Dynamic computing power scheduling method based on tenant business priority

Abstract

The invention relates to the technical field of computing power scheduling, in particular to a tenant business priority-based computing power dynamic scheduling method, which comprises the following steps of: acquiring business resource information of a plurality of cluster tenants in a target cluster to determine the business priority of each cluster tenant; constructing a corresponding computing power account for each cluster tenant based on the service priority, and determining an account interference feature of each cluster tenant; determining a plurality of cluster tenant matching pairs based on the account state and the service priority of each cluster tenant; on the basis of the account interference features of the cluster tenants, computing power resource division is carried out on the cluster tenant matching pairs, and computing power occupation types of the cluster tenants are determined; and determining whether to adjust the account state of each cluster tenant based on the computing power occupation type of each cluster tenant matching pair so as to complete the computing power dynamic scheduling of the target cluster. According to the method, computing power distribution can be carried out according to the actual priorities of the tenant businesses, computing power collaborative scheduling among the tenants is realized, and the multi-tenant computing power scheduling efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of computing power scheduling technology, and in particular to a method for dynamic computing power scheduling based on tenant service priorities. Background Technology

[0002] In cloud computing and data center environments, efficient scheduling of computing resources is one of the key technologies to ensure the efficient operation of multi-tenant services. With the widespread adoption of cloud computing, data centers face complex scenarios involving multiple tenants sharing resources. Each tenant has different business types, importance, urgency, and resource requirements, making it difficult for traditional computing scheduling methods to meet the needs of all tenants.

[0003] On the one hand, traditional computing power scheduling methods typically employ simple round-robin or random allocation strategies, which cannot allocate resources based on the actual priority of tenant services. For example, when resources are scarce, high-priority urgent services may be delayed due to insufficient resources, while low-priority non-urgent services may consume excessive resources. This irrational allocation of resources not only affects the operational efficiency of critical services but also reduces the overall user experience. Furthermore, existing technologies lack a dynamic evaluation mechanism for tenant service priorities. Tenants' service priorities may change with business development, market changes, or adjustments in user needs, but traditional scheduling methods cannot respond to these changes in a timely manner, leading to a disconnect between resource allocation and business requirements. On the other hand, traditional scheduling strategies lack inter-tenant resource collaboration mechanisms. Each tenant's computing power account is independent, and resources cannot be flexibly transferred across accounts. When a tenant experiences a shortage of computing power resources, it cannot quickly utilize the idle computing power of other tenants; conversely, tenants with idle resources cannot temporarily share redundant resources, resulting in generally low overall cluster computing power utilization, which severely restricts the business processing efficiency and operational stability of the cluster computing power platform. To address these issues, a method is needed that can dynamically adjust the allocation of computing resources based on tenant business priorities. Summary of the Invention

[0004] To address this, the present invention provides a dynamic computing power scheduling method based on tenant service priorities, which overcomes the problems in the prior art of being unable to allocate resources according to the actual priority of tenant services and being unable to achieve resource collaborative scheduling among tenants.

[0005] To achieve the above objectives, the present invention provides a method for dynamic scheduling of computing power based on tenant service priorities, comprising: Obtain business resource information of several cluster tenants within the target cluster to determine the business priority of each cluster tenant; Based on the business priority, a corresponding computing power account is constructed for each of the cluster tenants, and the account interference characteristics of each cluster tenant are determined. In response to receiving a computing power scheduling request, the computing power request data of each cluster tenant and the computing power load data of each cluster node of the target cluster are obtained to determine the account status of each cluster tenant. Based on the account status and business priority of each cluster tenant, several cluster tenant matching pairs are determined, wherein each cluster tenant matching pair includes any lending cluster tenant and the corresponding lending cluster tenant. Based on the account interference characteristics of each cluster tenant, the computing resources of each cluster tenant pair are allocated, and the computing power occupancy type of each cluster tenant is determined. Based on the computing power occupancy type of each cluster tenant pair, determine whether to adjust the account status of each cluster tenant in order to complete the dynamic scheduling of computing power of the target cluster.

[0006] Further, determining the business priority of any of the cluster tenants includes: Feature extraction is performed based on the business resource information of the cluster tenants to determine several key feature fields; The business priority representation value is determined based on the indicator mapping representation value corresponding to each of the key feature fields. The business priority of the cluster tenant is determined based on the business priority characterization value.

