Resource scheduling method and device, computer equipment and storage medium

Abstract

The invention relates to a resource scheduling method and device, computer equipment and a storage medium. The method belongs to the technical field of resource management. The method comprises the following steps: determining tenant feature representation of a target tenant and database feature representation of a cloud database; and performing fusion processing on the time sequence demand feature representation, the tenant portrait feature representation and the task association feature representation to obtain tenant comprehensive feature representation of the target tenant. And determining a resource demand prediction result of the target tenant according to the tenant comprehensive feature representation and the database feature representation. And performing resource scheduling for the target tenant according to the resource demand prediction result of the target tenant. According to the method, the tenant portrait information and the task association information of the target tenant and the resource feature information and the task feature information of the cloud database are considered, so that the resource demand of the target tenant can be predicted more accurately, and the accuracy and rationality of resource scheduling are improved; the database resource utilization rate is improved; and the SLA guarantee of the target tenant is realized.

Description

Technical Field

[0001] This application relates to the field of resource management technology, and in particular to a resource scheduling method, apparatus, computer equipment, and storage medium. Background Technology

[0002] The cross-tenant resource scheduling of the metering automation big data platform faces the dual challenges of dynamic supply and demand imbalance and multi-objective optimization. Tenants' resource demands are affected by the cycle of cross-database synchronization tasks (such as real-time metering data cross-database transmission, weekly cross-database data aggregation, and monthly cross-database archiving and backup), exhibiting quasi-periodic fluctuations. On the other hand, they are affected by sudden cross-database tasks (such as temporary cross-database data scheduling, cross-database data recovery and synchronization after failure, and emergency metering data cross-database reporting) and a surge in the amount of synchronized data (such as the doubling of cross-database synchronized data volume due to high-frequency reporting by metering equipment under extreme weather conditions), resulting in irregular and sudden fluctuations.

[0003] Currently, resource scheduling for tenants still relies on traditional fixed configuration allocation (such as pre-allocating CPU cores, memory capacity, cross-database synchronization bandwidth, storage quotas, and IO resources based on the tenant's database size) or simple queue rules (such as FIFO and priority queues). This cannot accurately predict the tenant's resource needs, resulting in low utilization of allocated resources (e.g., idle computing resources) or SLA (Service Level Agreement) breaches (e.g., cross-database synchronization delays and query response timeouts during peak periods). Summary of the Invention

[0004] Therefore, it is necessary to provide a resource scheduling method, apparatus, computer equipment, and storage medium that can effectively improve the accuracy of tenant resource demand prediction, thereby improving resource utilization and SLA guarantee, in order to address the above-mentioned technical problems.

[0005] Firstly, this application provides a resource scheduling method. The method includes:

[0006] Determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation and task association feature representation; the database feature representation includes resource feature representation and task feature representation;

[0007] The time-series demand feature representation, tenant profile feature representation, and task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant;

[0008] Based on the tenant's comprehensive characteristic representation and the database characteristic representation, the resource demand prediction results for the target tenant are determined;

[0009] Based on the resource demand forecast results of the target tenant, resource scheduling is performed for the target tenant.

[0010] In one embodiment, determining the tenant characteristic representation of the target tenant includes:

[0011] Based on the historical resource demand information of the target tenant, the temporal demand characteristics of the target tenant are determined; wherein, the historical resource demand information includes computing resource information, storage resource information, transmission resource information and IO resource information;

[0012] Based on the tenant database type, SLA level, and historical credit rating of the target tenant, determine the tenant profile features of the target tenant;

[0013] Based on the current pending cross-database task type, task priority, and cross-database difficulty value of the target tenant, determine the task association feature representation of the target tenant.

[0014] In one embodiment, the temporal demand characteristics of the target tenant are determined based on the target tenant's historical resource demand information, including:

[0015] Historical resource demand information is decomposed to obtain resource demand decomposition results;

[0016] Feature extraction is performed on the resource demand decomposition results to obtain resource demand extraction results; the resource demand extraction results include low-frequency trend components, periodic components and high-frequency burst components;

[0017] Based on the resource demand extraction results, the temporal demand characteristics of the target tenant are determined.

[0018] In one embodiment, the tenant profile feature representation of the target tenant is determined based on the target tenant's tenant database type, SLA level, and historical credit score, including:

[0019] The target tenant's tenant database type and SLA level are encoded to obtain an encoding vector;

[0020] The encoded vector and historical confidence scores are preprocessed to obtain the tenant profile feature representation of the target tenant; the preprocessing methods include encoding and normalization.

[0021] In one embodiment, determining the database feature representation of the cloud database includes:

[0022] Feature extraction is performed on the database resource information of the cloud database to obtain database feature information; among which, the database feature information includes real-time resource load information, resource capacity limit and historical load fluctuation statistics;

[0023] Feature extraction is performed on the database feature information to obtain the resource feature representation of cloud data;

[0024] Feature extraction is performed on cross-database task types in cloud databases to obtain database task information; the database task information includes standard information on task resource consumption, task execution cycle, and burst probability.

[0025] The database task information is encoded to obtain the task feature representation of the cloud database.

[0026] In one embodiment, the resource demand prediction result for the target tenant is determined based on the tenant comprehensive feature representation and the database feature representation, including:

[0027] The tenant comprehensive feature representation and the task feature representation are fused to obtain the first fused feature representation;

[0028] The resource feature representation and the resource competition feature representation are fused to obtain a second fused feature representation;

[0029] Based on SLA constraint feature representation, demand dependency feature representation, first fusion feature representation and second fusion feature representation, a multimodal fusion feature representation is obtained;

[0030] Based on the multimodal fusion feature representation, the resource demand prediction results for the target tenant are determined.

[0031] In one embodiment, the method further includes:

[0032] Based on the historical demand characteristics of the target tenant and the historical demand characteristics of each first tenant, the time lag correlation coefficient between the target tenant and each first tenant is determined; where the first tenant is the tenant other than the target tenant among all tenants of the cloud database.

[0033] The time-delay correlation coefficients of each first tenant are sorted in descending order to obtain the sorting results;

[0034] Based on the sorting results, determine the mean value of the coefficients of the first preset number of time-delay correlation coefficients;

[0035] By establishing a correlation between the mean coefficient and the historical demand characteristics of the target tenant, a demand dependency feature representation is obtained.

[0036] In one embodiment, the method further includes:

[0037] Based on the historical demand characteristics of the target tenant and the historical load information of each resource node in the cloud database, determine the matching degree between the target tenant and each resource node;

[0038] Based on the maximum matching degree and the average matching degree, the resource competition feature representation is determined.

[0039] In one embodiment, the method further includes:

[0040] Based on the mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database resource nodes, the SLA constraint characteristics are determined.

