Resource scheduling method and database

By predicting the load matching between computing nodes and resource groups, intelligent scheduling of resource groups is realized, which solves the problem of multi-tenant resource competition in cloud computing, and improves resource utilization and service quality.

WO2024230706A1PCT designated stage expired Publication Date: 2024-11-14ALIBABA CLOUD COMPUTING CO LTD

Patent Information

Application Number
PCT/CN2024/091531
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-08
Filing Date
2024-05-07
Publication Date
2024-11-14

AI Technical Summary

Technical Problem

In a cloud computing environment, when multi-tenant resource instances are deployed in a hybrid resource pool, there is a complex situation of resource competition, which makes it difficult to ensure service quality.

Method used

By predicting the resource load indicators when the resource group to be dispatched to the computing node, the matching degree between the node and the resource group is calculated, the target computing nodes that meet the preset scheduling conditions are selected, and the intelligent scheduling of the resource group is realized, and the multi-tenant The probability of loads affecting each other.

Benefits of technology

It improves resource utilization, reduces platform costs, improves service quality, and realizes resource over-issuance and load balancing in the resource pool.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2024091531_14112024_PF_FP_ABST
    Figure CN2024091531_14112024_PF_FP_ABST
Patent Text Reader

Abstract

Embodiments of the present description provide a resource scheduling method and a database. The resource scheduling method comprises: for a plurality of computing nodes, when a resource group to be scheduled is deployed onto the computing nodes, predicting resource load index predicted values of said resource group; for the plurality of computing nodes, calculating the matching degrees between the computing nodes and said resource group by utilizing resource load index actual values of the computing nodes and the resource load index predicted values of said resource group when said resource group is deployed onto the computing nodes; for said resource group, screening for a target computing node meeting a preset scheduling condition from the plurality of computing nodes by utilizing the matching degrees between the computing nodes and said resource group; and scheduling said resource group to the target computing node.
Need to check novelty before this filing date? Find Prior Art

Description

Resource scheduling method and database

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on May 8, 2023, with application number 202310519544.8 and application name “Resource Scheduling Method and Database”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The embodiments of this specification relate to the field of computer technology, and in particular to a resource scheduling method. Background Art

[0003] Cloud-native products built on cloud infrastructure can fully leverage cloud resources to provide users with instantaneous, massive data processing capabilities and a satisfying user experience. To use cloud resources for data processing, resource groups contained in resource instances need to be deployed to compute nodes. A resource instance can be understood as a computing module virtualized from cloud computing resources and provided to tenants. A resource group is the most basic execution unit in cloud computing and typically includes resources such as processors, storage, network, and logs. Therefore, a resource instance typically contains one or more resource groups. For example, in a container-based computing cluster, a resource group can be understood as a container group, also known as a pod.

[0004] Currently, when creating resource instances for customers, the resource groups contained in the user's resource instances are typically evenly distributed and deployed across the compute nodes in the resource pool based on resource supply and demand. However, in real-world applications, the mixed deployment of resource instances from numerous tenants within the resource pool creates complex resource contention, making it difficult to guarantee service quality.

[0005] Summary of the Invention

[0006] In view of this, embodiments of this specification provide a resource scheduling method. One or more embodiments of this specification also relate to a database, a computing device, a computer-readable storage medium, and a computer program to address technical deficiencies in the prior art.

[0007] According to a first aspect of an embodiment of the present specification, a resource scheduling method is provided, comprising: for multiple computing nodes, predicting predicted values ​​of resource load indicators of a resource group to be scheduled when the resource group to be scheduled is deployed on the computing node; for the multiple computing nodes, calculating a matching degree between the computing node and the resource group to be scheduled using the actual value of the resource load indicator of the computing node and the predicted value of the resource load indicator of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node; using the matching degree between the computing node and the resource group to be scheduled, screening out a target computing node that meets preset scheduling conditions for the resource group to be scheduled from the multiple computing nodes; and scheduling the resource group to be scheduled to the target computing node.

[0008] According to the second aspect of the embodiments of this specification, a database is provided, comprising: a database service node, a resource scheduling service node and a resource pool, wherein the resource pool comprises a plurality of computing nodes; the database service node is configured to receive a resource instance creation request, create a resource instance according to the resource instance creation request, and send a scheduling request to the resource scheduling service node, wherein the scheduling request is used to indicate scheduling of one or more resource groups of the resource instance; the resource scheduling service node is configured to receive the scheduling request and schedule the resource group indicated by the scheduling request according to the resource scheduling method as described in any embodiment of this specification.

[0009] According to the third aspect of the embodiments of this specification, a computing device is provided, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned resource scheduling method.

[0010] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the above-mentioned resource scheduling method are implemented.

[0011] According to a fifth aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned resource scheduling method.

[0012] One embodiment of the present specification implements a resource scheduling method. Since the method predicts, for multiple computing nodes, the predicted value of the resource load indicator of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, for the multiple computing nodes, the actual value of the resource load indicator of the computing node and the predicted value of the resource load indicator of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node are used to calculate the matching degree between the computing node and the resource group to be scheduled. The matching degree between the computing node and the resource group to be scheduled is used to screen out a target computing node that meets a preset scheduling condition for the resource group to be scheduled from the multiple computing nodes, and schedule the resource group to be scheduled to the target computing node. It can be seen that this method targets the mixed deployment of resource instances of many tenants in the resource pool, and there is a relatively complex resource contention situation. According to the correlation between the load of the resource group and the computing node, the resource load index prediction value of the resource group to be scheduled is predicted when the resource group to be scheduled is deployed on each computing node. The matching degree between the resource load index prediction value of the resource group to be scheduled and the computing node is calculated based on the resource load index prediction value, and then the target computing node that meets the preset scheduling conditions is screened out from multiple computing nodes using the matching degree. This is equivalent to breaking up the deployment of the resource group based on the load index prediction value, reducing the probability of mutual influence between multiple tenants, and scheduling is not limited to the matching between resource supply and demand, so that a certain amount of resource over-issuance can be made in the resource pool, reducing platform costs, improving resource utilization, and improving service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] FIG1 is a schematic diagram of an application scenario of a resource scheduling method provided by an embodiment of this specification;

[0014] FIG2 is a flow chart of a resource scheduling method provided by one embodiment of this specification;

[0015] FIG3 is a schematic diagram of the system architecture of a cloud-native data warehouse provided by one embodiment of this specification;

[0016] FIG4 is a schematic diagram of the architecture of a CPU and LLC on a computing node provided by one embodiment of this specification;

[0017] FIG5 is a schematic diagram of a POD scheduling effect provided by an embodiment of this specification;

[0018] FIG6 is a flowchart of a processing process of a resource scheduling method provided by one embodiment of this specification;

[0019] FIG7 is a schematic diagram of the structure of a resource scheduling device provided by an embodiment of this specification;

[0020] FIG8 is a schematic diagram of the structure of a database provided by one embodiment of this specification;

[0021] FIG9 is a structural block diagram of a computing device provided by one embodiment of this specification. DETAILED DESCRIPTION

[0022] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0023] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a," "the," and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0024] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0025] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0026] First, the terms involved in one or more embodiments of this specification are explained.

[0027] A resource instance is a computing module that is virtualized from cloud computing resources and provided to tenants. A resource instance typically contains one or more resource groups.