[0007] Furthermore, based on the business priority, a corresponding computing power account is constructed for the cluster tenant, including: The corresponding baseline resource quota is determined based on the business priority of the cluster tenant; The computing power account corresponding to the cluster tenant is constructed based on the baseline resource quota.

[0008] Further, the account interference characteristics of the cluster tenant are determined, including: The corresponding elastic quota index is determined based on the business priority representation value of the cluster tenant. The account interference characteristics of the cluster tenant are determined based on the baseline resource quota and the elastic quota index, wherein the account interference characteristics include high interference characteristics and low interference characteristics.

[0009] Further, determining the account status of the cluster tenant includes: The computing power request characterization value is determined based on the comparison results between the computing power request data of the cluster tenant and the standard request data; The cluster load characterization value is determined based on the comparison results between the computing power load data of the cluster nodes corresponding to the cluster tenant and the standard load data. The account status of the cluster tenant is determined based on the computing power request characterization value and the cluster load characterization value, wherein the account status includes a lendable status and a non-lendable status.

[0010] Further, determining several cluster tenant matching pairs, including: Based on the account status of each cluster tenant, the cluster tenants are classified into lendable tenants and non-lendable tenants. Several cluster tenant matching pairs are randomly generated based on the business priority of each cluster tenant. The lending cluster tenant is a non-lending tenant, and the lending cluster tenant is a lending tenant. The business priority of the lending cluster tenant is higher than that of the lending cluster tenant.

[0011] Furthermore, the cluster tenant matching pairs are allocated computing resources, including: The comprehensive interference characterization value is determined based on the account interference characteristics of the cluster tenant matching pairs; Based on the comprehensive interference characterization value, the computing resource allocation method corresponding to the cluster tenant matching pair is determined, including coarse-grained allocation, fine-grained allocation, and mixed-grained allocation.

[0012] Further, determining the computing power occupancy type of any of the aforementioned cluster tenants includes: The computing power occupancy type of the cluster tenant is determined based on the computing power request characterization value and the comprehensive interference characterization value, including preemptive occupancy and non-preemptive occupancy.

[0013] Further, determining whether to adjust the account status of each of the aforementioned cluster tenants includes: If the computing power occupancy type of each lending cluster tenant is preemptive occupancy, and the computing power occupancy type of each lending cluster tenant is non-preemptive occupancy, then the account status of each cluster tenant will not be adjusted.

[0014] Furthermore, if any of the lending cluster tenants has a non-preemptive computing power occupancy type, and the lending cluster tenant corresponding to the lending cluster tenant has a preemptive computing power occupancy type, then the account status of the lending cluster tenant corresponding to the lending cluster tenant will be adjusted.

[0015] Compared with existing technologies, the advantages of this invention are as follows: By acquiring business resource information of several cluster tenants within the target cluster, this invention can accurately identify business needs and rationally allocate business priorities. Determining the business priorities of each cluster tenant provides a clear basis for subsequent computing power scheduling, enabling collaborative computing power scheduling among tenants. By constructing corresponding computing power accounts for each cluster tenant based on business priorities, refined management of resource usage for each cluster tenant is achieved. By determining the account interference characteristics of each cluster tenant, the resource interaction boundaries between different tenant accounts can be clarified, laying the foundation for subsequent collaborative computing power scheduling. By acquiring computing power request data from each cluster tenant and computing power load data from each cluster node of the target cluster, changes in resource supply and demand can be perceived in real time, and the account status of each cluster tenant can be dynamically evaluated, improving the accuracy and efficiency of computing power scheduling. By determining several cluster tenant matching pairs based on the account status and business priorities of each cluster tenant, inter-tenant computing power scheduling can be achieved. By optimizing resource utilization efficiency and ensuring the priority of high-priority services through unified scheduling, and by allocating computing resources to matching tenants based on their account interference characteristics, the accuracy of computing resource allocation can be improved, ensuring compliance and fairness in resource allocation. Determining the computing power usage type of each tenant helps avoid resource conflicts and improves the efficiency and security of computing power utilization. By determining whether to adjust the account status of each tenant based on their computing power usage type, it is possible to ensure that computing power allocation is always in an optimal state, improving the flexibility and adaptability of computing power allocation and enhancing the accuracy of multi-tenant computing power scheduling.