[0041] Secondly, this application also provides a resource scheduling apparatus. The apparatus includes:

[0042] The first determining module is used to determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation and task association feature representation; the database feature representation includes resource feature representation and task feature representation;

[0043] The second determining module is used to fuse the temporal demand feature representation, the tenant profile feature representation, and the task association feature representation to obtain the comprehensive tenant feature representation of the target tenant.

[0044] The third determination module is used to determine the resource demand prediction results of the target tenant based on the tenant's comprehensive feature representation and the database feature representation;

[0045] The scheduling module is used to schedule resources for the target tenant based on the predicted resource demand results.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps.

[0047] Determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation and task association feature representation; the database feature representation includes resource feature representation and task feature representation;

[0048] The time-series demand feature representation, tenant profile feature representation, and task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant;

[0049] Based on the tenant's comprehensive characteristic representation and the database characteristic representation, the resource demand prediction results for the target tenant are determined;

[0050] Based on the resource demand forecast results of the target tenant, resource scheduling is performed for the target tenant.

[0051] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0052] Determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation and task association feature representation; the database feature representation includes resource feature representation and task feature representation;

[0053] The time-series demand feature representation, tenant profile feature representation, and task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant;

[0054] Based on the tenant's comprehensive characteristic representation and the database characteristic representation, the resource demand prediction results for the target tenant are determined;

[0055] Based on the resource demand forecast results of the target tenant, resource scheduling is performed for the target tenant.

[0056] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0057] Determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation and task association feature representation; the database feature representation includes resource feature representation and task feature representation;

[0058] The time-series demand feature representation, tenant profile feature representation, and task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant;

[0059] Based on the tenant's comprehensive characteristic representation and the database characteristic representation, the resource demand prediction results for the target tenant are determined;

[0060] Based on the resource demand forecast results of the target tenant, resource scheduling is performed for the target tenant.

[0061] The aforementioned resource scheduling method, apparatus, computer equipment, and storage medium determine the tenant characteristic representation of the target tenant and the database characteristic representation of the cloud database. The tenant characteristic representation includes a time-series demand characteristic representation, a tenant profile characteristic representation, and a task-related characteristic representation; the database characteristic representation includes a resource characteristic representation and a task characteristic representation. The time-series demand characteristic representation, the tenant profile characteristic representation, and the task-related characteristic representation are fused to obtain a comprehensive tenant characteristic representation of the target tenant. Based on the comprehensive tenant characteristic representation and the database characteristic representation, the resource demand prediction result of the target tenant is determined. Based on the resource demand prediction result of the target tenant, resource scheduling is performed for the target tenant. This application considers not only the time-series demand information of the target tenant, but also the tenant profile information, task-related information, and the resource characteristic information and task characteristic information of the cloud database, enabling more accurate prediction of the target tenant's resource demand, improving the accuracy and rationality of resource scheduling, thereby improving database resource utilization and achieving SLA assurance for the target tenant. Attached Figure Description

[0062] Figure 1 This is a diagram illustrating the application environment of the resource scheduling method provided in this embodiment.

[0063] Figure 2 This is a flowchart illustrating the first resource scheduling method provided in this embodiment;

[0064] Figure 3 This is a flowchart illustrating the process of determining the tenant characteristic representation of the target tenant in this embodiment;

[0065] Figure 4 This is a flowchart illustrating the process of determining the task feature representation of a cloud database, as provided in this embodiment.

[0066] Figure 5 This is a flowchart illustrating the process of determining the resource demand forecast results of the target tenant in this embodiment.

[0067] Figure 6 This is a partial flowchart of the second resource scheduling method provided in this embodiment;

[0068] Figure 7 This is another part of the flowchart of the second resource scheduling method provided in this embodiment;

[0069] Figure 8 This is a structural block diagram of a resource scheduling device provided in this embodiment;

[0070] Figure 9 This is an internal structural diagram of the computer device provided in this embodiment. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0072] The resource scheduling method provided in this application can be applied to, for example, Figure 1 In the application environment shown, taking a server as an example, server 104 receives an SLA guarantee request and a resource scheduling request for a target tenant from terminal device 102. Server 104 determines the tenant characteristic representation and the cloud database database characteristic representation of the target tenant. Server 104 performs fusion processing on the time-series demand characteristic representation, tenant profile characteristic representation, and task-related characteristic representation to obtain a comprehensive tenant characteristic representation of the target tenant. Based on the comprehensive tenant characteristic representation and the database characteristic representation, the resource demand prediction result of the target tenant is determined. Based on the resource demand prediction result of the target tenant, resource scheduling is performed for the target tenant. The tenant characteristic representation includes the time-series demand characteristic representation, the tenant profile characteristic representation, and the task-related characteristic representation; the database characteristic representation includes the resource characteristic representation and the task characteristic representation.

[0073] The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers. In this application, the server 104 can be a server for a metering automation big data platform. The terminal device 102 can be a smart terminal such as a mobile phone or computer.

[0074] In one embodiment, such as Figure 2 As shown, a resource scheduling method is provided, which can be applied to... Figure 1 Let's take the server in the example of this as an illustration. Figure 2 As shown, it includes the following steps:

[0075] S201, determine the tenant characteristic representation of the target tenant and the database characteristic representation of the cloud database.

[0076] Tenant feature representation includes time-series demand feature representation, tenant profile feature representation, and task-related feature representation; database feature representation includes resource feature representation and task feature representation. The target tenant refers to a user renting cloud database resources. Tenant feature representation is the mathematical representation of the target tenant's tenant-related features. Database feature representation is the mathematical representation of the cloud database's relevant features. Time-series demand feature representation is the mathematical representation of the relationship between time and demand. Tenant profile feature representation is the mathematical representation corresponding to the target tenant's user profile. Task-related feature representation is the mathematical representation of the features associated with the cross-database task currently being executed by the target tenant. All of the above mathematical representations can be vectors.

[0077] Optionally, in this embodiment, tenant-related information of the target tenant and database-related information of the cloud database are obtained. The obtained tenant-related information of the target tenant and database-related information of the cloud database are input into the feature extraction model, and the feature extraction model outputs the tenant feature representation of the target tenant and the database feature representation of the cloud database.

[0078] S202, the time-series demand feature representation, tenant profile feature representation and task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant.

[0079] Optionally, in this embodiment, the time-series demand feature representation, tenant profile feature representation, and task association feature representation are concatenated to obtain an initial comprehensive feature vector. The initial comprehensive feature vector is then input into the LayerNorm layer, which normalizes the initial comprehensive feature vector to obtain the tenant comprehensive feature vector, i.e., the tenant comprehensive feature representation.

[0080] S203, Based on the tenant's comprehensive characteristic representation and the database characteristic representation, determine the resource demand prediction results for the target tenant.

[0081] Among them, the resource demand forecast result refers to the predicted resource demand result of the target tenant in the future period.

[0082] Optionally, in this embodiment, the tenant's comprehensive feature representation and the database feature representation are input into a trained neural network model, and the neural network model outputs the resource demand prediction result of the target tenant.

