value is closest to the requirements of the computing task. In order to avoid waste of resources and make efficient use of node resources, the first computing node with the smallest distance value is determined as the target computing node.

[0088] According to the computing power resource scheduling method of this invention, based on the real-time performance data of multiple computing power nodes within a target computing power node group and their real-time computing power requirements, multiple first computing power nodes within the target computing power group that meet the task requirements are determined; the distance value between the real-time performance data of each first computing power node and its real-time computing power requirements is determined; and the first computing power node with the smallest distance value is determined as the target computing power node. This invention determines the target computing power node from the target computing power node group based on the minimum distance value, and schedules the target computing power node to execute computing power tasks, which can avoid resource waste and improve the utilization rate of computing power resources.

[0089] To implement the above embodiments, the present invention also provides a computing resource scheduling device.

[0090] Figure 6 This is a schematic diagram of the computing resource scheduling device provided in an embodiment of the present invention. Figure 6 As shown, the device includes a first determining module 610 and a scheduling module 620. The first determining module 610 is used to determine the resource requirements corresponding to the computing power task; the scheduling module 620 is used to schedule computing power nodes for the computing power task based on the basic computing power performance data and real-time computing power performance data of each of the multiple computing power nodes, as well as the resource requirements.

[0091] In some embodiments, resource requirements include basic computing power performance requirements and real-time computing power performance requirements; the scheduling module 620 is specifically used to: determine a target computing power node group from multiple computing power node groups based on the basic computing power performance requirements; wherein, the multiple computing power node groups are obtained by dividing the data based on the basic computing power performance data of each of the multiple computing power nodes, and each computing power node group contains multiple computing power nodes; determine a target computing power node that matches the computing power task based on the real-time computing power performance data of each of the multiple computing power nodes in the target computing power node group, and the real-time computing power performance requirements; and schedule the target computing power node to execute the computing power task.

[0092] In some embodiments, the device further includes a second determining module 630. The second determining module 630 is specifically configured to: determine the basic capability score of each computing power node based on the basic computing power performance data of each of the multiple computing power nodes; and divide the multiple computing power nodes into multiple computing power node groups based on the basic capability scores.

[0093] In some embodiments, the basic computing power performance data of each computing power node includes the index values ​​of multiple basic computing power performance indicators; the second determining module 630 is further configured to: analyze the basic computing power performance data of multiple computing power nodes based on the entropy weight method to determine the weight value of each basic computing power performance indicator; and for each computing power node, determine the basic capability score of the computing power node based on the weight value of each basic computing power performance indicator and the index value of each basic computing power performance indicator corresponding to the computing power node.

[0094] In some embodiments, the second determining module 630 is further configured to: divide multiple computing power nodes into multiple computing power node groups based on a decision tree algorithm and using basic capability scores as feature values.

[0095] In some embodiments, the scheduling module 620 is further configured to: determine multiple first computing power nodes in the target computing power group that meet the task requirements based on the real-time computing power performance data of each of the multiple computing power nodes in the target computing power node group and the real-time computing power requirements; determine the distance value between the real-time computing power performance data of each first computing power node and the real-time computing power requirements; and determine the first computing power node with the smallest distance value as the target computing power node.

[0096] According to an embodiment of the computing power resource scheduling device of the present invention, a first determining module is used to determine the resource requirements corresponding to a computing power task; a scheduling module is used to schedule computing power nodes for a computing power task based on the basic computing power performance data and real-time computing power performance data of multiple computing power nodes, as well as the resource requirements. The present invention combines basic computing power performance data and real-time computing power performance data to perform computing network resource scheduling, which can improve the matching between scheduled computing power nodes and computing power tasks, thereby improving resource utilization and avoiding resource waste.

[0097] It should be noted that the explanations and descriptions of the computing resource scheduling method in the above embodiments also apply to the computing resource scheduling device in the embodiments of the present invention, and will not be repeated here.

