Cross-domain resource scheduling method and cross-domain resource scheduling platform
Through cross-domain resource scheduling methods and platforms, the resource scheduling system and perception system compute resource scores are used to solve the problem of computing resource separation, realize cross-domain resource scheduling and efficient utilization, and meet user needs.
Patent Information
- Application Number
- CN202411132849.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-19
AI Technical Summary
The computing power resources of the existing computing power center are separated, and users cannot use it across domains, have low resource utilization, and cannot achieve unified resource scheduling, which is limited by the differentiation of data storage and resource heterogeneity.
Provide cross-domain resource scheduling methods and platforms, including resource request interface, resource scheduling system and resource perception system. By obtaining the type attribute information and resource quantity attribute information of each alternative resource, calculate resource quantity scores based on type attributes and weights, and determine the target Resources and send resource startup requests to the target computing power center to realize cross-domain resource scheduling.
The computing power resource scheduling and use across wide-area networks is realized, resource utilization is improved, and user needs can be met in a timely and effective manner. Users do not need to understand the type attributes and resource quantity attributes of each computing power center, but only need to select the resource type.
Smart Images

Figure CN118642864B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of resource scheduling, and in particular to a cross-domain resource scheduling method and a cross-domain resource scheduling platform. Background Art
[0002] With the continuous development of modern information technologies such as big data, artificial intelligence, and 5G, the demand for computing resources is becoming more and more diversified. However, problems such as the lack of foresight in the construction of computing centers and the slow adaptation to changing economic and social needs have gradually become prominent, resulting in the current inefficient use of computing resources, low resource utilization, and the inability to meet user needs in a timely and effective manner.
[0003] The main problems of existing computing centers are as follows: 1. The computing resources of computing centers are fragmented, and users cannot use computing resources across domains. In particular, existing users usually do not change computing centers for computing due to the huge cost of migrating applications and data. 2. The resources of each computing center cannot be effectively scheduled from the user's perspective. General computing applications such as HPC (High Performance Computing), AI (Artificial Intelligence), and cloud can only schedule resources in a single computing center. This is difficult to achieve unified resource scheduling due to the limitations of remote data storage, heterogeneous resources, and whether user needs are met.
[0004] In view of this, the present invention is proposed. Summary of the invention
[0005] In order to solve the above technical problems, the present invention provides a cross-domain resource scheduling method and a cross-domain resource scheduling platform to realize unified scheduling of resources across computing power centers and solve the problem that computing power resources cannot be used across domains in the prior art.
[0006] An embodiment of the present invention provides a cross-domain resource scheduling method, which is applied to a cross-domain resource scheduling platform. The cross-domain resource scheduling platform includes a resource request interface, a resource scheduling system, and a resource perception system. The method includes:
[0007] The resource scheduling system, in response to detecting the resource call request sent by the resource request interface, obtains type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through the resource perception system;
[0008] The resource scheduling system determines a valid resource from all candidate resources based on the type attribute information in the resource call request and the type attribute information of each candidate resource;
[0009] The resource scheduling system determines the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weight and user weight corresponding to each resource quantity attribute item, and determines the target resource based on all resource quantity scores;
[0010] The resource request interface sends a resource start request to the computing center corresponding to the target resource, so that the computing center starts the target resource and links to the user end.
[0011] An embodiment of the present invention provides a cross-domain resource scheduling platform, which includes a resource request interface, a resource scheduling system, and a resource perception system, wherein:
[0012] The resource perception system is used to obtain type attribute information and resource quantity attribute information of resources in each computing power center;
[0013] The resource request interface is used to obtain the resource call request sent by the user end, and send the resource call request to the resource scheduling system, obtain the target resource fed back by the resource scheduling system, and send a resource start request to the computing center corresponding to the target resource, so that the computing center starts the target resource and links to the user end;
[0014] The resource scheduling system is used to respond to the resource call request sent by the resource request interface, obtain the type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through the resource perception system, determine the valid resources among all the candidate resources based on the type attribute information in the resource call request and the type attribute information of each candidate resource, determine the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weight and user weight corresponding to each resource quantity attribute item, and determine the target resource based on all resource quantity scores, and feed the target resource back to the resource request interface.