[0083] S204, Based on the resource demand forecast results of the target tenant, perform resource scheduling for the target tenant.

[0084] Resource scheduling refers to resource scheduling between tenants, such as computing resources, network resources, and storage resources.

[0085] Optionally, in this embodiment, the resource demand information of the target tenant in the future period is determined based on the resource demand prediction results of the target tenant, and resource scheduling is performed in advance for the target tenant based on the resource demand information to ensure the business experience and SLA guarantee of the target tenant.

[0086] The aforementioned resource scheduling method determines the tenant characteristic representation of the target tenant and the database characteristic representation of the cloud database. The tenant characteristic representation includes a time-series demand characteristic representation, a tenant profile characteristic representation, and a task-related characteristic representation; the database characteristic representation includes a resource characteristic representation and a task characteristic representation. The time-series demand characteristic representation, the tenant profile characteristic representation, and the task-related characteristic representation are fused to obtain a comprehensive tenant characteristic representation of the target tenant. Based on the comprehensive tenant characteristic representation and the database characteristic representation, the resource demand prediction result of the target tenant is determined. Based on the resource demand prediction result of the target tenant, resource scheduling is performed for the target tenant. This application considers not only the time-series demand information of the target tenant, but also the tenant profile information, task-related information, and the resource characteristic information and task characteristic information of the cloud database. This allows for more accurate prediction of the target tenant's resource demand, improving the accuracy and rationality of resource scheduling, thereby improving database resource utilization and ensuring the SLA of the target tenant.

[0087] In one embodiment, to more accurately determine the tenant characteristic representation of the target tenant, such as Figure 3 As shown, one optional implementation of determining the tenant characteristic representation of the target tenant in S201 includes:

[0088] S301, Based on the historical resource demand information of the target tenant, determine the temporal demand characteristics of the target tenant.

[0089] Historical resource demand information includes computing resource information, storage resource information, transmission resource information, and I / O resource information.

[0090] Optionally, in this embodiment, the computing resource information includes CPU utilization, memory usage, and cross-library task processing time.

[0091] Optionally, in this embodiment, the storage resource information includes local data storage capacity, cross-database synchronous backup space usage, etc.

[0092] Optionally, in this embodiment, the transmission resource information includes cross-library synchronization bandwidth usage and data transmission rate, etc.

[0093] Optionally, in this embodiment, the IO resource information includes cross-database read / write throughput and IO latency, etc.

[0094] Optionally, in this embodiment, historical resource demand information is decomposed to obtain resource demand decomposition results. Feature extraction is performed on the resource demand decomposition results to obtain resource demand extraction results. Based on the resource demand extraction results, the temporal demand feature representation of the target tenant is determined. Specifically, the resource demand extraction results include low-frequency trend components, periodic components, and high-frequency burst components. The historical resource demand information is decomposed using VMD (Variational Mode Decomposition) to obtain resource demand decomposition results. Feature extraction is performed on the resource demand decomposition results to obtain low-frequency trend components (Trend), periodic components (Seasonal), and high-frequency burst components (Burst). The low-frequency trend components (Trend), periodic components (Seasonal), and high-frequency burst components (Burst) are input into a fully connected layer, which outputs a 64-dimensional embedding vector as the temporal demand feature representation of the target tenant.

[0095] S302, determine the tenant profile feature representation of the target tenant based on the tenant database type, SLA level, and historical credit rating of the target tenant.

[0096] Optionally, in this embodiment, the tenant database type is a core business database, a non-core business database, a third-party application database, or a temporary synchronization database.

[0097] Optionally, in this embodiment, the SLA level is divided into three levels: Gold, Silver, and Bronze. Gold requires a cross-database synchronization delay of ≤500ms. Silver requires a cross-database synchronization delay of ≤1s. Bronze requires a cross-database synchronization delay of ≤3s.

[0098] Optionally, in this embodiment, historical credit rating includes cross-database resource usage compliance rate, data transmission integrity, and synchronization task success rate.

[0099] Optionally, in this embodiment, the tenant database type and SLA level of the target tenant are encoded to obtain an encoded vector. The encoded vector and historical confidence scores are preprocessed to obtain a tenant profile feature representation of the target tenant; the preprocessing methods include encoding and normalization. Specifically, the tenant database type and SLA level of the target tenant are encoded through an embedding layer to obtain a 32-dimensional encoded vector. The historical confidence scores are normalized to obtain normalized features. The encoded vector and normalized features are input to a fully connected layer, which concatenates the encoded vector and normalized features to output a 32-dimensional embedding vector as the tenant profile feature representation of the target tenant.

[0100] S303, determine the task association feature representation of the target tenant based on the type of cross-database task to be executed, the task priority, and the cross-database difficulty value.

[0101] Optionally, in this embodiment, the type of cross-database task to be executed is real-time cross-database transmission or batch cross-database synchronization.

[0102] Optionally, in this embodiment, the cross-database difficulty value can be characterized by the cross-database data volume or the synchronization complexity.

[0103] Optionally, in this embodiment, the current cross-database task type, task priority, and cross-database difficulty value of the target tenant are encoded and normalized before being concatenated to obtain the task association feature representation of the target tenant. Specifically, the current cross-database task type, task priority, and cross-database difficulty value are input into the One-Hot layer, which encodes and normalizes these parameters. The output is then fed into the fully connected layer, where it is concatenated and outputs a 32-dimensional embedding vector as the task association feature representation of the target tenant.

[0104] In this embodiment, the temporal demand characteristic representation of the target tenant is determined based on the target tenant's historical resource demand information; wherein, the historical resource demand information includes computing resource information, storage resource information, transmission resource information, and I / O resource information. The tenant profile characteristic representation of the target tenant is determined based on the target tenant's tenant database type, SLA level, and historical credit rating. The task association characteristic representation of the target tenant is determined based on the target tenant's current pending cross-database task type, task priority, and cross-database difficulty value. Based on this embodiment, the temporal demand characteristic representation, tenant profile characteristic representation, and task association characteristic representation of the target tenant can be determined more accurately.

[0105] In one embodiment, to more accurately determine the database feature representation of the cloud database, such as Figure 4 As shown, one optional implementation of determining the database feature representation of the cloud database in S201 includes:

[0106] S401 extracts features from the database resource information of the cloud database to obtain database feature information.

[0107] The database feature information includes real-time resource load information, resource capacity limit, and historical load fluctuation statistics.

[0108] Optionally, in this embodiment, the database resource information includes resource information of units such as database cluster nodes, cross-database synchronization links, and storage arrays.

[0109] Optionally, the resource load information implemented in this embodiment includes CPU utilization, memory usage, and cross-library link bandwidth utilization.

[0110] Optionally, in this embodiment, the statistical measure in the historical load fluctuation statistics can be the standard deviation or the peak frequency.