[0028] A resource group is the basic execution unit of cloud computing. It typically includes resources such as storage, networking, metrics, logging, and cloud environments. For example, in a container-based computing cluster, a resource group can be understood as a group of containers, or pods. For example, in a distributed database implemented in a container-based computing cluster, resource instances created by tenants will correspondingly launch multiple pods to implement external services in the distributed cluster.

[0029] A multi-tenant resource pool is a resource pool built using a group of machines. Multiple tenants' resource instances can run in the resource pool. Generally, it is necessary to ensure that tenants do not affect each other.

[0030] Resource load metrics are metrics used to represent resource load performance. These metrics, such as CPU (Central Processing Unit) utilization and memory latency, represent the performance of a compute node's processor, memory, and input / output resources. These performance factors in the quality of service for a tenant's resource instances. For example, metrics like RT (Response Time) and QPS (Queries Per Second) measure the quality of query service.

[0031] Distributed deployment involves distributing the resource groups within resource instances of tenants with different characteristics to avoid resource conflicts and ensure service quality. For example, if some tenants' resource groups run on the same compute node, there will be no resource conflicts. However, if some tenants' resource groups are not suitable for running on the same compute node, there will be resource conflicts, affecting service quality.

[0032] Resource group scheduling refers to the allocation mechanism that finds computing nodes in the resource pool that meet the processing needs of the resource group and optimize execution efficiency.

[0033] With the development of cloud computing, cost reduction and efficiency improvement have become core demands for cloud vendors and cloud tenants. In some cloud service offerings, resource groups are deployed using a model that exclusively allocates elastic resources. This model, based on global resource pool utilization statistics, shows low peak utilization, with a significant amount of idle and wasted resources. Therefore, this deployment model, where resource groups exclusively allocate elastic resources, fails to fully leverage the performance advantages of machine hardware. Consequently, in other cloud service offerings, multi-tenant co-location is implemented at the resource pool level. Resource groups containing resource instances are evenly distributed across the compute nodes within the resource pool based on resource supply and demand. However, in real-world applications, the mixed deployment of resource instances from multiple tenants within a resource pool creates complex resource contention, making it difficult to ensure service quality. Therefore, ensuring the service quality of cloud service offerings in this co-location scenario has become a pressing issue.

[0034] In view of this, the embodiment of the present specification provides a resource scheduling method, which targets the mixed deployment of resource instances of multiple tenants in a resource pool, where there is a relatively complex resource contention situation. According to the correlation between the load of the resource group and the computing node, for multiple computing nodes, the resource load index prediction value of the resource group to be scheduled when the resource group to be scheduled is deployed on each computing node is predicted. Based on the actual value of the resource load index of the computing node and the resource load index prediction value of the resource group to be scheduled, the matching degree between the resource group to be scheduled and the computing node is calculated, and then the matching degree is used to filter out the target computing node that meets the preset scheduling conditions from the multiple computing nodes, which is equivalent to breaking up the deployment of the resource group based on the load index prediction value, reducing the probability of mutual influence between the loads of multiple tenants, and scheduling is not limited to matching between resource supply and demand, so that a certain amount of resource over-issuance can be made in the resource pool, reducing platform costs, improving resource utilization, and improving service quality. In a cloud service product that applies the resource scheduling method provided by the embodiment of the present specification, a resource pool with mixed deployment of multiple tenants can be constructed based on the resource pool level, so that low-utilization resource groups and high-utilization resource groups can be deployed together, reducing platform resource costs and improving the overall resource utilization of the platform.

[0035] Specifically, this specification provides a resource scheduling method, and this specification also involves a resource scheduling device, a database, a computing device, and a computer-readable storage medium, which are described in detail one by one in the following embodiments.

[0036] Referring to Figure 1 , Figure 1 illustrates a schematic diagram of an application scenario for a resource scheduling method according to one embodiment of this specification. As shown in Figure 1 , this application scenario includes a cloud service node, a resource scheduling service node, and a resource pool. The resource pool includes a solid resource pool and an elastic resource pool. The solid resource pool includes one or more compute nodes.

[0037] The cloud service node is configured to receive a resource instance creation request, create a resource instance based on the resource instance creation request, and send a scheduling request to the resource scheduling service node, wherein the scheduling request is used to instruct the scheduling of one or more resource groups of the resource instance. In Figure 1, a POD is used to illustrate a resource group.

[0038] A resource scheduling service node is used to receive the scheduling request, determine the POD to be scheduled according to the scheduling request, predict the resource load index predicted value of the POD to be scheduled when the POD to be scheduled is deployed on the computing node for multiple computing nodes, calculate the matching degree between the computing node and the POD to be scheduled using the actual value of the resource load index of the computing node and the predicted value of the resource load index of the POD to be scheduled when the POD to be scheduled is deployed on the computing node for the multiple computing nodes, and use the matching degree between the computing node and the POD to be scheduled to screen out a target computing node that meets the preset scheduling conditions from the multiple computing nodes for the POD to be scheduled, and schedule the POD to be scheduled to the target computing node.

[0039] For example, let's assume that Tenant 1 issues a request to create Resource Instance 1. The cloud service node receives this request, creates Resource Instance 1 based on it, and sends a scheduling request for the POD of Instance 1 to the resource scheduling service node. The resource scheduling service node receives this scheduling request, determines the POD of Instance 1 based on it, and predicts the resource load index predicted for each POD of Instance 1 when deployed on each compute node. Using the actual resource load index values ​​of the compute nodes and the predicted resource load index values ​​of each POD of Instance 1, the resource scheduling service node calculates the matching degree between each POD of Instance 1 and each compute node. Using this matching degree, the resource scheduling service node selects a target compute node from multiple compute nodes that meets the preset scheduling conditions for each POD of Instance 1. For example, if Compute Node 1 is selected for PODs 1 and 2 of Instance 1, and Compute Node 2 is selected for PODs 3 and 4 of Instance 1, PODs 1 and 2 of Instance 1 are scheduled to Compute Node 1, and PODs 3 and 4 of Instance 1 are scheduled to Compute Node 2.

[0040] The computing nodes can be physical servers or cloud servers in the cloud that provide various services. For example, they can be servers that provide communication services, data processing services, computing services, storage services, database services, and other services to multiple clients. It should be noted that the computing nodes in the cloud can be implemented as a distributed server cluster consisting of multiple servers or as a single server. The server can also be a server in a distributed system or a server integrated with a blockchain. The server can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0041] It is understandable that the above application scenarios are only used to illustrate the methods provided in the embodiments of this specification and do not constitute a limitation on the methods provided in the embodiments of this specification. The methods provided in the embodiments of this specification can perform resource scheduling for cloud-native products of computing clusters with various master-slave architectures. For example, the methods provided in the embodiments of this specification can be applied to a multi-tenant elastic pooling scheduling platform to schedule elastic resources for multi-tenant cloud-native data warehouses; for another example, the methods provided in the embodiments of this specification can be applied to resource scheduling of multi-cloud databases to meet the needs of data warehouse analysis and computing.

[0042] Referring to FIG. 2 , FIG. 2 shows a flow chart of a resource scheduling method provided according to an embodiment of this specification, which specifically includes the following steps.

[0043] Step 202: For multiple computing nodes, predict resource load indicator predicted values ​​of the resource groups to be scheduled when the resource groups to be scheduled are deployed on the computing nodes.

[0044] The resource load indicator is a metric used to represent the resource load performance of a computing node or resource group. For example, the resource load indicator may include load indicators for any one or more of processor resources, memory resources, input / output resources, and network bandwidth resources. For example, the resource load indicator may be CPU utilization, memory access latency, etc. As decentralized scheduling becomes more refined, any one or more resource load indicators may be increased or decreased as needed.