[0015] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a program or instruction, wherein the program or instruction enables a computer to execute the steps of the cross-domain resource scheduling method described in any embodiment.
[0016] The embodiments of the present invention have the following technical effects:
[0017] The resource scheduling system of the cross-domain resource scheduling platform, when detecting a resource call request sent by a resource request interface, obtains the type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through the resource perception system, and then determines the valid resource based on the type attribute information in the resource call request and the type attribute information of each candidate resource, and for each valid resource, calculates the resource quantity score according to the resource quantity attribute information, and the preset weight and user weight corresponding to each resource quantity attribute item, determines the target resource through the resource quantity score, and finally the resource request interface sends a resource start request to the computing power center corresponding to the target resource, so that the computing power center starts the target resource and links to the user end, realizes the scheduling and use of computing power resources across the wide area network, and can perform user-level scheduling on resources of different architectures and different systems on different computing power centers. Users do not need to understand the type attributes and resource quantity attributes of each computing power center, but only need to select the type attributes of the resource, which ensures the efficient use of computing power resources and improves resource utilization. At the same time, it can also meet user needs in a timely and effective manner. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0019] Figure 1 is a flow chart of a cross-domain resource scheduling method provided by an embodiment of the present invention;
[0020] Figure 2 It is a schematic diagram of a resource perception system acquiring information provided by an embodiment of the present invention;
[0021] Figure 3 This is a schematic diagram of resource scheduling provided by an embodiment of the present invention;
[0022] Figure 4 It is a schematic diagram of a process of writing data in a data transmission interface provided by an embodiment of the present invention;
[0023] Figure 5 is a schematic diagram of a consistency judgment process provided by an embodiment of the present invention;
[0024] Figure 6 It is a structural diagram of a cross-domain resource scheduling platform provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0025] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.
[0026] The cross-domain resource scheduling method provided by the embodiment of the present invention is applied to a cross-domain resource scheduling platform. The cross-domain resource scheduling platform includes a resource request interface, a resource scheduling system and a resource perception system. Figure 1 is a flow chart of a cross-domain resource scheduling method provided by an embodiment of the present invention. Figure 1 , the cross-domain resource scheduling method specifically includes:
[0027] S110. In response to detecting a resource call request sent by a resource request interface, the resource scheduling system obtains type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through a resource perception system.
[0028] The resource request interface may be a communication interface for the user end in the cross-domain resource scheduling platform, which is used to receive resource call requests and resource link requests sent by the user end, and to feed back resource link usage information to the user end. The user end may specifically run in the form of a browser, and generate corresponding resource call requests or resource link requests according to user operations and send them to the resource request interface.
[0029] In the embodiment of the present invention, in response to receiving a resource call request sent by the user end, the resource request interface may forward the resource call request to the resource scheduling system so as to schedule corresponding target resources for the user through a scheduling algorithm in the resource scheduling system.
[0030] Specifically, after receiving the resource call request, the resource scheduling system may first obtain the type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through the resource perception system in the cross-domain resource scheduling platform.
[0031] Among them, the candidate resources corresponding to the user can be the resources that the user can use, such as the resources purchased in advance by the user. It should be noted that the resource can be a single node, that is, a single physical machine, or the resource can also include multiple nodes, such as an HPC platform, etc. The embodiment of the present invention can be scheduled to the target resource. If the target resource is a single node, the target resource can directly provide computing services to the user. If the target resource includes multiple nodes, the target resource can continue to further determine the target node through an internal scheduling algorithm, and the target node provides computing services to the user.
[0032] Specifically, the type attribute information can describe the values of all type attribute items of the resource. For example, the type attribute item can be the platform type, computing architecture, intelligent computing architecture, storage architecture, operating system or software version.