[0111] S402, extract features from database feature information to obtain resource feature representation of cloud data.

[0112] Optionally, in this embodiment, the database feature information is standardized to obtain processed database feature information. The processed database feature information is then input into the CNN-1d layer, which extracts local dependency features and outputs a 64-dimensional resource feature vector as the resource feature representation of the cloud data.

[0113] S403 extracts features from the cross-database task types of the cloud database to obtain database task information.

[0114] The database task information includes standard information on task resource consumption, task execution cycle, and probability of sudden occurrences.

[0115] Optionally, in this embodiment, the standard information on task resource consumption includes average bandwidth usage and CPU consumption coefficient, etc.

[0116] Optionally, in this embodiment, based on the cross-database task type of the cloud database, relevant information about the cross-database task type of the cloud database is obtained, and feature extraction is performed on the relevant information to obtain a resource feature representation of the cloud data.

[0117] S404 encodes the database task information to obtain the task feature representation of the cloud database.

[0118] Optionally, in this embodiment, the database task information is sequentially input into the Embedding layer and the fully connected layer, and the fully connected layer outputs a 64-dimensional task feature vector as the task feature representation of the cloud database.

[0119] In this embodiment, feature extraction is performed on the database resource information of the cloud database to obtain database feature information, which includes real-time resource load information, resource capacity limits, and historical load fluctuation statistics. Feature extraction is then performed on the database feature information to obtain a resource feature representation of the cloud data. Feature extraction is also performed on the cross-database task types of the cloud database to obtain database task information, which includes task resource consumption standards, task execution cycle, and burst probability. The database task information is then encoded to obtain a task feature representation of the cloud database. This embodiment effectively improves the accuracy of the obtained database feature representation.

[0120] In one embodiment, to obtain more accurate resource demand prediction results for the target tenant, such as... Figure 5 As shown, one optional implementation method for determining the resource demand forecast result of a target tenant based on the tenant comprehensive feature representation and the database feature representation includes:

[0121] S501, the tenant comprehensive feature representation and the task feature representation are fused to obtain the first fused feature representation.

[0122] Optionally, in this embodiment, a feature fusion network is constructed based on a three-layer gated recurrent unit (GRU). Specifically, based on the first layer of the three-layer gated recurrent unit, the tenant comprehensive feature representation and the task feature representation are fused to obtain a first fused feature representation.

[0123] S502, the resource feature representation and the resource competition feature representation are fused to obtain the second fused feature representation.

[0124] Optionally, in this embodiment, the resource feature representation and the resource competition feature representation are fused based on the second layer of the three-layer gated loop unit to obtain the second fused feature representation.

[0125] Optionally, in this embodiment, one possible implementation for determining the resource contention feature representation is to determine the matching degree between the target tenant and each resource node based on the historical demand characteristics of the target tenant and the historical load information of each resource node in the cloud database. The resource contention feature representation is then determined based on the maximum matching degree and the average matching degree. The historical demand characteristics can be obtained based on historical demand information. In this embodiment, another possible implementation for determining the matching degree between the target tenant and each resource node based on the historical demand characteristics is to determine the demand value based on the historical demand characteristics. The remaining resource capacity of each resource node is determined based on the historical load information of each resource node. The ratio of the demand value to the remaining resource capacity is used as the matching degree between the target tenant and that resource node. Then, the maximum matching degree and the average matching degree are taken as the resource contention feature representation.

[0126] S503, based on SLA constraint feature representation, demand dependency feature representation, first fusion feature representation and second fusion feature representation, obtains multimodal fusion feature representation.

[0127] Optionally, in this embodiment, based on the third layer of the three-layer gated loop unit, the SLA constraint feature representation, demand dependency feature representation, first fusion feature representation, and second fusion feature representation are fused to obtain a multimodal fusion feature representation. The fusion processing method can be weighted summation.

[0128] Optionally, in this embodiment, the SLA constraint feature representation can be determined by using a mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database's resource nodes. Specifically, the mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database's resource nodes is used as the SLA constraint feature.

[0129] Optionally, in this embodiment, an optional implementation method for determining the demand dependency feature representation is as follows: Based on the historical demand characteristics of the target tenant and the historical demand characteristics of each first tenant, determine the time-lag correlation coefficient between the target tenant and each first tenant; wherein, the first tenant is the tenant other than the target tenant among all tenants of the cloud database. Sort the time-lag correlation coefficients of each first tenant in descending order to obtain a sorting result. Based on the sorting result, determine the average coefficient of the first preset number (e.g., 10) time-lag correlation coefficients. Establish a correlation between the average coefficient and the historical demand characteristics of the target tenant to obtain the demand dependency feature representation. Alternatively, in this embodiment, an optional implementation method for establishing a correlation between the average coefficient and the historical demand characteristics of the target tenant to obtain the demand dependency feature representation is to perform feature concatenation on the average coefficient feature and the historical demand characteristics of the target tenant to obtain the demand dependency feature representation.

[0130] S504. Based on the multimodal fusion feature representation, determine the resource demand prediction results for the target tenant.

[0131] Optionally, in this embodiment, the predicted value of periodic demand is determined based on an LSTM (Long Short-Term Memory) network and multimodal fusion feature representation. Specifically, a stacked LSTM network is used to capture the quasi-periodic features of demand: the temporal trend component and periodic component are input from the multimodal fusion features, and a long-term temporal dependency is modeled through two layers of bidirectional LSTM (64 neurons per layer) to capture multi-periodic patterns. The LSTM network introduces a time attention mechanism, assigning adaptive weights to features at different time steps, which can be represented by the following formula: In the formula, h t Let W be the hidden state of the LSTM at time t. a b a These are learnable parameters. Finally, the predicted periodic demand is output through a fully connected layer.

[0132] Optionally, in this embodiment, a DNN (Deep Neural Network) is used to determine the predicted value of sudden demand. Specifically, a task-enhanced self-attention network is designed to capture sudden demand and dynamic fluctuations: high-frequency sudden components, task features, and SLA constraint features are input from the multimodal fusion features, and the nonlinear correlation between features is modeled through a multi-head self-attention mechanism (4 attention heads); wherein, the attention calculation formula is: Q = W q *X, K = W k *X, V =W v *X;Attention (Q,K,V) = Softmax (QK T )*V; where X is the input feature matrix, W q W k W v These are learnable parameters; finally, high-order fluctuation features are extracted through two fully connected layers of the DNN network to output the predicted value of sudden demand.

[0133] Optionally, in this embodiment, the resource demand forecast result is determined based on the cyclical demand forecast value and the burst demand forecast value. Specifically, the resource demand forecast result can be determined using the following formula: ; where Y pred Y represents the resource demand forecast result; λ represents the fusion weight; P Y represents the cyclical demand forecast; v This represents the forecast value for sudden demand.