[0016] Furthermore, this invention extracts features based on the business resource information of the cluster tenants to determine several key feature fields. This transforms abstract business attributes into quantifiable and comparable specific features, avoiding the subjectivity and randomness of manual priority determination and improving the accuracy of priority assessment. By determining business priority representation values ​​based on the indicator mapping representation values ​​corresponding to each key feature field, multiple core attributes of the business can be comprehensively weighed, avoiding the one-sidedness of judgment based on a single feature dimension and enhancing the flexibility and adaptability of priority ranking. By determining the business priority of cluster tenants based on business priority representation values, the automatic connection between priority determination and resource scheduling is achieved, improving the overall efficiency of cluster computing power scheduling and providing a clear decision-making basis for the subsequent differentiated allocation of computing power resources.

[0017] Furthermore, this invention determines corresponding baseline resource quotas based on the business priorities of cluster tenants, achieving differentiated and precise allocation of resource quotas. This provides a quantitative benchmark for computing resource management. Setting baseline resource quotas based on priority hierarchies creates a clear resource allocation hierarchy, reducing the complexity of cluster resource scheduling and improving computing resource scheduling efficiency. The computing account uses the baseline resource quota as its core parameter to achieve refined measurement and management of tenant resource usage, providing a platform for computing power collaboration among tenants and enhancing the flexibility and adjustability of computing resource scheduling.

[0018] Furthermore, this invention determines the corresponding elastic quota index based on the business priority characterization value to achieve dynamic elastic adaptation of quota allocation. It also determines the account interference characteristics of cluster tenants based on the baseline resource quota and the elastic quota index, ensuring the stability of each tenant's business and providing a standardized execution basis for dynamic scheduling.

[0019] Furthermore, by comparing actual computing power request data with standard request data, this invention can accurately reflect the difference between a tenant's current resource needs and normal needs, quantify the fluctuation state of a tenant's resource needs, accurately identify the supply and demand gap of tenant resources, improve the efficiency and accuracy of subsequent tenant matching, and further improve the accuracy of computing power scheduling among tenants. By comparing real-time computing power load data of nodes with standard load data, the current load pressure of nodes can be quantified. The comprehensive determination of account status by combining computing power request characterization values ​​and cluster load characterization values ​​avoids the limitations of single-dimensional judgment, improves the accuracy of status determination, and provides a precise basis for subsequent tenant matching pair construction.

[0020] Furthermore, this invention categorizes tenants into lendable and non-lendable classes based on their account status, clearly defining the permission boundaries for different tenants to participate in the sharing of computing resources, improving the efficiency and accuracy of subsequent tenant matching, and randomly generating several cluster tenant matching pairs based on the business priorities of each cluster tenant, reducing the complexity of matching rules, improving scheduling flexibility, and ensuring the efficiency of computing resource scheduling among tenants.

[0021] Furthermore, by determining the comprehensive interference characterization value, this invention achieves the quantitative integration of interference rules between matching pairs, strengthens the correlation between interference features and the fit of matching pairs, and determines the computing power resource allocation method corresponding to the matching pairs of cluster tenants based on the comprehensive interference characterization value. This enables differentiated and precise adaptation of resource allocation, balances resource flow efficiency and tenant rights protection, and improves the efficiency of computing power resource scheduling among tenants. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the dynamic computing power scheduling method based on tenant service priority according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the process of determining the service priority of any cluster tenant in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the process of determining the account status of a cluster tenant according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the process of determining several cluster tenant matching pairs in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0024] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0025] Please see Figure 1 The diagram shown is a flowchart illustrating a dynamic computing power scheduling method based on tenant service priority according to an embodiment of the present invention. The present invention provides a dynamic computing power scheduling method based on tenant service priority, comprising: Step S1: Obtain the business resource information of several cluster tenants within the target cluster to determine the business priority of each cluster tenant; Please see Figure 2 The diagram illustrates the process of determining the service priority of any cluster tenant according to an embodiment of the present invention. Specifically, in step S1, determining the service priority of any cluster tenant includes: Step S11: Based on the business resource information of the cluster tenant, feature extraction is performed to determine several key feature fields; Step S12: Determine the business priority representation value based on the indicator mapping representation value corresponding to each of the key feature fields; Step S13: Determine the service priority of the cluster tenant based on the service priority characterization value.