[0033] As shown in Table 1, the values of each type attribute item in the type attribute information are shown, and the type attribute information includes platform type, computing architecture, intelligent computing architecture, storage architecture, operating system and software version. Among them, the platform type can be HPC, AI, Cloud or others; the computing architecture can be X86, ARM (Advanced RISC Machine, reduced instruction set machine) or others; the intelligent computing architecture can be H100, 910b or A100; the storage architecture can be HDFS (Hadoop Distributed File System, distributed file system), Lustre, Block, Object or NFS (Network File System, network file system); the operating system can be RedHat, Ubuntu or CentOS; the software version can be Ansys1.0, Fluent2.0 or Jupyter3.0.
[0034] Table 1 Values of each type attribute item in type attribute information
[0035]
[0036] For example, Table 2 shows the type attribute information of all candidate resources of computing center A, and Table 3 shows the type attribute information of all candidate resources of computing center B.
[0037] Table 2 Type attribute information of all candidate resources of computing center A
[0038]
[0039] Table 3 Type attribute information of all candidate resources of computing center B
[0040]
[0041] In addition, resource quantity attribute information can describe the values of all resource quantity attribute items of the resource (including the total amount and the remaining amount). For example, the resource quantity attribute items can be the number of nodes, the number of CPU (Central Processing Unit) cores, the number of GPU (Graphics Processing Unit) cores, memory, storage, and storage bandwidth.
[0042] Specifically, the resource perception system can pull resource information, including type attribute information and resource quantity attribute information, from the monitoring interface of each computing center in real time, and store it in the database, and provide RPC (Remote Procedure Call) service to the resource scheduling system so that the resource scheduling system can call the resource perception system to obtain the type attribute information and resource quantity attribute information of each alternative resource from the database.
[0043] like Figure 2 As shown, Figure 2 This is a schematic diagram of the resource perception system obtaining information provided by an embodiment of the present invention. Taking the number of computing centers as 3 as an example, the resource perception system can periodically obtain the type attribute information and resource quantity attribute information of each resource in the center from computing center A, computing center B, and computing center C through a scheduled task, and write it into the database. The resource perception system can also provide an RPC service for the resource scheduling system. If the resource scheduling system responds to a resource call request and calls the RPC service, the resource perception system can read the type attribute information and resource quantity attribute information of the resource from the database.
[0044] Among them, the cross-domain resource scheduling platform provided by the embodiment of the present invention may also include a resource monitoring system, and the resource perception system can obtain the type attribute information and resource quantity attribute information of all schedulable resources of each computing power center through the resource monitoring system for use by the cross-domain resource scheduling platform. The resource monitoring system can be composed of a main service and a resource monitoring agent component (agent) deployed in each computing power center. The main service can be responsible for obtaining the information of all schedulable resources of each computing power center in real time by monitoring the overhead data interface provided by each agent, and provide it to the resource perception system.
[0045] S120: The resource scheduling system determines a valid resource from all candidate resources based on the type attribute information in the resource call request and the type attribute information of each candidate resource.
[0046] In an embodiment of the present invention, after obtaining the type attribute information and resource quantity attribute information of each candidate resource, the resource scheduling system can first select resources whose type attribute information meets the user's needs from the candidate resources as valid resources. Valid resources can be candidate resources that can provide computing services to users and meet the user's needs for resource types.
[0047] The resource call request may include the type attribute information and resource quantity attribute information requested by the user. The resource scheduling system may select the candidate resources whose type attribute information meets the type attribute information of the user request from all candidate resources as valid resources.
[0048] In a specific implementation, the resource scheduling system determines a valid resource from all candidate resources based on the type attribute information in the resource call request and the type attribute information of each candidate resource, including:
[0049] The resource scheduling system constructs a type compliance vector for each candidate resource based on the type attribute information in the resource call request and the type attribute information of each candidate resource, where each element in the type compliance vector indicates whether the platform type, computing architecture, intelligent computing architecture, storage architecture, operating system and software version of the candidate resource meet the resource call request; based on the type compliance vector of each candidate resource, valid resources are determined among all candidate resources.