[0134] Based on the above embodiments, an optional implementation method for resource scheduling for target tenants, according to the resource demand prediction results, is as follows: First, establish the state space and reward function. Second, establish the action space and policy network. In this embodiment, the state space includes the tenant's cross-database demand prediction value, the real-time load of cloud database resources (CPU / memory / cross-database bandwidth / IO / connection pool), the tenant's SLA level, and the resource contention intensity, totaling a 32-dimensional state vector. The reward function adopts a multi-objective weighted reward to balance various indicators. The expression of the reward function is: In the formula, R U = 1 - Cloud database resource idle rate (average idle percentage of compute / cross-database bandwidth / storage / IO / connection pool); R SLA = Tenant cross-database task compliance rate (the percentage of cross-database synchronization latency ≤ SLA commitment value, data transmission integrity ≥ 99.99%); R F= 1 - Tenant resource allocation variance coefficient (to avoid core tenants' cross-database resource occupation or non-core tenants' insufficient resources); ω1, ω2, and ω3 are adaptive weights. The action space includes 6 types of discrete actions and 3 types of continuous actions, constituting a hybrid action space: Discrete actions: resource node switching (3 selectable nodes), task priority adjustment (3 levels); Continuous actions: CPU / memory quota adjustment ratio (0.5-2 times), cross-database bandwidth allocation ratio (0.3-1.5 times), synchronization timing offset (-120~120 minutes). The policy network adopts a fully connected neural network to construct an Actor-Critic architecture; where the Actor network includes 3 fully connected layers (128→64→output layer), outputting the action probability distribution and continuous action values; the Critic network: a dual Critic architecture (to avoid overestimation), with 128 neurons per layer, outputting the state-action value Q value. Therefore, in this embodiment, based on the resource demand prediction results, state space, reward function, action space, and policy network, resource scheduling for the target tenant can be executed efficiently and accurately.

[0135] In this embodiment, the tenant comprehensive feature representation and task feature representation are fused to obtain a first fused feature representation. The resource feature representation and resource contention feature representation are fused to obtain a second fused feature representation. Based on the SLA constraint feature representation, demand dependency feature representation, the first fused feature representation, and the second fused feature representation, a multimodal fused feature representation is obtained. Based on the multimodal fused feature representation, the resource demand prediction result for the target tenant is determined. This embodiment effectively improves the determined resource demand prediction result, thereby enabling more accurate resource scheduling for the target tenant, improving resource utilization, and ensuring the SLA of the target tenant is guaranteed.

[0136] In one embodiment, such as Figure 6 , Figure 7 As shown, an optional implementation of a resource scheduling method includes:

[0137] S601 decomposes the historical resource demand information to obtain the resource demand decomposition result. The historical resource demand information includes computing resource information, storage resource information, transmission resource information, and I / O resource information.

[0138] S602, feature extraction is performed on the resource demand decomposition results to obtain the resource demand extraction results. These results include low-frequency trend components, periodic components, and high-frequency burst components.

[0139] S603, based on the resource requirement extraction results, determine the temporal requirement characteristics of the target tenant.

[0140] S604 encodes the tenant database type and SLA level of the target tenant to obtain an encoding vector.

[0141] S605 involves preprocessing the encoded vector and historical confidence scores to obtain the tenant profile feature representation of the target tenant. The preprocessing methods include encoding and normalization.

[0142] S606, Based on the current cross-database task type to be executed, task priority, and cross-database difficulty value of the target tenant, determine the task association feature representation of the target tenant.

[0143] S607 extracts features from the cloud database's resource information to obtain database feature information. This database feature information includes real-time resource load information, resource capacity limits, and historical load fluctuation statistics.

[0144] S608 extracts features from the database feature information to obtain a resource feature representation of the cloud data.

[0145] S609 performs feature extraction on cross-database task types in the cloud database to obtain database task information. This database task information includes standard information on task resource consumption, task execution cycle, and burst probability.

[0146] S610 encodes the database task information to obtain the task feature representation of the cloud database.

[0147] S611 integrates the temporal demand feature representation, tenant profile feature representation, and task association feature representation to obtain the comprehensive tenant feature representation of the target tenant.

[0148] S612, based on the historical demand characteristics of the target tenant and the historical demand characteristics of each first tenant, determine the time lag correlation coefficient between the target tenant and each first tenant. Here, the first tenant refers to all tenants of the cloud database other than the target tenant.

[0149] S613, sort the time delay correlation coefficients of each first tenant in descending order to obtain the sorting result.

[0150] S614, Based on the sorting results, determine the mean value of the coefficients of the first preset number of time-delay correlation coefficients.

[0151] S615 establishes a correlation between the mean coefficient and the historical demand characteristics of the target tenant to obtain a demand dependency characteristic representation.

[0152] S616 determines the matching degree between the target tenant and each resource node based on the historical demand characteristics of the target tenant and the historical load information of each resource node in the cloud database.

[0153] S617, Based on the maximum matching degree and the average matching degree, determine the resource competition feature representation.

[0154] S618 determines the SLA constraint characteristics based on the mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database's resource nodes.

[0155] S619, the tenant comprehensive feature representation and the task feature representation are fused to obtain the first fused feature representation.

[0156] S620, the resource feature representation and the resource competition feature representation are fused to obtain the second fused feature representation.

[0157] S621, based on SLA constraint feature representation, demand dependency feature representation, first fusion feature representation and second fusion feature representation, a multimodal fusion feature representation is obtained.

[0158] S622, Based on the multimodal fusion feature representation, determine the resource demand prediction results for the target tenant.

[0159] S623, based on the resource demand forecast results of the target tenant, performs resource scheduling for the target tenant.

[0160] This embodiment determines the tenant feature representation of the target tenant and the database feature representation of the cloud database. The tenant feature representation includes time-series demand feature representation, tenant profile feature representation, and task-related feature representation; the database feature representation includes resource feature representation and task feature representation. The time-series demand feature representation, tenant profile feature representation, and task-related feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant. Based on the comprehensive tenant feature representation and the database feature representation, the resource demand prediction result of the target tenant is determined. Based on the resource demand prediction result of the target tenant, resource scheduling is performed for the target tenant. This application considers not only the time-series demand information of the target tenant, but also the tenant profile information, task-related information, and the resource feature information and task feature information of the cloud database. This allows for more accurate prediction of the target tenant's resource demand, improves the accuracy and rationality of resource scheduling, and thus improves database resource utilization and achieves SLA assurance for the target tenant.

[0161] Based on the same inventive concept, this application also provides a resource scheduling apparatus for implementing the resource scheduling method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, specific limitations in a resource scheduling apparatus embodiment provided below can be found in the limitations of the resource scheduling method described above, and will not be repeated here.