[0045] The resource scheduling methods provided in the embodiments of this specification are not limited to the timing of scheduling resource groups. For example, upon receiving a scheduling request, a resource group to be scheduled can be determined based on the scheduling request and then scheduled. For another example, the resource scheduling service can monitor the current resource load indicators of each computing node in real time. When one or more resource load indicators of a computing node are found to meet preset rescheduling conditions, the resource group to be scheduled can be determined from the computing node and rescheduled.

[0046] After determining the resource group to be scheduled, the resource load index predicted value of the resource group to be scheduled can be predicted for multiple computing nodes when the resource group to be scheduled is deployed on the computing node. The specific method of predicting the resource load index predicted value is not limited. For example, the resource load index predicted value of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node can include:

[0047] Using a load indicator collector deployed on each computing node, collecting resource load indicators of the computing node and the resource group deployed on the computing node, and generating resource load indicator historical data;

[0048] The resource load indicator historical data is used to predict, for multiple computing nodes, predicted resource load indicator values ​​of the resource groups to be scheduled when the resource groups to be scheduled are deployed on the computing nodes.

[0049] In this embodiment, the specific implementation method for performing predictions based on historical resource load indicator data is not limited. For example, historical data can be sorted to summarize the correlation rules between resource load indicators between computing nodes and resource groups, and the resource load indicator prediction value of the resource group to be scheduled can be predicted based on the correlation rules. In another example, a machine learning model can be trained using historical data as samples to perform pattern training on the correlation between resource load indicators between computing nodes and resource groups, and the trained machine learning model can then be used to predict the resource load indicator prediction value of the resource group to be scheduled.

[0050] Among them, the resource load indicator historical data may include: the resource instances related to the resource group to be scheduled and the resource load indicator historical data of the related computing nodes. The resource instances related to the resource group to be scheduled include the resource instance where the resource group to be scheduled is located and other resource instances similar to the resource instance where the resource group to be scheduled is located. Specifically, according to implementation needs, the resource load indicator historical data of a preset time window can be used to predict the future resource load indicators and service quality of the resource group to be scheduled. In addition, the future resource load indicators and service quality of the computing nodes can also be predicted. Among them, in the case where there is no resource load indicator historical data for a newly created resource instance, the average resource load indicator historical data of similar instances can be matched according to the ratio of the number of storage resource groups and the number of computing resource groups of the resource instance as the resource load indicator data historical data used to participate in the prediction.

[0051] The historical resource load indicator data can be obtained through a load indicator collector deployed on each compute node in the resource pool. The load indicator collector collects resource load indicators for the compute node and the resource groups deployed on the compute node through operating system calls. Depending on the needs of the prediction implementation, the resource load indicators collected by the load indicator collector can also be aggregated. This aggregation can be performed based on one or more dimensions, such as compute node, resource instance, resource group, resource type, and data generation time, to obtain historical resource load indicator data for one or more dimensions.

[0052] Step 204: For the multiple computing nodes, use the actual value of the resource load indicator of the computing node and the predicted value of the resource load indicator of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node to calculate the matching degree between the computing node and the resource group to be scheduled.

[0053] It can be understood that the more the predicted value of the resource load index of the resource group to be scheduled and the resource load index of the computing node meet the implementation needs of the application scenario, the higher the matching degree between the two. Therefore, the specific evaluation method of the matching degree between the computing node and the resource group to be scheduled can be set according to the implementation needs of the application scenario, and this manual does not limit this.

[0054] For example, in one or more embodiments of this specification, a predicted resource load index value that the computing node will reach in the future is calculated based on the actual value of the computing node's current resource load indicator and the predicted resource load index value of the resource group to be scheduled, and the degree of matching is determined based on the predicted resource load index value that the computing node will reach in the future. In this embodiment, the lower the predicted resource load index value that the computing node will reach in the future, the more balanced the resource load of the entire system will be if the resource group to be scheduled is deployed on the computing node, and the higher the degree of matching between the two.

[0055] For another example, in one or more embodiments of this specification, the actual value of the current resource load indicator of the computing node is compared with the predicted value of the resource load indicator of the resource group to be scheduled to determine whether the resource indicator predicted value requirements are met, and the matching degree is determined based on the frequency with which the computing node participates in scheduling. In this embodiment, the higher the resource load indicator of the computing node meets the requirements of the resource group to be scheduled and the higher the frequency of its participation in scheduling, the more frequently the resources of the computing node are idle. Deploying the resource group to be scheduled to the computing node will enable the idle resources of the computing node to be used as soon as possible, thereby improving the resource utilization of the entire system. In this case, the higher the matching degree between the two.

[0056] Step 206: Using the matching degree between the computing node and the resource group to be scheduled, select a target computing node that meets a preset scheduling condition from the plurality of computing nodes for the resource group to be scheduled.

[0057] The preset scheduling conditions are information that stipulates the conditions that the matching degree between the target computing node and the resource group to be scheduled should meet, and are set according to the needs of the implementation scenario. For example, the preset scheduling conditions may include: the matching degree between the target computing node and the resource group to be scheduled should be the highest matching degree among multiple computing nodes. For another example, the preset scheduling conditions may include: the matching degree between the target computing node and the resource group to be scheduled should be greater than a preset matching degree threshold. For another example, the preset scheduling conditions may include: sorting from high to low by matching degree, and the computing nodes that are ranked first by a preset number are the target computing nodes.

[0058] Step 208: Schedule the to-be-scheduled resource group to the target computing node.

[0059] For example, taking a POD to be scheduled as an example, scheduling the resource group to be scheduled to the target computing node can be understood as launching the POD to be scheduled on the target computing node. The specific implementation method of launching the POD on the computing node can refer to the implementation method of a general container cluster and is not limited in this specification.

[0060] It can be seen that this method targets the mixed deployment of resource instances of many tenants in the resource pool, and there is a relatively complex resource contention situation. According to the correlation between the load of the resource group and the computing node, the resource load index predicted value of the resource group to be scheduled is predicted when the resource group to be scheduled is deployed on each computing node. The matching degree between the resource load index actual value of the computing node and the resource load index predicted value of the resource group to be scheduled and the computing node is calculated, and then the matching degree is used to screen out the target computing node that meets the preset scheduling conditions from multiple computing nodes. This is equivalent to breaking up the deployment of the resource group based on the load index prediction value, reducing the probability of mutual influence between multiple tenants, and scheduling is not limited to the matching between resource supply and demand, so that a certain amount of resource over-issuance can be made in the resource pool, reducing platform costs, improving resource utilization, and improving service quality.

[0061] In one or more embodiments of this specification, after a cluster runs some resource instances, as the load of user resource instances, such as queries and analyses, changes, the problem of uneven load levels on computing nodes may occur. The method can also reschedule the deployed resource groups to maintain a balanced load level on computing nodes in the cluster and better ensure service quality. Specifically, the method also includes:

[0062] Monitoring actual values ​​of resource load indicators of the plurality of computing nodes;

[0063] Filtering out, from the plurality of computing nodes, rescheduling nodes whose actual values ​​of resource load indicators meet preset rescheduling conditions;

[0064] Filtering the resource group that needs to be rescheduled from the resource groups deployed on the rescheduling node;

[0065] The resource group that needs to be rescheduled is taken as the resource group to be scheduled, and the step of obtaining the resource load indicators of multiple computing nodes and the resource load indicators of the resource groups deployed on the multiple computing nodes is re-entered to reschedule the resource group to be scheduled.