[0026] In implementation, business resource information includes, but is not limited to, CPU utilization, memory usage, network bandwidth, service runtime, network I / O rate, and disk I / O throughput. This information can be collected in real-time using monitoring tools (such as Prometheus, Ganglia, and Kubernetes). Since the business types of each cluster tenant may differ, the business resource information for each cluster tenant may also differ. Therefore, there are no restrictions on the feature extraction method. For example, for real-time services (such as online transaction systems), CPU and network bandwidth are extracted as key feature fields; for batch processing services (such as data analysis), memory and storage requirements are extracted as key feature fields; for services with specific deadlines, the deadline and current progress are extracted as key feature fields; and for services contributing to the tenant's core business... For major business operations, extract business value assessment indicators (such as business revenue share) as key characteristic fields. Implementers can set up a key characteristic field lookup table based on the actual situation. Each key characteristic field has a corresponding indicator mapping value. These values ​​can be comprehensively set based on the degree of business relevance, impact, and urgency. The weighted sum of the indicator mapping values ​​corresponding to each key characteristic field is used to determine the business priority value. Implementers can set up a priority value lookup table based on the actual situation. For example, for a business priority level one, the corresponding business priority value ranges from 0 to 0.1; for a business priority level two, the corresponding business priority value ranges from 0.1 to 1.0; for a business priority level three, the corresponding business priority value ranges from 1.0 to 2.0, and so on.

[0027] Specifically, this invention extracts features based on the business resource information of the cluster tenants to determine several key feature fields. This transforms abstract business attributes into quantifiable and comparable specific features, avoiding the subjectivity and randomness of manual priority determination and improving the accuracy of priority assessment. By determining business priority representation values ​​based on the index mapping representation values ​​corresponding to each key feature field, multiple core attributes of the business can be comprehensively weighed, avoiding the one-sidedness of judgment based on a single feature dimension and enhancing the flexibility and adaptability of priority ranking. By determining the business priority of cluster tenants based on business priority representation values, the automatic connection between priority determination and resource scheduling is achieved, improving the overall efficiency of cluster computing power scheduling and providing a clear decision-making basis for the subsequent differentiated allocation of computing power resources.

[0028] Step S2: Based on the business priority, construct corresponding computing power accounts for each cluster tenant and determine the account interference characteristics of each cluster tenant; Specifically, in step S2, constructing a corresponding computing power account for the cluster tenant based on the business priority includes: Step S21: Determine the corresponding baseline resource quota based on the business priority of the cluster tenant; Step S22: Construct the computing power account corresponding to the cluster tenant based on the baseline resource quota.

[0029] In implementation, each service priority has a corresponding baseline coefficient, which is positively correlated with the service priority. For example, a first-level service priority has a baseline coefficient ranging from 0.1 to 0.5, a second-level priority from 0.5 to 1.0, a third-level priority from 1.0 to 2.0, and so on. For any given service priority, the baseline resource quota is determined by multiplying the baseline coefficient by the standard resource quota. This resource quota represents overall computing power, including but not limited to: CPU resources (total number of cores, single-core computing power); memory resources (total memory capacity (GB), memory read / write speed); GPU resources (number of GPU cards, single-card computing power, video memory capacity); and storage resources (available storage space, IOPS). Implementers can set the standard resource quota based on actual conditions or the average resource quota of cluster tenants that have passed qualification checks in historical data.

[0030] Understandably, a computing power account is built for each cluster tenant based on the baseline resource quota of each cluster tenant, so as to supply computing power resources to each cluster tenant. The computing power resources of each cluster tenant can be scheduled through the computing power account.

[0031] Specifically, this invention determines corresponding baseline resource quotas based on the business priorities of cluster tenants, achieving differentiated and precise allocation of resource quotas. This provides a quantitative benchmark for computing resource management. By setting baseline resource quotas based on priority levels, a clear hierarchical system for resource allocation can be formed, reducing the complexity of cluster resource scheduling and improving the efficiency of computing resource scheduling. The computing account uses the baseline resource quota as the core parameter to achieve refined measurement and management of tenant resource usage, providing a platform for computing power collaboration among tenants and enhancing the flexibility and adjustability of computing resource scheduling.

[0032] Specifically, in step S2, determining the account interference characteristics of the cluster tenant includes: Step S23: Determine the corresponding elastic quota index based on the business priority characterization value of the cluster tenant; Step S24: Determine the account interference characteristics of the cluster tenant based on the baseline resource quota and the elastic quota index, wherein the account interference characteristics include high interference characteristics and low interference characteristics.

[0033] In implementation, for any cluster tenant, the product of the business priority representation value corresponding to the cluster tenant and the standard resource quota is determined as the elastic quota amount corresponding to the cluster tenant. The lower limit of the corresponding elastic quota indicator is the difference between the base resource quota and the elastic quota amount, and the upper limit of the elastic quota indicator is the sum of the base resource quota and the elastic quota amount.