[0050] For each candidate resource, the resource scheduling system may construct a type compliance vector indicating whether the candidate resource contains the type attribute information requested by the user; in the type compliance vector, 1 represents that the requested value is contained, and 0 represents that the requested value is not contained.
[0051] For example, assuming that the type attribute information requested by the user in the resource call request is: hpc-x86-''-lustre-redhat-ansys1.0, the type matching vectors of all three candidate resources under the computing center A in the above example can form a matrix Ra:
[0052] ;
[0053] The type matching vectors of all two candidate resources under computing center B can form a matrix Rb:
[0054] ;
[0055] Furthermore, the resource scheduling system can select candidate resources whose types match that of all elements in the vector are 1 as valid resources. For example, the first candidate resource in computing center A and the first candidate resource in computing center B in the above example can be used as valid resources. The valid resource ID set can be expressed as:
[0056] ;
[0057] S130. The resource scheduling system determines the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weight and user weight corresponding to each resource quantity attribute item, and determines the target resource based on all resource quantity scores.
[0058] Specifically, after determining each valid resource, the resource scheduling system may further calculate a resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, as well as a preset weight and a user weight.
[0059] Among them, the preset weight is a weight pre-set for the resource quantity attribute item, which can be calibrated according to historical scheduling data, and the user weight can be a weight set by the user for the resource quantity attribute item.
[0060] In an embodiment of the present invention, the preset weight can be used to weight the total amount of the resource attribute item, and the user weight can be used to weight the remaining amount of the resource attribute item. As shown in Table 4, Table 4 shows the resource attribute information of each valid resource, including the total amount Tv and the remaining amount Rv of each resource attribute item, where Ra-1 is the first candidate resource (i.e., valid resource) in computing center A, and Rb-1 is the first candidate resource (i.e., valid resource) in computing center B.
[0061] Table 4 Resource quantity attribute information of each effective resource
[0062]
[0063] Table 5 shows the preset weight Ws of each resource attribute item, which can be maintained according to historical data. The value range of the preset weight Ws is [0,10]. Table 6 shows the user weight Wu of each resource attribute item. The value range of the user weight Wu is [0,10].
[0064] Table 5 Preset weights of various resource attribute items
[0065]
[0066] Table 6 User weights of each resource attribute item
[0067]
[0068] In a specific implementation, the resource scheduling system determines the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weight and user weight corresponding to each resource quantity attribute item, including:
[0069] For each valid resource, the resource scheduling system performs weighted calculation on the total amount and remaining amount of each resource quantity attribute item in the resource quantity attribute information according to the preset weight and user weight corresponding to the resource quantity attribute item, and obtains the score of the resource quantity attribute item; the resource scheduling system determines the resource quantity score of the valid resource based on the scores of all resource quantity attribute items.
[0070] Among them, the preset weight corresponding to the resource quantity attribute item can be multiplied by the total amount of the resource quantity attribute item, and the user weight corresponding to the resource quantity attribute item can be multiplied by the remaining amount of the resource quantity attribute item, and then the two multiplied results can be added together to obtain the score of the resource quantity attribute item.
[0071] Furthermore, the scores of all resource quantity attribute items can be added together to obtain the resource quantity score of the effective resource. As shown in the following formula:
[0072] ;
[0073] In the formula, is the resource score, k is the index, U-1 is the number of resource attribute items, is the total amount of the k-th resource attribute item, is the preset weight corresponding to the k-th resource attribute item, is the remaining amount of the k-th resource attribute item, is the user weight corresponding to the kth resource quantity attribute item.
[0074] Through the above implementation, the resource quantity score of each effective resource can be determined based on the preset weight and the user weight to evaluate the resource quantity attribute item of each effective resource. The effective resources that can better meet user needs can be selected based on the resource quantity score to ensure the user's usage experience.