[0162] In one embodiment, such as Figure 8 As shown, a resource scheduling device 1 is provided, comprising:

[0163] The first determining module 10 is used to determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes the time-series demand feature representation, the tenant profile feature representation, and the task association feature representation; the database feature representation includes the resource feature representation and the task feature representation;

[0164] The second determining module 20 is used to perform fusion processing on the time-series demand feature representation, the tenant profile feature representation, and the task association feature representation to obtain the comprehensive tenant feature representation of the target tenant;

[0165] The third determination module 30 is used to determine the resource demand prediction results of the target tenant based on the tenant comprehensive feature representation and the database feature representation;

[0166] The scheduling module 40 is used to schedule resources for the target tenant based on the predicted resource demand of the target tenant.

[0167] In one embodiment, the first determining module is further specifically used for:

[0168] Based on the historical resource demand information of the target tenant, the temporal demand characteristics of the target tenant are determined; wherein, the historical resource demand information includes computing resource information, storage resource information, transmission resource information and IO resource information;

[0169] Based on the tenant database type, SLA level, and historical credit rating of the target tenant, determine the tenant profile features of the target tenant;

[0170] Based on the current pending cross-database task type, task priority, and cross-database difficulty value of the target tenant, determine the task association feature representation of the target tenant.

[0171] In one embodiment, the first determining module is further specifically used for:

[0172] Historical resource demand information is decomposed to obtain resource demand decomposition results;

[0173] Feature extraction is performed on the resource demand decomposition results to obtain resource demand extraction results; the resource demand extraction results include low-frequency trend components, periodic components and high-frequency burst components;

[0174] Based on the resource demand extraction results, the temporal demand characteristics of the target tenant are determined.

[0175] In one embodiment, the first determining module is further specifically used for:

[0176] The target tenant's tenant database type and SLA level are encoded to obtain an encoding vector;

[0177] The encoded vector and historical confidence scores are preprocessed to obtain the tenant profile feature representation of the target tenant; the preprocessing methods include encoding and normalization.

[0178] In one embodiment, the first determining module is further specifically used for:

[0179] Feature extraction is performed on the database resource information of the cloud database to obtain database feature information; among which, the database feature information includes real-time resource load information, resource capacity limit and historical load fluctuation statistics;

[0180] Feature extraction is performed on the database feature information to obtain the resource feature representation of cloud data;

[0181] Feature extraction is performed on cross-database task types in cloud databases to obtain database task information; the database task information includes standard information on task resource consumption, task execution cycle, and burst probability.

[0182] The database task information is encoded to obtain the task feature representation of the cloud database.

[0183] In one embodiment, the third determining module is further specifically used for:

[0184] The tenant comprehensive feature representation and the task feature representation are fused to obtain the first fused feature representation;

[0185] The resource feature representation and the resource competition feature representation are fused to obtain a second fused feature representation;

[0186] Based on SLA constraint feature representation, demand dependency feature representation, first fusion feature representation and second fusion feature representation, a multimodal fusion feature representation is obtained;

[0187] Based on the multimodal fusion feature representation, the resource demand prediction results for the target tenant are determined.

[0188] In one embodiment, the third determining module is further specifically used for:

[0189] Based on the historical demand characteristics of the target tenant and the historical demand characteristics of each first tenant, the time lag correlation coefficient between the target tenant and each first tenant is determined; where the first tenant is the tenant other than the target tenant among all tenants of the cloud database.

[0190] The time-delay correlation coefficients of each first tenant are sorted in descending order to obtain the sorting results;

[0191] Based on the sorting results, determine the mean value of the coefficients of the first preset number of time-delay correlation coefficients;

[0192] By establishing a correlation between the mean coefficient and the historical demand characteristics of the target tenant, a demand dependency feature representation is obtained.

[0193] In one embodiment, the third determining module is further specifically used for:

[0194] Based on the historical demand characteristics of the target tenant and the historical load information of each resource node in the cloud database, determine the matching degree between the target tenant and each resource node;

[0195] Based on the maximum matching degree and the average matching degree, the resource competition feature representation is determined.

[0196] In one embodiment, the third determining module is further specifically used for:

[0197] Based on the mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database resource nodes, the SLA constraint characteristics are determined.

[0198] Each module in the aforementioned resource scheduling device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0199] In one embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 9As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a resource scheduling method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0200] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0201] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0202] Determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation and task association feature representation; the database feature representation includes resource feature representation and task feature representation;

[0203] The time-series demand feature representation, tenant profile feature representation, and task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant;

[0204] Based on the tenant's comprehensive characteristic representation and the database characteristic representation, the resource demand prediction results for the target tenant are determined;

[0205] Based on the resource demand forecast results of the target tenant, resource scheduling is performed for the target tenant.

[0206] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining a tenant characteristic representation of the target tenant, including:

[0207] Based on the historical resource demand information of the target tenant, the temporal demand characteristics of the target tenant are determined; wherein, the historical resource demand information includes computing resource information, storage resource information, transmission resource information and IO resource information;

[0208] Based on the tenant database type, SLA level, and historical credit rating of the target tenant, determine the tenant profile features of the target tenant;

[0209] Based on the current pending cross-database task type, task priority, and cross-database difficulty value of the target tenant, determine the task association feature representation of the target tenant.

[0210] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining the temporal demand characteristic representation of the target tenant based on the target tenant's historical resource demand information, including:

[0211] Historical resource demand information is decomposed to obtain resource demand decomposition results;

[0212] Feature extraction is performed on the resource demand decomposition results to obtain resource demand extraction results; the resource demand extraction results include low-frequency trend components, periodic components and high-frequency burst components;

[0213] Based on the resource demand extraction results, the temporal demand characteristics of the target tenant are determined.

[0214] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining a tenant profile feature representation of the target tenant based on the target tenant's tenant database type, SLA level, and historical credit rating, including:

[0215] The target tenant's tenant database type and SLA level are encoded to obtain an encoding vector;

[0216] The encoded vector and historical confidence scores are preprocessed to obtain the tenant profile feature representation of the target tenant; the preprocessing methods include encoding and normalization.

[0217] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining a database feature representation of the cloud database, including:

[0218] Feature extraction is performed on the database resource information of the cloud database to obtain database feature information; among which, the database feature information includes real-time resource load information, resource capacity limit and historical load fluctuation statistics;

[0219] Feature extraction is performed on the database feature information to obtain the resource feature representation of cloud data;

[0220] Feature extraction is performed on cross-database task types in cloud databases to obtain database task information; the database task information includes standard information on task resource consumption, task execution cycle, and burst probability.

[0221] The database task information is encoded to obtain the task feature representation of the cloud database.

[0222] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining the resource demand prediction result of the target tenant based on the tenant comprehensive feature representation and the database feature representation, including:

[0223] The tenant comprehensive feature representation and the task feature representation are fused to obtain the first fused feature representation;

[0224] The resource feature representation and the resource competition feature representation are fused to obtain a second fused feature representation;

[0225] Based on SLA constraint feature representation, demand dependency feature representation, first fusion feature representation and second fusion feature representation, a multimodal fusion feature representation is obtained;

[0226] Based on the multimodal fusion feature representation, the resource demand prediction results for the target tenant are determined.