[0034] It is understandable that the account interference representation value is determined based on the baseline resource quota and elastic quota indicators corresponding to the cluster tenant, where the baseline resource quota ZY = (ZY1, ZY2, ..., ZY...). j , ..., ZY m ), Flexible quota index ZE = (ZE1, ZE2, ..., ZE) j , ..., ZE m ), ZE j =(ZE j,min ,ZE j,max If the account interference characteristic value ZP = (∑ m j=1 (((ZY j / ZE j,max )+(ZY j / ZE j,min )) / 2)) / m, where j=1, 2, ..., m; m is the resource quota dimension, ZE j,min ZE represents the lower limit of the elastic quota indicator corresponding to the j-th dimension of the baseline quota. j,max Let be the upper limit of the elastic quota indicator corresponding to the j-th dimension benchmark resource quota.

[0035] Understandably, the account interference characteristics of a cluster tenant are determined based on the comparison between the account interference characterization value and the preset interference characterization value. If the account interference characterization value is greater than the preset interference characterization value, the account interference characteristics of the cluster tenant are determined to be high interference characteristics. If the account interference characterization value is less than or equal to the preset interference characterization value, the account interference characteristics of the cluster tenant are determined to be low interference characteristics. In practice, the implementers can set the preset interference characterization value based on the actual situation or the average value of the account interference characterization values ​​of cluster tenants that have passed the qualification test in historical data.

[0036] Specifically, this invention determines the corresponding elastic quota index based on the business priority characterization value to achieve dynamic elastic adaptation of quota allocation. It also determines the account interference characteristics of cluster tenants based on the baseline resource quota and the elastic quota index to ensure the stability of each tenant's business and provides a standardized execution basis for dynamic scheduling.

[0037] Step S3: In response to receiving a computing power scheduling request, obtain the computing power request data of each cluster tenant and the computing power load data of each cluster node of the target cluster to determine the account status of each cluster tenant. Please see Figure 3 The diagram illustrates the process of determining the account status of a cluster tenant according to an embodiment of the present invention. Specifically, in step S3, determining the account status of the cluster tenant includes: Step S31: Determine the computing power request characterization value based on the comparison result between the computing power request data of the cluster tenant and the standard request data; Step S32: Determine the cluster load characterization value based on the comparison result between the computing power load data of the cluster node corresponding to the cluster tenant and the standard load data. Step S33: Determine the account status of the cluster tenant based on the computing power request characterization value and the cluster load characterization value, wherein the account status includes a lendable status and a non-lendable status.

[0038] In implementation, computing power request data includes, but is not limited to, the number of CPU cores requested, the memory capacity requested, the computing power utilization rate, and the number of GPUs requested. The computing power request data is standardized to eliminate the differences in the units of measurement for different resource indicators. For any cluster tenant, the computing power request data is SY1, SY2, ..., SY... i , ..., SY n With standard request data SE1, SE2, ..., SE i SE n Then the computing power request representation value SP = (∑ n i=1 SY i ×SE i ) / (sqrt(∑ n i=1 (SY i ) 2 )×sqrt(∑ n i=1 (SE i ) 2 )), where i=1,2,…,n; n is the dimension of computing power request data, and sqrt() is the preset square root determination function; the actual implementers can set the standard request data based on the actual situation or the maximum value of the cluster tenant computing power request data that has passed the qualification test in historical data.

[0039] It is understood that each cluster tenant has a corresponding cluster node, and the computing load data includes, but is not limited to, CPU utilization, memory utilization, disk I / O utilization, network bandwidth utilization, and task queuing length. For any cluster tenant, the computing load data is FY1, FY2, ..., FY g , ..., FY h With standard load data FE1, FE2, ..., FE g , ..., FE h Then the cluster load characterization value FP = (∑h g=1 FY g ×FE g ) / (sqrt(∑ h g=1 (FY g ) 2 )×sqrt(∑ h g=1 (FE g ) 2 )), where g=1,2,…,h; h is the computing load data dimension. The actual implementers can set the standard load data based on the actual situation or the average computing load data of cluster tenants that have passed the qualification test in historical data.