[0098] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call the computer program in the memory 730 to execute the steps of the computing resource scheduling method in the above embodiments.

[0099] For example, the method includes: determining the resource requirements corresponding to the computing power task; scheduling computing power nodes for the computing power task based on the basic computing power performance data and real-time computing power performance data of each of the multiple computing power nodes, as well as the resource requirements.

[0100] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] On the other hand, embodiments of the present invention also provide a computer program product, the computer program product including a computer program, the computer program being stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute the steps of the computing resource scheduling method provided in the above embodiments.

[0102] On the other hand, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing a computer program, the computer program being used to cause a processor to execute the steps of the computing resource scheduling method provided in the above embodiments.

[0103] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0104] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for scheduling computing resources, characterized in that, include: Determine the resource requirements corresponding to the computing power task; Based on the basic computing power performance data and real-time computing power performance data of each of the multiple computing power nodes, as well as the resource requirements, computing power nodes are scheduled for the computing power task.

2. The method according to claim 1, characterized in that, The resource requirements include basic computing power performance requirements and real-time computing power performance requirements; the scheduling of computing power nodes for the computing power task based on the basic computing power performance data and real-time computing power performance data of multiple computing power nodes, and the resource requirements, includes: Based on the aforementioned basic computing power performance requirements, a target computing power node group is determined from multiple computing power node groups; wherein, the multiple computing power node groups are obtained by dividing the computing power basic performance data of each of the multiple computing power nodes, and each computing power node group contains multiple computing power nodes; Based on the real-time computing power performance data of each of the multiple computing power nodes in the target computing power node group, and the real-time computing power performance requirements, the target computing power node that matches the computing power task is determined. The target computing node is scheduled to execute the computing task.

3. The method according to claim 2, characterized in that, The multiple computing node groups are obtained through the following method: Based on the basic computing power performance data of each of the multiple computing power nodes, a basic capability score for each computing power node is determined; Based on the aforementioned basic capability scores, the multiple computing power nodes are divided into multiple computing power node groups.

4. The method according to claim 3, characterized in that, The basic computing power performance data of each computing power node includes the index values ​​of multiple basic computing power performance indicators; The step of determining the basic capability score of each computing power node based on its respective basic computing power performance data includes: The basic computing power performance data of each of the multiple computing power nodes are analyzed based on the entropy weight method to determine the weight value of each of the basic computing power performance indicators. For each computing power node, a basic capability score is determined based on the weight value of each basic computing power performance indicator and the index value of each basic computing power performance indicator corresponding to the computing power node.

5. The method according to claim 3, characterized in that, The process of dividing the plurality of computing power nodes into plurality of computing power node groups based on the basic capability scores includes: Based on the decision tree algorithm, the basic capability score is used as the feature value to divide the multiple computing power nodes into multiple computing power node groups.

6. The method according to any one of claims 2 to 5, characterized in that, The step of determining the target computing power node matching the computing power task based on the real-time computing power performance data of each of the multiple computing power nodes within the target computing power node group and the real-time computing power performance requirements includes: Based on the real-time performance data of each computing node in the target computing power node group, and the real-time computing power requirements, multiple first computing power nodes in the target computing power group that meet the task requirements are determined. Determine the distance between the real-time performance data of each first computing node and the real-time computing requirement; The first computing power node with the smallest distance value is determined as the target computing power node.

7. A computing resource scheduling device, characterized in that, include: The first determination module is used to determine the resource requirements corresponding to the computing power task; The scheduling module is used to schedule computing nodes for the computing task based on the basic computing power performance data and real-time computing power performance data of each of the multiple computing power nodes, as well as the resource requirements.

8. An electronic device, characterized in that, It includes a processor and a memory storing a computer program, wherein the processor executes the program to implement the computing resource scheduling method according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the computing resource scheduling method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the computing resource scheduling method according to any one of claims 1 to 6.

Images