[0075] It should be noted that in the embodiment of the present invention, the total amount is weighted by the preset weight, and the remaining amount is weighted by the user weight, so that the user can adaptively adjust the user weight according to his or her own needs for the number of nodes, the number of CPU cores, the number of GPU cores, memory, storage, and storage bandwidth. For example, if the user has a higher demand for the number of GPU cores and expects faster data processing efficiency, or the user has a higher demand for storage and expects to store large quantities of data, in this way, resource scheduling can be made more targeted, and the scheduled resources can be made more in line with user needs. In addition, the preset weight can also weight the total amount to simultaneously consider the impact of the total amount such as the number of nodes and the number of CPU cores. While ensuring that the scheduled resources meet the user's needs, it can also try to schedule a larger total amount of resources for the user to further improve resource utilization.
[0076] Specifically, after obtaining the resource quantity scores of all valid resources, they may be sorted in descending order according to the resource quantity scores, and the valid resource with the highest resource quantity score may be determined as the target resource.
[0077] In an embodiment of the present invention, considering that each computing power center may be in a different wide area network, the connectivity, latency and other parameters from different computing power centers to the user request location may be different, and these parameters may affect the efficiency and stability of the user-side connection to the resources in the computing power center. Therefore, before determining the resource quantity score and scheduling the computing power center resources according to the resource quantity score, the computing power center can also be scheduled first.
[0078] Optionally, before the resource scheduling system determines the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weight and user weight corresponding to each resource quantity attribute item, the following steps are further included:
[0079] Step 11: The resource scheduling system determines whether all valid resources are from the same computing center. If not, the computing center corresponding to each valid resource is determined as the computing center to be determined.
[0080] Step 12: The resource scheduling system obtains the connection attribute information of each computing power center to be determined through the resource perception system, and determines the target computing power center based on the connection attribute information of each computing power center to be determined;
[0081] Step 13: If there are multiple valid resources in the target computing power center, the resource scheduling system determines the resource quantity score of each valid resource in the target computing power center based on the resource quantity attribute information of each valid resource, and the preset weights and user weights corresponding to each resource quantity attribute item.
[0082] Among them, in step 11, the resource scheduling system can first determine whether all valid resources come from the same computing power center. If so, the resource quantity score of each valid resource can be directly determined, thereby realizing resource scheduling within the same computing power center.
[0083] If all valid resources do not come from the same computing center, it means that the valid resources are distributed in multiple computing centers, and the computing centers can be scheduled first. Specifically, the computing centers corresponding to each valid resource can be determined as the computing centers to be determined.
[0084] Furthermore, in step 12, the resource scheduling system can obtain the connection attribute information of each computing power center to be judged through the resource perception system, wherein the connection attribute information is used to describe the values of each connection attribute item of the computing power center; the connection performance of the computing power center to be judged can be scored through the connection attribute information, thereby screening out the target computing power center.
[0085] With respect to the above step 12, in one example, the connection attribute information includes connectivity value, network delay, network bandwidth, network stability, and historical resource average score. Based on the connection attribute information of each computing power center to be determined, the target computing power center is determined, including:
[0086] The computing power centers to be judged whose connectivity values are represented as disconnected are eliminated. For each remaining computing power center to be judged, the central connection score of the computing power center to be judged is determined based on the network delay, network bandwidth, network stability and historical resource average score of the computing power center to be judged; the computing power center to be judged with the largest central connection score is determined as the target computing power center.
[0087] Among them, the connectivity value Np is 0 for connected, and the connectivity value Np is 1 for disconnected. Network delay Nd and network bandwidth Nb are real-time data. The user end can initiate a test to each computing center to be judged to calculate the delay and real-time bandwidth, and send it to the cross-domain resource scheduling platform after the calculation is completed. Network stability Ns is the calculated value of historical data, and the historical resource average score Cc is the average value of the resource quantity score of the resources scheduled in the computing center to be judged, which can be understood as the computing resource compliance, that is, reflecting the degree to which the resources in the computing center to be judged meet user needs.