[0040] It is understood that for any cluster tenant, if the computing power request characteristic value is less than the preset request characteristic value and the cluster load characteristic value is greater than the preset load characteristic value, then the cluster tenant's account status is determined to be available for lending; if the computing power request characteristic value is greater than or equal to the preset request characteristic value, or the cluster load characteristic value is less than or equal to the preset load characteristic value, then the cluster tenant's account status is determined to be unavailable for lending. Preferably, the preset request characteristic value is set to a range of 0.5 to 0.6, and the preset load characteristic value is set to a range of 0.8 to 0.9.

[0041] Specifically, this invention compares actual computing power request data with standard request data to accurately reflect the difference between a tenant's current resource needs and normal needs, quantifies the fluctuation of tenant resource needs, accurately identifies the supply and demand gap of tenant resources, improves the efficiency and accuracy of subsequent tenant matching, and further enhances the accuracy of computing power scheduling among tenants. By comparing real-time computing power load data of nodes with standard load data, the current load pressure of nodes can be quantified. The comprehensive determination of account status using both computing power request characterization values ​​and cluster load characterization values ​​avoids the limitations of single-dimensional judgment, improves the accuracy of status determination, and provides a precise basis for subsequent tenant matching pair construction.

[0042] Step S4: Determine several cluster tenant matching pairs based on the account status and business priority of each cluster tenant, wherein each cluster tenant matching pair includes any lending cluster tenant and the corresponding lending cluster tenant. Please see Figure 4 As shown, this is a schematic diagram of the process for determining several cluster tenant matching pairs according to an embodiment of the present invention; specifically, in step S4, determining several cluster tenant matching pairs includes: Step S41: Classify each cluster tenant based on the account status of each cluster tenant to obtain tenants that can be lent and tenants that cannot be lent. Step S42: Randomly generate several cluster tenant matching pairs based on the business priority of each cluster tenant, wherein the lending cluster tenant is a non-lending tenant, the lending cluster tenant is a lending tenant, and the business priority of the lending cluster tenant is higher than that of the lending cluster tenant.

[0043] In implementation, for any cluster tenant, if the account status is available for lending, the cluster tenant is classified as a lendable tenant; if the account status is not available for lending, the cluster tenant is classified as a non-lendable tenant. High-priority cluster tenants are prioritized for matching, and low-priority cluster tenants are prioritized for matching high-priority cluster tenants with low-priority cluster tenants.

[0044] Specifically, this invention categorizes tenants into lendable and non-lendable classes based on their account status, clearly defining the permission boundaries for different tenants to participate in the sharing of computing resources, improving the efficiency and accuracy of subsequent tenant matching, and randomly generating several cluster tenant matching pairs based on the business priorities of each cluster tenant, reducing the complexity of matching rules, improving scheduling flexibility, and ensuring the efficiency of computing resource scheduling among tenants.

[0045] Step S5: Based on the account interference characteristics of each cluster tenant, allocate computing resources for each cluster tenant matching pair and determine the computing power occupancy type of each cluster tenant. Specifically, in step S5, the computing resources are allocated to the cluster tenant matching pairs, including: Step S51: Determine the comprehensive interference characterization value based on the account interference characteristics of the cluster tenant matching pair; Step S52: Based on the comprehensive interference characterization value, determine the computing resource allocation method corresponding to the cluster tenant matching pair, including coarse-grained allocation, fine-grained allocation, and mixed-grained allocation.

[0046] In implementation, each account's interference characteristic has a corresponding interference representation value. For example, the interference representation value corresponding to a high interference characteristic is set to 1.0, and the interference representation value corresponding to a low interference characteristic is set to 0.2. The comprehensive interference representation value is determined based on the average interference representation values ​​of the lending and borrowing cluster tenants and the lending cluster tenants in the cluster tenant matching pair. For example, if the lending cluster tenant in the cluster tenant matching pair has a high interference characteristic, then the corresponding interference representation value is set to 1.0; if the lending cluster tenant has a high interference characteristic, then the corresponding interference representation value is set to 1.0. Therefore, the comprehensive interference representation value is (1.0). +1.0) / 2=1.0; If the lending cluster tenant in the matching pair has low interference characteristics, the corresponding interference characterization value is set to 0.2, and the lending cluster tenant has high interference characteristics, the corresponding interference characterization value is set to 1.0, then the comprehensive interference characterization value is (0.2+1.0) / 2=0.6; If the lending cluster tenant in the matching pair has low interference characteristics, the corresponding interference characterization value is set to 0.2, and the lending cluster tenant has low interference characteristics, the corresponding interference characterization value is set to 0.2, then the comprehensive interference characterization value is (0.2+0.2) / 2=0.2.