[0066] For example, the load indicator collector deployed on each of the aforementioned computing nodes can be used to monitor the actual values ​​of the current resource load indicators of multiple computing nodes through operating system calls, select computing nodes with high load levels, and then filter out the resource groups that cause the high load levels for rescheduling.

[0067] In one or more embodiments of this specification, in view of the different resource requirements of different types of resource groups, weight coefficients corresponding to various resources are set for the resource groups. The weight coefficients corresponding to various resources of different types of resource groups can be different. In this way, the weight coefficients corresponding to various resources of the resource group to be scheduled can be used to calculate the overall matching degree, so that the resource group can be scheduled to a computing node that is more suitable for its operation. Specifically, the actual value of the resource load indicator of the computing node and the predicted value of the resource load indicator of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node are used to calculate the matching degree between the computing node and the resource group to be scheduled, including:

[0068] The comprehensive matching degree between the computing node and the resource group to be scheduled on the multiple resources is calculated by using the actual value of the resource load indicator of the computing node, the predicted values ​​of the resource load indicators of the multiple resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, and the weight coefficients corresponding to the multiple resources of the resource group to be scheduled.

[0069] The types of the multiple resources are not limited and may include, for example, at least two of processor resources, memory resources, input / output resources, and network bandwidth resources. Based on the resource types provided in this embodiment, load spreading can be achieved based on the perception of resource load indicators of at least two of the processor resources, memory resources, input / output resources, and network bandwidth resources.

[0070] Below, the specific implementation of calculating the comprehensive matching degree by the method provided in the embodiment of this specification is exemplified by taking the multiple resources, including processor resources and memory access resources, as an example.

[0071] Specifically, calculating the comprehensive matching degree between the computing node and the resource group to be scheduled on the multiple resources by using the actual value of the resource load indicator of the computing node, the predicted values ​​of the resource load indicators of the multiple resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, and the weight coefficients corresponding to the multiple resources of the resource group to be scheduled, may include:

[0072] Calculating a matching degree between the computing node and the resource group to be scheduled in terms of processor resources by using the actual value of the current resource load indicator of the processor resources of the computing node, the predicted value of the resource load indicator of the processor resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, the logical processing core allocation value of the resource group to be scheduled, and the total number of logical processing cores of the computing node;

[0073] Calculate the matching degree of memory access resources between the computing node and the resource group to be scheduled by using the actual value of the current resource load indicator of the memory access resources of the computing node and the predicted value of the resource load indicator of the memory access resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node;

[0074] The comprehensive matching degree between the computing node and the resource group to be scheduled is calculated by using the matching degree between the computing node and the resource group to be scheduled in terms of processor resources, the matching degree in terms of memory resources, the weight coefficient corresponding to the processor resources of the resource group to be scheduled, and the weight coefficient corresponding to the memory resources of the resource group to be scheduled.

[0075] For example, according to the above embodiment, the calculation of the matching degree between the computing node and the to-be-scheduled resource group in terms of processor resources may be performed using the following formula 1.

[0076] Formula 1: Node_Request_CPU_Score=Node_CPU_Usage+(Pod_CPU_Usage*Pod_CPU_Cores) / Node_CPU_Cores

[0077] Among them, Node_Request_CPU_Score represents the predicted value of the resource load index of the processor resources of the computing node when the resource group to be scheduled is deployed on the computing node, Node_CPU_Usage represents the actual value of the resource load index, Pod_CPU_Usage represents the predicted value of the resource load index, Pod_CPU_Cores represents the logical processing core allocation value of the resource group to be scheduled, and Node_CPU_Cores represents the total number of logical processing cores of the computing node.

[0078] The above formula 1 calculates the ratio of the logical processing core allocation value of the resource group to be scheduled to the total number of logical processing cores of the computing node, and converts the resource load index prediction value of the processor resources of the resource group to be scheduled from a measurement value in the logical processing core dimension to a measurement value in the physical processor dimension, thereby adding the current actual value of the resource load index of the processor resources of the computing node and the resource load index prediction value of the processor resources of the resource group to be scheduled in the physical processor dimension to obtain the resource load index prediction value of the processor resources of the computing node.

[0079] In the above embodiment, considering that the lower the predicted value of the resource load index of the processor resources of the computing node, the more balanced the overall load level of the platform will be after the resource group to be scheduled is deployed on the computing node, the predicted value of the resource load index of the processor resources of the computing node, Node_Request_CPU_Score, is used as the matching degree of the processor resources between the computing node and the resource group to be scheduled. In this embodiment, the smaller the value of the predicted value of the resource load index of the processor resources of the computing node, Node_Request_CPU_Score, the higher the matching degree.

[0080] For another example, according to the above embodiment, the calculation of the matching degree between the computing node and the to-be-scheduled resource group in terms of memory access resources may be performed using the following formula 2.

[0081] Formula 2: IMC_Latency_ns_Score=Node_IMC_Latency_ns-Pod_IMC_Latency_ns

[0082] Among them, IMC_Latency_ns_Score represents the difference between the actual value of the current resource load index of the memory access resources of the computing node and the predicted value of the resource load index of the memory access resources of the resource group to be scheduled, Node_IMC_Latency_ns represents the actual value of the current resource load index of the memory access resources of the computing node, and Pod_IMC_Latency_ns represents the predicted value of the resource load index of the memory access resources of the resource group to be scheduled. Among them, Pod_IMC_Latency_ns can be the memory access delay that the resource group to be scheduled expects to have no impact on the service quality, which can be specifically determined using historical data on memory access delay. For example, based on the historical data on memory access delay, an expected memory access delay threshold is found. As long as the memory access delay of the computing node is below the memory access delay threshold, the service quality requirements can be met.

[0083] In the above embodiment, considering that the smaller the gap between the actual value of the current resource load index of the memory access resource of the computing node and the predicted value of the resource load index of the memory access resource of the resource group to be scheduled, the more the memory access needs of the resource group to be scheduled can be met, the gap (IMC_Latency_ns_Score) between the actual value of the current resource load index of the memory access resource of the computing node and the predicted value of the resource load index of the memory access resource of the resource group to be scheduled is used as the matching degree between the computing node and the resource group to be scheduled in terms of memory access resources. In this embodiment, the smaller the value of IMC_Latency_ns_Score, the higher the matching degree.

[0084] According to the above formula 1 and formula 2, the comprehensive matching degree between the computing node and the to-be-scheduled resource group can be calculated, illustratively, using the following formula 3.

[0085] Formula 3: Qos_Match_Score=x*Node_Request_CPU_Usage+y*IMC_Latency_ns_Score

[0086] Where Qos_Match_Score represents the overall matching degree, x represents the weight coefficient corresponding to the processor resources of the resource group to be scheduled, and y represents the weight coefficient corresponding to the memory resources of the resource group to be scheduled. It can be understood that Qos_Match_Score is the numerical value corresponding to the overall matching degree. According to the meaning of the above formula, the smaller the numerical value, the higher the overall matching degree. The higher the overall matching degree between a resource group and a computing node, the smaller the impact of scheduling on service quality.

[0087] The specific values ​​of the weight coefficient corresponding to the processor resources of the resource group to be scheduled and the weight coefficient corresponding to the memory resources of the resource group to be scheduled can be set according to the different requirements of the resource group to be scheduled for processor resources and memory resources in the implementation scenario.