[0088] Specifically, the resource scheduling system can first remove the computing power centers to be judged whose connectivity values are represented as disconnected, and then calculate the center connection score of each computing power center to be judged based on the network delay, network bandwidth, network stability and historical resource average score of the remaining computing power centers to be judged. For example, the center connection score CCs = Nd*P1 + Nb *P2 + Ns *P3 + Cc * P4, where P1-P4 are empirical parameters.
[0089] Furthermore, the computing power center to be judged with the largest center connection score can be used as the target computing power center to realize the scheduling of the computing power center, which can ensure the stability and real-time connection between the computing power center where the resources are finally scheduled and the user's requested location, thereby ensuring the reliability of subsequent data processing and data transmission.
[0090] After determining the target computing power center, further, in step 13, the resource scheduling system can continue to determine whether the number of valid resources in the target computing power center is multiple. If so, the resource scheduling system further determines the resource quantity score of each valid resource to achieve resource scheduling. If not, the valid resources in the target computing power center can be directly determined as the target resources without the need to calculate the resource quantity score.
[0091] Through the above steps 11 to 13, before scheduling computing power resources, the computing power center can be scheduled first, and the target computing power center can be determined through the connection attribute information of the computing power center corresponding to each effective resource, and then resource scheduling can be performed in the target computing power center. While ensuring that the amount of resources meets user needs, the stability and real-time nature of the connection between resources and the user end can be guaranteed, thereby ensuring the reliability and efficiency of data transmission.
[0092] S140. The resource request interface sends a resource start request to the computing center corresponding to the target resource, so that the computing center starts the target resource and links it to the user end.
[0093] Specifically, after the resource scheduling system determines the target resource, the resource scheduling system can further feedback the resource identifier of the target resource to the resource request interface, and the resource request interface can send a resource start request to the computing power center corresponding to the target resource. For example, the resource request interface can obtain the API where the target resource is located, and request the computing power center to start the target resource through the API.
[0094] In a specific implementation, the resource request interface sends a resource start request to the computing center corresponding to the target resource, so that the computing center starts the target resource and links it to the user end, including:
[0095] The resource request interface sends a resource start request to the computing power center corresponding to the target resource, and obtains the resource link usage information fed back by the computing power center; the resource request interface pushes the resource link usage information to the user end, and obtains the resource link request fed back by the user end, and in response to the resource link request, links the target resource to the user end.
[0096] Among them, the resource request interface can send a resource start request to the computing power center corresponding to the target resource, so that the computing power center starts the target resource and returns the resource link usage information to the resource request interface.
[0097] Furthermore, the resource request interface can send resource link usage information to the user end, and wait to receive the resource link request from the user end, and in response to receiving the resource link request, link the target resource to the user end for use. The resource request interface can be a standard RESTful (Representational State Transfer) API.
[0098] As Figure 3 As shown, Figure 3 This is a resource scheduling diagram provided by an embodiment of the present invention. First, the resource request interface can send the received resource scheduling request to the resource scheduling system. The resource scheduling system determines each valid resource, obtains the type attribute information and resource quantity attribute information of the valid resource from the resource perception system, and then calculates the resource quantity score, determines the target resource, and requests the start of the target resource through the resource request API. After starting the target resource, the computing power center can feedback the resource link and usage information to the resource request interface. Among them, the resource link and usage information can be a URL (Uniform Resource Locator).
[0099] In an embodiment of the present invention, it is considered that data between different computing power centers cannot be efficiently communicated in the prior art. For example, it is difficult to repeat the calculation of data calculated by a user in computing power center A in computing power center B, because data needs to be transmitted across domains between the two computing power centers, and the user must manually copy and transmit the data. Calculations between different computing power centers require uploading data multiple times, resulting in a waste of resources; and, the prior art cannot effectively realize efficient request and use of resources between cross-domain computing power centers. For example, if the data and services of users of computing power center A want to request the same computing power resources for calculation in computing power center B, it is necessary to first deploy relevant computing software in computing power center B, and then suspend the service from computing power center A, download the data, transmit the data to computing power center B, install the computing software in computing power center B, and restart the computing task. The process is very complicated and the flow is too cumbersome.