[0047] Understandably, coarse-grained partitioning corresponds to the smallest comprehensive interference characteristic value, while fine-grained partitioning corresponds to the largest. All other partitioning is of mixed granularity. Fine-grained partitioning is based on the smallest computing power unit; coarse-grained partitioning is based on 10% to 20% of the baseline resource quota; in mixed granularity partitioning, if the tenant lending the cluster has high interference characteristics, it is partitioned based on the smallest computing power unit; if the tenant lending the cluster has low interference characteristics, it is partitioned based on 10% to 20% of the baseline resource quota.

[0048] Specifically, this invention achieves quantitative integration of interference rules between matching pairs by determining a comprehensive interference characterization value, strengthens the correlation between interference features and matching pair fit, and determines the computing power resource allocation method corresponding to the matching pairs of cluster tenants based on the comprehensive interference characterization value. This enables differentiated and precise adaptation of resource allocation, balances resource flow efficiency and tenant rights protection, and improves the efficiency of computing power resource scheduling among tenants.

[0049] Specifically, in step S5, determining the computing power occupancy type of any of the cluster tenants includes: The computing power occupancy type of the cluster tenant is determined based on the computing power request characterization value and the comprehensive interference characterization value, including preemptive occupancy and non-preemptive occupancy.

[0050] In implementation, for any cluster tenant, if both the computing power request characteristic value and the comprehensive interference characteristic value are greater than the preset characteristic value, the computing power occupancy type for that cluster tenant is determined to be preemptive occupancy. If the computing power request characteristic value is less than or equal to the preset request characteristic value, or the comprehensive interference characteristic value is less than or equal to the preset characteristic value, the computing power occupancy type for that cluster tenant is determined to be non-preemptive occupancy. Implementers can set the preset characteristic value based on the actual situation or the minimum comprehensive interference characteristic value that passed the qualification test in historical data.

[0051] Step S6: Determine whether to adjust the account status of each cluster tenant based on the computing power occupancy type of each cluster tenant matching pair, so as to complete the dynamic scheduling of computing power of the target cluster.

[0052] Specifically, in step S6, determining whether to adjust the account status of each cluster tenant includes: If the computing power occupancy type of each lending cluster tenant is preemptive occupancy, and the computing power occupancy type of each lending cluster tenant is non-preemptive occupancy, then the account status of each cluster tenant will not be adjusted.

[0053] If any of the lending cluster tenants has a non-preemptive computing power occupancy type, and the corresponding lending cluster tenant has a preemptive computing power occupancy type, then the account status of the corresponding lending cluster tenant will be adjusted.

[0054] It is understandable that the computing resources of the lending cluster tenant's computing power account are temporarily allocated to the lending cluster tenant to ensure the business stability of the lending cluster tenant.

[0055] This invention, by acquiring business resource information from several cluster tenants within a target cluster, can accurately identify business needs and rationally prioritize business operations. Determining the business priorities of each cluster tenant provides a clear basis for subsequent computing power scheduling, enabling collaborative computing power scheduling among tenants. By constructing corresponding computing power accounts for each cluster tenant based on business priorities, refined management of resource usage for each cluster tenant is achieved. Determining the account interference characteristics of each cluster tenant clarifies the resource interaction boundaries between different tenant accounts, laying the foundation for subsequent collaborative computing power scheduling. By acquiring computing power request data from each cluster tenant and computing power load data from each cluster node of the target cluster, changes in resource supply and demand can be perceived in real time, and the account status of each cluster tenant can be dynamically evaluated, improving the accuracy and efficiency of computing power scheduling. By determining several cluster tenant matching pairs based on the account status and business priorities of each cluster tenant, inter-tenant computing power scheduling can be achieved. By optimizing resource utilization efficiency and ensuring the priority of high-priority services through unified scheduling, and by allocating computing resources to matching tenants based on their account interference characteristics, the accuracy of computing resource allocation can be improved, ensuring compliance and fairness in resource allocation. Determining the computing power usage type of each tenant helps avoid resource conflicts and improves the efficiency and security of computing power utilization. By determining whether to adjust the account status of each tenant based on their computing power usage type, it is possible to ensure that computing power allocation is always in an optimal state, improving the flexibility and adaptability of computing power allocation and enhancing the accuracy of multi-tenant computing power scheduling.