[0088] For example, when the resource group to be scheduled is a computing resource group, the weight coefficient corresponding to the processor resource of the resource group to be scheduled is greater than the weight coefficient corresponding to the memory access resource.

[0089] For another example, when the resource group to be scheduled is a storage resource group, the weight coefficient corresponding to the processor resource of the resource group to be scheduled is smaller than the weight coefficient corresponding to the memory access resource.

[0090] It is understandable that resource instances generally include two types of resource groups: one is a storage resource group of storage type, and the other is a computing resource group of computing type. The computing resource group, commonly referred to as an "Executor Pod", mainly reads the data stored in the storage resource group and puts it into memory for calculation; the storage resource group, commonly referred to as a "Worker Pod", mainly performs a large number of memory access operations. The method provided in the embodiments of this specification, based on periodic load performance evaluation, found that the service quality of the computing resource group is highly correlated with the resource load index of the processor resource, and the service quality of the storage resource group is highly correlated with the resource load index of the memory resource. Therefore, in the above embodiment, when the resource group to be scheduled is a computing resource group, the weight coefficient corresponding to the processor resource of the resource group to be scheduled is greater than the weight coefficient corresponding to the memory resource, so that the matching degree evaluation focuses more on the load of the processor resource, which helps to improve the service quality of the computing resource group; when the resource group to be scheduled is a storage resource group, the weight coefficient corresponding to the processor resource of the resource group to be scheduled is less than the weight coefficient corresponding to the memory resource, so that the matching degree evaluation focuses more on the load of the memory resource, which helps to improve the service quality of the storage resource group. For example, the weight coefficients for the compute resource group could be (x, y) = (0.7, 0.3), giving the CPU a higher weight; the weight coefficients for the storage resource group could be (x, y) = (0.3, 0.7), giving memory access a higher weight. Of course, in actual applications, the weight coefficients can be more finely tuned based on load requirements, so we won't go into detail here.

[0091] In order to more accurately express the resource load index of the processor resources, in one or more embodiments of this specification, the current resource load index of the processor resources of the computing node includes: the current processor resource utilization of the computing node; the resource load index prediction value of the processor resources of the resource group to be scheduled includes: the processor resource utilization prediction value of the resource group to be scheduled.

[0092] In order to more accurately express the resource load index of the memory access resources, in one or more embodiments of the present specification, the actual value of the current resource load index of the memory access resources of the computing node includes: the current memory access delay of the computing node; the predicted value of the resource load index of the memory access resources of the resource group to be scheduled includes: the predicted value of the memory access delay of the resource group to be scheduled.

[0093] It is understandable that the resource load index of the processor resources and the resource load index of the memory access resources can also be expressed using other indicators besides resource utilization and memory access latency, and this specification does not limit this.

[0094] In order to make the overall resource load level of the platform more balanced, in one or more embodiments of this specification, using the matching degree between the computing node and the resource group to be scheduled to select a target computing node that meets the preset scheduling conditions from the multiple computing nodes for the resource group to be scheduled includes:

[0095] By utilizing the comprehensive matching degree between the computing node and the resource group to be scheduled, a target computing node with the lowest comprehensive processor resource utilization and memory access latency is screened out from the multiple computing nodes for the resource group to be scheduled.

[0096] For example, based on the above formulas 1, 2, and 3, a computing node with the smallest Qos_Match_Score value may be screened out from the multiple computing nodes for the resource group to be scheduled as the target computing node.

[0097] The following, in conjunction with Figure 3, takes the application of the resource scheduling method provided in this specification in a cloud-based database as an example to further illustrate the resource scheduling method. For example, the cloud-based database can be a cloud-native data warehouse. A cloud-native data warehouse is a data warehouse built on cloud infrastructure with unlimited elasticity of computing resources and storage resources. Specifically, Figure 3 shows a schematic diagram of the system architecture of a cloud-native data warehouse provided by an embodiment of this specification. As shown in Figure 3, the cloud-native data warehouse builds a multi-tenant elastic pool at a regional level based on a container cluster. Each solid-state resource pool contains computing nodes corresponding to multiple physical machines, and an on-demand exclusive elastic resource pool is also set up. The multi-tenant mixed resource pool in the cloud-native data warehouse can use a solid-state resource pool. According to the method provided in the embodiment of this specification, the load-sharing scheduling of the multi-tenant mixed resource pool of the cloud-native data warehouse is realized, which reduces the probability of mutual influence of loads between multiple tenants, and can make a certain amount of resource over-issuance in the resource pool, reduce platform costs, improve resource utilization, and improve service quality.

[0098] As shown in the system architecture of the cloud native data warehouse in Figure 3, if a user wants to use the cloud native data warehouse, he must first create a resource instance through the database service node of the cloud native data warehouse. In response to receiving the resource instance creation request, the database service node of the cloud native data warehouse sends the scheduling request of the Pod of the resource instance to the resource scheduling service node, and the resource scheduling service node schedules the Pod to a specific computing node for execution according to the resource scheduling method provided in the embodiment of this specification. Afterwards, the user's query request will be sent to the corresponding Pod for execution. Since the complexity, concurrency and other aspects of the query requests issued by different users are different, the load of each user's Pod will also be different. Since the entire resource pool supports instance resources created by all users, the resource pool of the cloud native data warehouse is divided into a solid resource pool and an elastic resource pool. The solid resource pool consists of multiple servers. A server can be understood as a computing node. Multiple users' Pods can run on a computing node at the same time. Therefore, users will have an impact on each other.

[0099] Based on this, according to the method provided in the embodiment of this specification, for the mixed deployment of resource instances of many users in the resource pool, there is a relatively complex resource competition situation. According to the correlation between the load of POD and the computing node, the resource load index prediction value of the POD to be scheduled is predicted when it is deployed on each computing node. The matching degree between the resource load index prediction value of the POD to be scheduled and the computing node is calculated based on the resource load index prediction value of the POD to be scheduled, and then the matching degree is used to filter out the target computing node that meets the preset scheduling conditions from multiple computing nodes. This is equivalent to breaking up the deployment of POD based on the load index prediction value, reducing the probability of mutual influence between multiple users' loads, and scheduling is not limited to matching between resource supply and demand, so that a certain amount of resource over-issuance can be made in the resource pool, reducing platform costs, improving resource utilization, and improving service quality. In Figure 3, the scheduling effect is schematically demonstrated by taking the pre-dispersion and in-process balanced dispersion of PODs deployed in a solid-state resource pool as an example.

[0100] FIG4 shows a schematic diagram of the architecture of the CPU and LLC on a computing node.

[0101] As shown in Figure 4, the CPU and LLC architecture diagram shows that within the resource pool, each compute node can have the performance of a physical machine, including resources such as the CPU, LLC (last level cache), and memory bandwidth. In multi-user scenarios, these resources are typically shared, meaning that multiple pods share CPU and LLC resources. When resource utilization exceeds a certain threshold, significant contention overhead is incurred, impacting the load on each pod and the service quality of the cloud-native data warehouse.