[0227] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0228] Based on the historical demand characteristics of the target tenant and the historical demand characteristics of each first tenant, the time lag correlation coefficient between the target tenant and each first tenant is determined; where the first tenant is the tenant other than the target tenant among all tenants of the cloud database.

[0229] The time-delay correlation coefficients of each first tenant are sorted in descending order to obtain the sorting results;

[0230] Based on the sorting results, determine the mean value of the coefficients of the first preset number of time-delay correlation coefficients;

[0231] By establishing a correlation between the mean coefficient and the historical demand characteristics of the target tenant, a demand dependency feature representation is obtained.

[0232] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0233] Based on the historical demand characteristics of the target tenant and the historical load information of each resource node in the cloud database, determine the matching degree between the target tenant and each resource node;

[0234] Based on the maximum matching degree and the average matching degree, the resource competition feature representation is determined.

[0235] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0236] Based on the mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database resource nodes, the SLA constraint characteristics are determined.

[0237] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0238] Determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation and task association feature representation; the database feature representation includes resource feature representation and task feature representation;

[0239] The time-series demand feature representation, tenant profile feature representation, and task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant;

[0240] Based on the tenant's comprehensive characteristic representation and the database characteristic representation, the resource demand prediction results for the target tenant are determined;

[0241] Based on the resource demand forecast results of the target tenant, resource scheduling is performed for the target tenant.

[0242] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining a tenant characteristic representation of the target tenant, including:

[0243] Based on the historical resource demand information of the target tenant, the temporal demand characteristics of the target tenant are determined; wherein, the historical resource demand information includes computing resource information, storage resource information, transmission resource information and IO resource information;

[0244] Based on the tenant database type, SLA level, and historical credit rating of the target tenant, determine the tenant profile features of the target tenant;

[0245] Based on the current pending cross-database task type, task priority, and cross-database difficulty value of the target tenant, determine the task association feature representation of the target tenant.

[0246] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining the temporal demand characteristic representation of the target tenant based on the target tenant's historical resource demand information, including:

[0247] Historical resource demand information is decomposed to obtain resource demand decomposition results;

[0248] Feature extraction is performed on the resource demand decomposition results to obtain resource demand extraction results; the resource demand extraction results include low-frequency trend components, periodic components and high-frequency burst components;

[0249] Based on the resource demand extraction results, the temporal demand characteristics of the target tenant are determined.

[0250] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining a tenant profile feature representation of the target tenant based on the target tenant's tenant database type, SLA level, and historical credit score, including:

[0251] The target tenant's tenant database type and SLA level are encoded to obtain an encoding vector;

[0252] The encoded vector and historical confidence scores are preprocessed to obtain the tenant profile feature representation of the target tenant; the preprocessing methods include encoding and normalization.

[0253] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining a database feature representation of the cloud database, including:

[0254] Feature extraction is performed on the database resource information of the cloud database to obtain database feature information; among which, the database feature information includes real-time resource load information, resource capacity limit and historical load fluctuation statistics;

[0255] Feature extraction is performed on the database feature information to obtain the resource feature representation of cloud data;

[0256] Feature extraction is performed on cross-database task types in cloud databases to obtain database task information; the database task information includes standard information on task resource consumption, task execution cycle, and burst probability.

[0257] The database task information is encoded to obtain the task feature representation of the cloud database.

[0258] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining the resource demand prediction result of the target tenant based on the tenant comprehensive feature representation and the database feature representation, including:

[0259] The tenant comprehensive feature representation and the task feature representation are fused to obtain the first fused feature representation;

[0260] The resource feature representation and the resource competition feature representation are fused to obtain a second fused feature representation;

[0261] Based on SLA constraint feature representation, demand dependency feature representation, first fusion feature representation and second fusion feature representation, a multimodal fusion feature representation is obtained;

[0262] Based on the multimodal fusion feature representation, the resource demand prediction results for the target tenant are determined.

[0263] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0264] Based on the historical demand characteristics of the target tenant and the historical demand characteristics of each first tenant, the time lag correlation coefficient between the target tenant and each first tenant is determined; where the first tenant is the tenant other than the target tenant among all tenants of the cloud database.

[0265] The time-delay correlation coefficients of each first tenant are sorted in descending order to obtain the sorting results;

[0266] Based on the sorting results, determine the mean value of the coefficients of the first preset number of time-delay correlation coefficients;

[0267] By establishing a correlation between the mean coefficient and the historical demand characteristics of the target tenant, a demand dependency feature representation is obtained.

[0268] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0269] Based on the historical demand characteristics of the target tenant and the historical load information of each resource node in the cloud database, determine the matching degree between the target tenant and each resource node;

[0270] Based on the maximum matching degree and the average matching degree, the resource competition feature representation is determined.

[0271] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0272] Based on the mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database resource nodes, the SLA constraint characteristics are determined.

[0273] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0274] Determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation and task association feature representation; the database feature representation includes resource feature representation and task feature representation;

[0275] The time-series demand feature representation, tenant profile feature representation, and task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant;

[0276] Based on the tenant's comprehensive characteristic representation and the database characteristic representation, the resource demand prediction results for the target tenant are determined;

[0277] Based on the resource demand forecast results of the target tenant, resource scheduling is performed for the target tenant.

[0278] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining a tenant characteristic representation of the target tenant, including:

[0279] Based on the historical resource demand information of the target tenant, the temporal demand characteristics of the target tenant are determined; wherein, the historical resource demand information includes computing resource information, storage resource information, transmission resource information and IO resource information;

[0280] Based on the tenant database type, SLA level, and historical credit rating of the target tenant, determine the tenant profile features of the target tenant;

[0281] Based on the current pending cross-database task type, task priority, and cross-database difficulty value of the target tenant, determine the task association feature representation of the target tenant.

[0282] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining the temporal demand characteristic representation of the target tenant based on the target tenant's historical resource demand information, including:

[0283] Historical resource demand information is decomposed to obtain resource demand decomposition results;

[0284] Feature extraction is performed on the resource demand decomposition results to obtain resource demand extraction results; the resource demand extraction results include low-frequency trend components, periodic components and high-frequency burst components;

[0285] Based on the resource demand extraction results, the temporal demand characteristics of the target tenant are determined.

[0286] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining a tenant profile feature representation of the target tenant based on the target tenant's tenant database type, SLA level, and historical credit score, including:

[0287] The target tenant's tenant database type and SLA level are encoded to obtain an encoding vector;

[0288] The encoded vector and historical confidence scores are preprocessed to obtain the tenant profile feature representation of the target tenant; the preprocessing methods include encoding and normalization.