[0056] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for dynamic scheduling of computing power based on tenant service priorities, characterized in that, include: Obtain business resource information of several cluster tenants within the target cluster to determine the business priority of each cluster tenant; Based on the business priority, a corresponding computing power account is constructed for each of the cluster tenants, and the account interference characteristics of each cluster tenant are determined. In response to receiving a computing power scheduling request, the computing power request data of each cluster tenant and the computing power load data of each cluster node of the target cluster are obtained to determine the account status of each cluster tenant. Based on the account status and business priority of each cluster tenant, several cluster tenant matching pairs are determined, wherein each cluster tenant matching pair includes any lending cluster tenant and the corresponding lending cluster tenant. Based on the account interference characteristics of each cluster tenant, the computing resources of each cluster tenant pair are allocated, and the computing power occupancy type of each cluster tenant is determined. Based on the computing power occupancy type of each cluster tenant pair, determine whether to adjust the account status of each cluster tenant in order to complete the dynamic scheduling of computing power of the target cluster.

2. The dynamic computing power scheduling method based on tenant service priority according to claim 1, characterized in that, Determining the business priority of any of the aforementioned cluster tenants includes: Feature extraction is performed based on the business resource information of the cluster tenants to determine several key feature fields; The business priority representation value is determined based on the indicator mapping representation value corresponding to each of the key feature fields. The business priority of the cluster tenant is determined based on the business priority characterization value.

3. The dynamic computing power scheduling method based on tenant service priority according to claim 2, characterized in that, Based on the business priority, corresponding computing power accounts are constructed for the cluster tenants, including: The corresponding baseline resource quota is determined based on the business priority of the cluster tenant; The computing power account corresponding to the cluster tenant is constructed based on the baseline resource quota.

4. The dynamic computing power scheduling method based on tenant service priority according to claim 3, characterized in that, Determining the account interference characteristics of the cluster tenant includes: The corresponding elastic quota index is determined based on the business priority representation value of the cluster tenant. The account interference characteristics of the cluster tenant are determined based on the baseline resource quota and the elastic quota index, wherein the account interference characteristics include high interference characteristics and low interference characteristics.

5. The dynamic computing power scheduling method based on tenant service priority according to claim 4, characterized in that, Determining the account status of the cluster tenant includes: The computing power request characterization value is determined based on the comparison results between the computing power request data of the cluster tenant and the standard request data; The cluster load characterization value is determined based on the comparison results between the computing power load data of the cluster nodes corresponding to the cluster tenant and the standard load data. The account status of the cluster tenant is determined based on the computing power request characterization value and the cluster load characterization value, wherein the account status includes a lendable status and a non-lendable status.

6. The dynamic computing power scheduling method based on tenant service priority according to claim 5, characterized in that, Determine several cluster tenant matching pairs, including: Based on the account status of each cluster tenant, the cluster tenants are classified into lendable tenants and non-lendable tenants. Several cluster tenant matching pairs are randomly generated based on the business priority of each cluster tenant. The lending cluster tenant is a non-lending tenant, and the lending cluster tenant is a lending tenant. The business priority of the lending cluster tenant is higher than that of the lending cluster tenant.

7. The dynamic computing power scheduling method based on tenant service priority according to claim 6, characterized in that, The cluster tenant matching pairs are allocated computing resources, including: The comprehensive interference characterization value is determined based on the account interference characteristics of the cluster tenant matching pairs; Based on the comprehensive interference characterization value, the computing resource allocation method corresponding to the cluster tenant matching pair is determined, including coarse-grained allocation, fine-grained allocation, and mixed-grained allocation.

8. The dynamic computing power scheduling method based on tenant service priority according to claim 7, characterized in that, Determining the computing power usage type of any of the aforementioned cluster tenants includes: The computing power occupancy type of the cluster tenant is determined based on the computing power request characterization value and the comprehensive interference characterization value, including preemptive occupancy and non-preemptive occupancy.

9. The method for dynamic scheduling of computing power based on tenant service priority according to claim 8, characterized in that, Determine whether to adjust the account status of each of the aforementioned cluster tenants, including: If the computing power occupancy type of each lending cluster tenant is preemptive occupancy, and the computing power occupancy type of each lending cluster tenant is non-preemptive occupancy, then the account status of each cluster tenant will not be adjusted.

10. The dynamic computing power scheduling method based on tenant service priority according to claim 9, characterized in that, If any of the lending cluster tenants has a non-preemptive computing power occupancy type, and the corresponding lending cluster tenant has a preemptive computing power occupancy type, then the account status of the corresponding lending cluster tenant will be adjusted.

Images