[0102] Specifically, for example, CPU sharing can be managed through the operating system's scheduling module, which manages multiple logical cores of the CPU. The scheduling module allocates a certain number of logical cores to each pod based on its resource configuration. For example, the scheduling module may allocate 15 logical cores to one pod and 8 logical cores to another pod based on their resource configuration. The scheduling module allocates time slices to different pods based on their logical cores. When the CPU load of each pod is high, each pod will queue in the scheduling module. Excessive queuing time can affect the service quality of CPU-intensive pods. For example, CPU utilization can reflect CPU busyness. When the CPU utilization of the entire machine is below a certain value, service quality is generally not affected. However, when the CPU utilization of the entire machine exceeds this value, service quality is affected. In this case, pods can be rescheduled to balance the processor resource load across the platform's computing nodes and ensure service quality.

[0103] For example, memory resources are generally divided into three tiers, with LLC being the last tier shared by multiple physical cores. LLC is a shared resource that primarily impacts the service quality of pods (Pods). In cloud-native data warehouse scenarios, LLC contention can be measured using memory latency, a resource load metric. When the overall machine's memory latency is below a certain threshold, service quality is generally not affected. However, when the latency exceeds this threshold, service quality is affected. In this case, Pods can be rescheduled to balance the memory resource load across the platform's compute nodes, ensuring service quality.

[0104] Based on this, according to the method provided in the embodiments of this specification, in the cloud native data warehouse application scenario, the actual value of the resource load indicator of each computing node of the platform can be monitored. When the load water level of the platform is found to be unbalanced based on the actual value of the resource load indicator of each computing node, the rescheduling node whose actual value of the resource load indicator reaches the preset rescheduling condition can be screened out from multiple computing nodes. For example, the preset rescheduling condition may include a preset threshold. When the actual value of the resource load indicator of a computing node reaches the preset threshold, the computing node is the rescheduling node. For another example, the preset rescheduling condition may include selecting a preset number of computing nodes with a high load water level as rescheduling nodes. PODs that cause high resource load water levels are screened out from the rescheduling nodes for rescheduling. The method of screening PODs that need to be rescheduled can be similar to the method of selecting rescheduling nodes, such as screening according to a certain load water level threshold, or screening according to the load water level ranking. Other methods can also be used, and this specification does not limit this. For example, with respect to CPU utilization, when the CPU utilization of a computing node is less than 25%, there is no impact on the service quality. When the CPU utilization of a computing node is greater than or equal to 25%, the actual value of the resource load indicator and service type of the POD of the computing node can be perceived to select PODs for rescheduling. For another example, with respect to memory access latency, when the memory access latency of a computing node is less than 300ns, there is no impact on the service quality. When the memory access latency of a computing node is greater than or equal to 300ns, the actual value of the resource load indicator and service type of the POD of the computing node can be perceived to select PODs for rescheduling.

[0105] The following is an example of how to schedule PODs in a cloud-native data warehouse, with reference to the schematic diagram of POD scheduling effects shown in FIG5 .

[0106] In cloud-native data warehouses, services are provided to each user through resource instances. Each resource instance can contain two types of pods: storage pods, called worker pods, and compute pods, called executor pods. The service quality of an executor pod is highly correlated with processor resource load metrics, such as CPU utilization; the service quality of a worker pod is highly correlated with memory resource load metrics, such as memory access latency. Therefore, worker pods are more suitable for scheduling to compute nodes with lower memory access latency, while executor pods are more suitable for scheduling to compute nodes with lower CPU utilization.

[0107] Based on this, according to the method provided in the embodiment of this specification, in the cloud native data warehouse scenario, the comprehensive matching degree between the computing node and the POD to be scheduled can be calculated based on Formula 1, Formula 2 and Formula 3 provided in the above embodiment. For the Executor Pod, the weight coefficient corresponding to its CPU utilization is greater than the weight coefficient corresponding to the memory access latency. For the Worker Pod, the weight coefficient corresponding to its CPU utilization is less than the weight coefficient corresponding to the memory access latency. In this way, the comprehensive matching degree between the computing node and the POD to be scheduled can be used to screen out the target computing node with the lowest CPU utilization and memory access latency from multiple computing nodes for the POD to be scheduled. According to the setting of the weight coefficient, the Worker Pod can be scheduled to the computing node with lower memory access latency, and the Executor Pod can be scheduled to the computing node with lower CPU utilization. For example, the resource instance 2 shown in Figure 5 includes five Executor Pods and six Worker Pods. According to the method provided in the embodiments of this specification, the Worker Pod is scheduled to a computing node with lower memory access latency, and the Executor Pod is scheduled to a computing node with lower CPU utilization, so that the three Executor Pods of resource instance 2 are scheduled to computing node 1, and the other two Executor Pods are scheduled to computing node 2; one Worker Pod of resource instance 2 is scheduled to computing node 1, three Worker Pods are scheduled to computing node 2, and the other two Worker Pods are scheduled to computing node 3.

[0108] The following further illustrates the processing process of applying the resource scheduling method in a cloud-native data warehouse with reference to FIG6 The resource scheduling architecture implemented by the resource scheduling method provided in this embodiment may include: a load indicator collector, a profiling module, a POD scheduling module, and a rescheduling module.

[0109] The load metric collector, deployed on compute nodes in the resource pool, collects resource load metrics for the compute nodes and the pods running on them through operating system calls. Resource load metrics can include CPU utilization and memory access latency. In actual applications, as decentralized scheduling becomes more refined, more resource load metrics can be added as needed.

[0110] The profiling module aggregates the resource load metrics collected by the load metric collector and uses them to predict the resource load metrics of the pods to be scheduled. This prediction is generally based on historical data from the most recent time window.

[0111] The POD scheduling module is used to find the computing node whose comprehensive matching degree with the POD to be scheduled meets the preset scheduling conditions based on the predicted value of the resource load index predicted by the portrait module. For example, it can be the computing node with the highest comprehensive matching degree, thereby reducing the probability that the POD is affected by other tenants on the corresponding computing node. The POD scheduling module can further include three submodules: a scoring submodule, a sorting submodule and a scheduling submodule. Among them, the scoring submodule is used to calculate the comprehensive matching degree between the POD to be scheduled and each computing node. The sorting submodule is used to sort the comprehensive matching degrees calculated by the scoring submodule. The scheduling submodule is used to find the computing node with the highest comprehensive matching degree from the comprehensive matching degrees sorted by the sorting submodule, and schedule the POD to be scheduled to the computing node. Among them, according to the above formulas 1, 2 and 3, the computing node with the highest comprehensive matching degree is also the computing node with the smallest value of Qos_Match_Score.

[0112] The rescheduling module is used to monitor the actual values ​​of resource load indicators of multiple computing nodes. When the load levels of various load indicators of multiple computing nodes are unbalanced, the node with the high actual value of resource load indicator is selected. For example, the node with high CPU utilization and high memory access latency is selected. Then, the POD that causes the high value is selected as the POD to be scheduled, and a scheduling request is sent to the POD scheduling module to reschedule the POD, thereby periodically ensuring the balance of the cluster load level.

[0113] Based on the above scheduling architecture, FIG6 shows a processing flow chart of a resource scheduling method provided in an embodiment of this specification, which specifically includes the following steps.

[0114] Step 602: The load index collector collects resource load indexes of each computing node and POD, and sends the collected resource load indexes to the profiling module.

[0115] Step 604: The POD scheduling module receives the POD scheduling request. When the user issues a resource instance creation request, the POD scheduling request is generated by the instance creation management module in the cloud native data warehouse service node.

[0116] Step 606: The POD scheduling module calls the profiling module to predict the resource load index prediction value of the POD to be scheduled, and calculates the comprehensive matching degree between the POD to be scheduled and each computing node, so as to use the comprehensive matching degree to select a computing node whose comprehensive matching degree meets the preset scheduling conditions from multiple computing nodes.