[0100] Therefore, in order to solve the above problems, in an embodiment of the present invention, the cross-domain resource scheduling platform may further include a cloud storage module and a data transmission interface. The data transmission interface may be based on the cloud storage module to provide users with a unified data transmission service. When users use the data transmission interface, from the user's perspective, the data is only transmitted to the scheduled computing power center. In fact, the data transmission interface will automatically synchronize the data to the cloud storage module for storage. The cloud storage module supports dynamic high-speed bandwidth networks and uses asynchronous transmission. Users cannot actually see the synchronization process, which reduces the complexity of user data transmission.
[0101] In a specific implementation, the cross-domain resource scheduling platform further includes a cloud storage module and a data transmission interface, and the method provided by the embodiment of the present invention further includes the following steps:
[0102] Step 21, in response to detecting the data transmission request, the data transmission interface calls the unified resource storage API to write the data to be transmitted into the corresponding resource, and creates stream data according to the data to be transmitted;
[0103] Step 22: The data transmission interface calls the asynchronous task service to create a data upload task, obtains the data to be transmitted from the stream data through the asynchronous task service, and calls the cloud storage API to write the data to be transmitted into the cloud storage module.
[0104] Among them, the data transmission request can be generated when the user specifies the resources of the computing power center in the user terminal, or it can also be automatically generated in response to the detection of data generation during the operation of the target resource.
[0105] Specifically, in step 21, the data transmission interface can call the uniform resource storage API in response to detecting a data transmission request to write the data to be transmitted to the resource specified in the data transmission request, for example, the resource selected by the user or the target resource; and the uniform resource storage API can also create stream data based on the data to be transmitted.
[0106] Furthermore, in step 22, the data transmission interface may call the asynchronous task service to create a data upload task, obtain the data to be transmitted from the stream data through the asynchronous task service, and then call the cloud storage API to write it into the cloud storage module.
[0107] like Figure 4 As shown, Figure 4 This is a schematic diagram of the process of writing data in a data transmission interface provided by an embodiment of the present invention. First, the data transmission interface can receive a user's request to transmit data, call the unified resource storage API, transfer the data to be transmitted in blocks to the resources of the computing power center, and create streaming data in the form of data streams; further, the data transmission interface calls the asynchronous task service to create a task, and the asynchronous task service can read data blocks from the streaming data, and transfer the read data to be transmitted in blocks to the cloud storage module through the cloud storage API. The process of uploading to the cloud storage module is transparent to the user.
[0108] Through the above steps 21 and 22, users only need to transfer data once to resources and high-speed cloud storage modules, which reduces the complexity of user data transmission. It can also help to effectively realize efficient request and use of data between cross-domain computing centers and improve the efficiency of resource data processing.
[0109] In the embodiment of the present invention, an exemplary description may be given of obtaining data to be processed by the target resource from the cloud storage module. Optionally, the method provided in the embodiment of the present invention further includes:
[0110] In response to detecting a pending task for a target resource, the resource scheduling system determines whether the task processing data corresponding to the pending task is cross-domain data, where the cross-domain data is data processed or obtained by resources of other computing centers that are different from the computing center of the target resource; if so, the resource scheduling system obtains the task processing data from the cloud storage module and sends the task processing data to the target resource.
[0111] The task to be executed may be a data processing task set by the user at the user end. Specifically, in response to detecting the task to be executed of the target resource, the resource scheduling system may determine whether the task processing data of the task to be executed is cross-domain data, that is, whether it is data processed or processed by resources of other computing centers.