[0289] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining a database feature representation of the cloud database, including:

[0290] Feature extraction is performed on the database resource information of the cloud database to obtain database feature information; among which, the database feature information includes real-time resource load information, resource capacity limit and historical load fluctuation statistics;

[0291] Feature extraction is performed on the database feature information to obtain the resource feature representation of cloud data;

[0292] Feature extraction is performed on cross-database task types in cloud databases to obtain database task information; the database task information includes standard information on task resource consumption, task execution cycle, and burst probability.

[0293] The database task information is encoded to obtain the task feature representation of the cloud database.

[0294] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: determining the resource demand prediction result of the target tenant based on the tenant comprehensive feature representation and the database feature representation, including:

[0295] The tenant comprehensive feature representation and the task feature representation are fused to obtain the first fused feature representation;

[0296] The resource feature representation and the resource competition feature representation are fused to obtain a second fused feature representation;

[0297] Based on SLA constraint feature representation, demand dependency feature representation, first fusion feature representation and second fusion feature representation, a multimodal fusion feature representation is obtained;

[0298] Based on the multimodal fusion feature representation, the resource demand prediction results for the target tenant are determined.

[0299] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0300] Based on the historical demand characteristics of the target tenant and the historical demand characteristics of each first tenant, the time lag correlation coefficient between the target tenant and each first tenant is determined; where the first tenant is the tenant other than the target tenant among all tenants of the cloud database.

[0301] The time-delay correlation coefficients of each first tenant are sorted in descending order to obtain the sorting results;

[0302] Based on the sorting results, determine the mean value of the coefficients of the first preset number of time-delay correlation coefficients;

[0303] By establishing a correlation between the mean coefficient and the historical demand characteristics of the target tenant, a demand dependency feature representation is obtained.

[0304] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0305] Based on the historical demand characteristics of the target tenant and the historical load information of each resource node in the cloud database, determine the matching degree between the target tenant and each resource node;

[0306] Based on the maximum matching degree and the average matching degree, the resource competition feature representation is determined.

[0307] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0308] Based on the mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database resource nodes, the SLA constraint characteristics are determined.

[0309] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0310] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0311] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A resource scheduling method, characterized in that, The method includes: Determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes time-series demand feature representation, tenant profile feature representation, and task association feature representation; the database feature representation includes resource feature representation and task feature representation; The time-series demand feature representation, the tenant profile feature representation, and the task association feature representation are fused to obtain the comprehensive tenant feature representation of the target tenant; Based on the tenant's comprehensive feature representation and the database feature representation, the resource demand prediction result of the target tenant is determined; Based on the resource demand forecast results of the target tenant, resource scheduling is performed for the target tenant.

2. The method according to claim 1, characterized in that, The tenant characteristic representation for determining the target tenant includes: Based on the historical resource demand information of the target tenant, the temporal demand characteristics of the target tenant are determined; wherein, the historical resource demand information includes computing resource information, storage resource information, transmission resource information, and I / O resource information; Based on the tenant database type, service level agreement (SLA) level, and historical credit rating of the target tenant, determine the tenant profile feature representation of the target tenant; Based on the current cross-database task type, task priority, and cross-database difficulty value of the target tenant, the task association feature representation of the target tenant is determined.

3. The method according to claim 2, characterized in that, The step of determining the temporal demand characteristic representation of the target tenant based on the target tenant's historical resource demand information includes: The historical resource demand information is decomposed to obtain the resource demand decomposition result; Feature extraction is performed on the resource demand decomposition results to obtain resource demand extraction results; wherein, the resource demand extraction results include low-frequency trend components, periodic components and high-frequency burst components; Based on the resource demand extraction results, the temporal demand characteristics of the target tenant are determined.

4. The method according to claim 2, characterized in that, The step of determining the tenant profile feature representation of the target tenant based on the tenant database type, SLA level, and historical credit score of the target tenant includes: The target tenant's database type and SLA level are encoded to obtain an encoding vector; The encoded vector and the historical confidence score are preprocessed to obtain the tenant profile feature representation of the target tenant; wherein the preprocessing method includes encoding processing and normalization processing.

5. The method according to claim 1, characterized in that, Determine the database characteristic representation of the cloud database, including: Feature extraction is performed on the database resource information of the cloud database to obtain database feature information; wherein, the database feature information includes real-time resource load information, resource capacity limit and historical load fluctuation statistics; Feature extraction is performed on the database feature information to obtain the resource feature representation of the cloud data; Feature extraction is performed on the cross-database task types of the cloud database to obtain database task information; wherein, the database task information includes task resource consumption standard information, task execution cycle and burst probability; The database task information is encoded to obtain the task feature representation of the cloud database.

6. The method according to claim 1, characterized in that, The step of determining the resource demand prediction result of the target tenant based on the tenant comprehensive feature representation and the database feature representation includes: The tenant comprehensive feature representation and the task feature representation are fused to obtain a first fused feature representation; The resource feature representation and the resource competition feature representation are fused to obtain a second fused feature representation; Based on the SLA constraint feature representation, demand dependency feature representation, the first fusion feature representation, and the second fusion feature representation, a multimodal fusion feature representation is obtained; Based on the multimodal fusion feature representation, the resource demand prediction result of the target tenant is determined.

7. The method according to claim 6, characterized in that, The method further includes: Based on the historical demand characteristics of the target tenant and the historical demand characteristics of each first tenant, the time lag correlation coefficient between the target tenant and each first tenant is determined; wherein, the first tenant is the tenant other than the target tenant among all tenants of the cloud database; The time delay correlation coefficients of each of the first tenants are sorted in descending order to obtain the sorting results; Based on the sorting results, determine the mean value of the coefficients of the first preset number of time-delay correlation coefficients; A correlation is established between the mean of the coefficients and the historical demand characteristics of the target tenant to obtain the demand dependency feature representation.

8. The method according to claim 6, characterized in that, The method further includes: Based on the historical demand characteristics of the target tenant and the historical load information of each resource node in the cloud database, the matching degree between the target tenant and each resource node is determined; Based on the maximum matching degree and the average matching degree, the resource competition feature representation is determined.

9. The method according to claim 6, characterized in that, The method further includes: Based on the mapping coefficient between the target tenant's SLA level and the service capabilities of the cloud database's resource nodes, the SLA constraint characteristics are determined.

10. A resource scheduling device, characterized in that, The device includes: The first determining module is used to determine the tenant feature representation of the target tenant and the database feature representation of the cloud database; wherein, the tenant feature representation includes a time-series demand feature representation, a tenant profile feature representation, and a task association feature representation; the database feature representation includes a resource feature representation and a task feature representation; The second determining module is used to perform fusion processing on the time-series demand feature representation, the tenant profile feature representation, and the task association feature representation to obtain the comprehensive tenant feature representation of the target tenant; The third determining module is used to determine the resource demand prediction result of the target tenant based on the tenant comprehensive feature representation and the database feature representation; The scheduling module is used to schedule resources for the target tenant based on the predicted resource demand of the target tenant.

Images