[0117] Step 608: The POD scheduling module schedules the POD to be scheduled to the selected computing node.

[0118] While the POD is executing on the computing node, the load indicator collector can continue to collect resource load indicators of the POD and computing node, and periodically send them to the profiling module.

[0119] Step 610: During the POD execution process of the entire resource pool, the rescheduling module periodically obtains the actual value of the resource load index of each computing node and each POD from the profiling module, and calculates the balance of the resource load index of each computing node.

[0120] Step 612: When the balance degree of the resource load indicators of each computing node reaches the preset rescheduling condition, the rescheduling module selects the computing node whose actual value of the resource load indicator reaches the preset rescheduling condition, selects the POD that needs to be rescheduled from the computing node, and sends a rescheduling request to the POD scheduling module, which recalculates the comprehensive matching degree between the POD and each computing node accordingly so as to reschedule the POD.

[0121] As can be seen from the above process, this embodiment implements a fully self-feedback scheduling link based on data profiling, which can distribute the CPU and memory access load. Furthermore, depending on the implementation environment, load awareness of resources such as input / output and network bandwidth can be added as needed, and scheduling can be performed based on the load indicators of these resources to distribute the load.

[0122] Corresponding to the above method embodiment, this specification also provides an embodiment of a resource scheduling device. FIG7 shows a schematic diagram of the structure of a resource scheduling device provided by one embodiment of this specification. As shown in FIG7, the device includes:

[0123] The load prediction module 702 is configured to predict, for multiple computing nodes, predicted resource load index values ​​of the resource groups to be scheduled when the resource groups to be scheduled are deployed on the computing nodes.

[0124] a matching calculation module 704 configured to calculate, for the plurality of computing nodes, a matching degree between the computing nodes and the resource groups to be scheduled using actual resource load index values ​​of the computing nodes and predicted resource load index values ​​of the resource groups to be scheduled when the resource groups to be scheduled are deployed on the computing nodes;

[0125] The node screening module 706 is configured to use the matching degree between the computing node and the resource group to be scheduled to screen a target computing node that meets a preset scheduling condition from the plurality of computing nodes for the resource group to be scheduled;

[0126] The scheduling execution module 708 is configured to schedule the resource group to be scheduled to the target computing node.

[0127] In one or more embodiments of this specification, the device further includes:

[0128] A monitoring module, configured to monitor actual values ​​of resource load indicators of the plurality of computing nodes;

[0129] a rescheduling node screening module configured to screen out, from the plurality of computing nodes, rescheduling nodes whose actual values ​​of resource load indicators meet preset rescheduling conditions;

[0130] a rescheduling resource group screening module, configured to screen out resource groups that need to be rescheduled from resource groups deployed on the rescheduling node;

[0131] The rescheduling trigger module is configured to take the resource group that needs to be rescheduled as the resource group to be scheduled, and re-enter the step of predicting the resource load indicator prediction value of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing nodes for multiple computing nodes, so as to reschedule the resource group to be scheduled.

[0132] In one or more embodiments of the present specification, the matching calculation module is configured to use the actual value of the resource load index of the computing node, the predicted values ​​of the resource load indicators of multiple resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, and the weight coefficients corresponding to the multiple resources of the resource group to be scheduled, to calculate the comprehensive matching degree between the computing node and the resource group to be scheduled on the multiple resources.

[0133] In one or more embodiments of this specification, the multiple resources include: processor resources and memory resources; the matching calculation module includes:

[0134] a processor resource matching submodule configured to calculate a matching degree of processor resources between the computing node and the resource group to be scheduled using a current actual value of a resource load indicator of the processor resources of the computing node, a predicted value of a resource load indicator of the processor resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, an allocation value of logical processing cores of the resource group to be scheduled, and a total number of logical processing cores of the computing node;

[0135] a memory access resource matching submodule configured to calculate a matching degree of memory access resources between the computing node and the resource group to be scheduled by using a current actual value of a resource load indicator of the memory access resources of the computing node and a predicted value of a resource load indicator of the memory access resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node;

[0136] The comprehensive matching degree calculation submodule is configured to calculate the comprehensive matching degree between the computing node and the resource group to be scheduled by using the matching degree between the computing node and the resource group to be scheduled in terms of processor resources, the matching degree in terms of memory resources, the weight coefficient corresponding to the processor resources of the resource group to be scheduled, and the weight coefficient corresponding to the memory resources of the resource group to be scheduled.

[0137] In one or more embodiments of the present specification, when the resource group to be scheduled is a computing resource group, the weight coefficient corresponding to the processor resource of the resource group to be scheduled is greater than the weight coefficient corresponding to the memory access resource.

[0138] In one or more embodiments of the present specification, when the resource group to be scheduled is a storage resource group, the weight coefficient corresponding to the processor resource of the resource group to be scheduled is smaller than the weight coefficient corresponding to the memory access resource.

[0139] In one or more embodiments of the present specification, the node screening module is configured to utilize the comprehensive matching degree between the computing node and the resource group to be scheduled to screen out the target computing node with the lowest comprehensive processor resource utilization and memory access latency from the multiple computing nodes for the resource group to be scheduled.

[0140] In one or more embodiments of this specification, the load prediction module includes:

[0141] a data collection submodule configured to collect resource load indicators of the computing nodes and resource groups deployed on the computing nodes using load indicator collectors deployed on the computing nodes, and generate resource load indicator historical data;

[0142] The load prediction submodule is configured to use the resource load index historical data to predict, for multiple computing nodes, the resource load index predicted value of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node.

[0143] The above is a schematic scheme of a resource scheduling device of this embodiment. It should be noted that the technical scheme of the resource scheduling device and the technical scheme of the resource scheduling method described above are of the same concept. For details not described in detail in the technical scheme of the resource scheduling device, please refer to the description of the technical scheme of the resource scheduling method described above.

[0144] Corresponding to the above method embodiment, this specification also provides a database embodiment. Figure 8 shows a schematic diagram of the structure of a database provided by one embodiment of this specification. As shown in Figure 8, the database includes:

[0145] A database service node 802, a resource scheduling service node 804 and a resource pool 806, wherein the resource pool includes multiple computing nodes.

[0146] The database service node 802 is configured to receive a resource instance creation request, create a resource instance according to the resource instance creation request, and send a scheduling request to the resource scheduling service node, where the scheduling request is used to instruct scheduling of one or more resource groups of the resource instance.

[0147] The resource scheduling service node 804 is configured to receive the scheduling request and schedule the resource group indicated by the scheduling request according to the resource scheduling method described in any embodiment of this specification.

[0148] The above is a schematic solution of a database of this embodiment. It should be noted that the technical solution of the database and the technical solution of the resource scheduling method described above are of the same concept. For details not described in detail in the technical solution of the database, please refer to the description of the technical solution of the resource scheduling method described above.

[0149] Figure 9 shows a block diagram of a computing device 900 according to one embodiment of this specification. Components of the computing device 900 include, but are not limited to, a memory 910 and a processor 920. The processor 920 is connected to the memory 910 via a bus 930, and a database 950 is used to store data.

[0150] The computing device 900 also includes an access device 940 that enables the computing device 900 to communicate via one or more networks 960. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 940 may include one or more of any type of network interface (e.g., a network interface card (NIC)) whether wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, or a near field communication (NFC) interface.

[0151] In one embodiment of the present specification, the aforementioned components of computing device 900 and other components not shown in FIG9 may also be connected to each other, for example, via a bus. It should be understood that the computing device structure block diagram shown in FIG9 is for illustrative purposes only and does not limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0152] The computing device 900 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 900 may also be a mobile or stationary server.

[0153] The processor 920 is configured to execute the following computer-executable instructions, which implement the steps of the above-mentioned resource scheduling method when executed by the processor.

[0154] The above is a schematic solution of a computing device of this embodiment. It should be noted that the technical solution of the computing device and the technical solution of the resource scheduling method described above are of the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the resource scheduling method described above.

[0155] An embodiment of the present specification further provides a computer-readable storage medium storing computer-executable instructions, which implement the steps of the above-mentioned resource scheduling method when executed by a processor.

[0156] The above is a schematic solution of a computer-readable storage medium of this embodiment. It should be noted that the technical solution of the storage medium and the technical solution of the resource scheduling method described above are of the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the resource scheduling method described above.

[0157] An embodiment of the present specification further provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned resource scheduling method.

[0158] The above is an illustrative solution of a computer program of this embodiment. It should be noted that the technical solution of the computer program and the technical solution of the resource scheduling method described above are of the same concept. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the resource scheduling method described above.

[0159] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0160] The computer instructions include computer program code, which may be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media does not include electric carrier signals and telecommunication signals.

[0161] It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0162] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0163] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A resource scheduling method, comprising: For multiple computing nodes, predicting resource load indicator predicted values ​​of the resource groups to be scheduled when the resource groups to be scheduled are deployed on the computing nodes; For the multiple computing nodes, using the actual values ​​of the resource load indicators of the computing nodes and the predicted values ​​of the resource load indicators of the resource groups to be scheduled when the resource groups to be scheduled are deployed on the computing nodes, calculating the matching degree between the computing nodes and the resource groups to be scheduled; Using the matching degree between the computing node and the resource group to be scheduled, for the resource group to be scheduled, selecting a target computing node that meets a preset scheduling condition from the plurality of computing nodes; The to-be-scheduled resource group is scheduled to the target computing node.

2. The method according to claim 1, further comprising: Monitoring actual values ​​of resource load indicators of the plurality of computing nodes; Filter out the rescheduling nodes whose actual values ​​of resource load indicators meet the preset rescheduling conditions from the multiple computing nodes; Filter out the resource groups that need to be rescheduled from the resource groups deployed by the rescheduling node; The resource group that needs to be rescheduled is taken as the resource group to be scheduled, and the step of predicting the resource load indicator prediction value of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing nodes for multiple computing nodes is re-entered, so as to reschedule the resource group to be scheduled.

3. The method according to claim 1, wherein the use of the actual value of the resource load indicator of the computing node and the predicted value of the resource load indicator of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node to calculate the matching degree between the computing node and the resource group to be scheduled comprises: By using the actual value of the resource load index of the computing node, the predicted values ​​of the resource load indexes of the various resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, and the weight coefficients corresponding to the various resources of the resource group to be scheduled, the comprehensive matching degree between the computing node and the resource group to be scheduled on the multiple resources is calculated.

4. The method according to claim 3, wherein the plurality of resources comprises: At least two resources selected from the group consisting of processor resources, memory access resources, input / output resources, and network bandwidth resources.

5. The method according to claim 4, wherein the plurality of resources comprises: Processor resources and memory access resources; The method of calculating the comprehensive matching degree between the computing node and the resource group to be scheduled on the multiple resources by using the actual value of the resource load index of the computing node, the predicted values ​​of the resource load index of multiple resources of the resource group to be scheduled, and the weight coefficients corresponding to the multiple resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, includes: Calculate the matching degree between the computing node and the resource group to be scheduled in terms of processor resources by using the actual value of the current resource load index of the processor resources of the computing node, the predicted value of the resource load index of the processor resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node, the logical processing core allocation value of the resource group to be scheduled, and the total number of logical processing cores of the computing node; Calculate the matching degree of the memory access resources between the computing node and the resource group to be scheduled by using the actual value of the current resource load index of the memory access resources of the computing node and the predicted value of the resource load index of the memory access resources of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node; The comprehensive matching degree between the computing node and the resource group to be scheduled is calculated by using the matching degree between the computing node and the resource group to be scheduled in terms of processor resources, the matching degree in terms of memory resources, the weight coefficient corresponding to the processor resources of the resource group to be scheduled, and the weight coefficient corresponding to the memory resources of the resource group to be scheduled. 6 . The method according to claim 5 , wherein when the resource group to be scheduled is a computing resource group, the weight coefficient corresponding to the processor resources of the resource group to be scheduled is greater than the weight coefficient corresponding to the memory access resources.

7. The method according to claim 5, when the resource group to be scheduled is a storage resource group, the weight coefficient corresponding to the processor resource of the resource group to be scheduled is smaller than the weight coefficient corresponding to the memory access resource.

8. According to the method of claim 5, the current resource load index of the processor resource of the computing node comprises: The current processor resource utilization of the computing node; The resource load index prediction value of the processor resource of the resource group to be scheduled includes: the processor resource utilization prediction value of the resource group to be scheduled; The actual value of the current resource load indicator of the memory access resource of the computing node includes: the current memory access delay of the computing node; The resource load indicator prediction value of the memory access resource of the to-be-scheduled resource group includes: a memory access delay prediction value of the to-be-scheduled resource group.

9. The method according to claim 8, wherein the step of using the matching degree between the computing node and the resource group to be scheduled to select a target computing node that meets a preset scheduling condition from the plurality of computing nodes for the resource group to be scheduled comprises: By using the comprehensive matching degree between the computing node and the resource group to be scheduled, a target computing node with the lowest comprehensive processor resource utilization and memory access delay is screened out from the multiple computing nodes for the resource group to be scheduled.

10. The method according to claim 1, wherein for multiple computing nodes, predicting the resource load indicator prediction value of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node comprises: Using a load indicator collector deployed on each computing node, collecting resource load indicators of the computing node and the resource group deployed on the computing node, and generating resource load indicator historical data; By using the resource load index historical data, for multiple computing nodes, predicted values ​​of the resource load index of the resource group to be scheduled when the resource group to be scheduled is deployed on the computing node are predicted.

11. A database comprising: A database service node, a resource scheduling service node and a resource pool, wherein the resource pool includes a plurality of computing nodes; The database service node is configured to receive a resource instance creation request, create a resource instance according to the resource instance creation request, and send a scheduling request to the resource scheduling service node, wherein the scheduling request is used to instruct to schedule one or more resource groups of the resource instance; The resource scheduling service node is configured to receive the scheduling request and schedule the resource group indicated by the scheduling request according to the resource scheduling method according to any one of claims 1 to 10.

12. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the resource scheduling method according to any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the resource scheduling method according to any one of claims 1 to 10.

Citation Information

Patent Citations

  • Container scheduling method and device and computer readable storage medium

    CN110888714A

  • Cloud computing resource allocation method and device

    CN111225050A

  • Resource allocation method in cloud computing system

    CN112948122A

  • Dynamic load balancing scheduling method and system based on Kubernetes platform resources

    CN113010270A

  • Resource scheduling method and database

    CN116743760A

Cited By

  • Multi-block chain load balancing method, gateway and building construction data management system

    CN120416246A

  • Scheduling method, electronic device, medium and computer program product

    CN122614526A