[0112] If the task processing data is cross-domain data, the resource scheduling system can obtain the task processing data from the cloud storage module and then send it to the target resource.
[0113] Through the above method, when the target resource needs to process data processed or obtained by other computing power centers, the corresponding data can be directly pulled from the cloud storage module without the need for user uploading. This enables efficient request and use of data between cross-domain computing power centers, reducing the complexity of operations for users while improving the efficiency of task execution.
[0114] In an embodiment of the present invention, considering that if the task processing data of the task to be executed by the target resource is cross-domain data, there may be a situation where the target resource and the resources of other computing power centers execute the same task for the same batch of data. For such a situation, it is also possible to verify whether the data processed by the target resource is consistent with the data processed by other computing power centers.
[0115] In one example, after sending the task processing data to the target resource, the following steps are also included:
[0116] Step 31, the resource scheduling system determines whether the task to be executed is a task that has been executed by resources of other computing centers. If so, the actual result data of the task to be executed is obtained during the execution of the task to be executed;
[0117] Step 32: The resource scheduling system obtains historical result data of other computing centers from the cloud storage module or other computing centers, and determines whether the actual result data and the historical result data meet the consistency condition;
[0118] Step 33: If not, the resource scheduling system displays the coverage request to the user terminal through the resource request interface, and obtains the coverage confirmation instruction fed back by the user terminal;
[0119] Step 34: In response to the overwrite confirmation instruction, the resource scheduling system uploads the actual result data to the cloud storage module and overwrites the historical result data.
[0120] Among them, in step 31, the resource scheduling system can determine whether the task to be executed is a task that has been executed by resources of other computing centers. If so, the actual result data generated by the task can be obtained during the execution process.
[0121] Furthermore, in step 32, the resource scheduling system can obtain the historical result data generated by other computing centers executing the task from the cloud storage module or other computing centers, and perform consistency calculation on the historical result data and the actual result data to determine whether the consistency condition is met. For example, the similarity between the two is calculated, and if the similarity is greater than a preset threshold, it can be determined that the consistency condition is met.
[0122] Furthermore, in step 33, if the consistency condition is met, the task to be executed can continue to be processed. If the consistency condition is not met, it means that there is a difference between the historical result data and the actual result data. At this time, the resource scheduling system can display an overwrite request to the user terminal through the resource request interface to ask the user whether the actual result data needs to be used to overwrite the historical result data.
[0123] If the resource scheduling system obtains the overwrite confirmation instruction from the user, the actual result data can be uploaded to the cloud storage module and the historical result data can be overwritten. At the same time, the resource scheduling system can also send the actual result data to other computing power centers so that the resources of other computing power centers can use the actual result data to overwrite the historical result data.
[0124] like Figure 5 As shown, Figure 5 This is a flow chart of consistency judgment provided by an embodiment of the present invention. First, the resource scheduling interface (which may be in the resource scheduling system) responds to the cross-domain resource call request and can determine the data selected by the user. Assuming that the data selected by the user is from computing center A, and the target resource is from computing center B, the data of computing center A can be pulled from the cloud storage module.
[0125] Furthermore, the pulled data is transmitted across domains to computing center B, and in the process of computing center B executing tasks, the consistency between the pulled data and the data generated by computing center B is calculated. If the consistency condition is not met, it can be blocked to wait for computing center B to complete the transmission, and then computing center B continues to execute the task. After the task is completed, the resource scheduling interface can obtain all actual result data through the reading interface and upload the result data to the cloud storage module. At the same time, computing center B itself can also synchronize the results to the cloud storage module. It should be noted that Figure 5 The resource scheduling interface pulls data from the cloud storage module and uploads result data, which can be transmitted at a high rate and high bandwidth.
[0126] Through the above steps 31 to 34, when the target resource processes cross-domain data, consistency judgment can be performed based on the data obtained by other computing centers that execute the same task, thereby ensuring the accuracy of the data generated by task execution.
[0127] The present invention has the following technical effects: when a resource call request sent by a resource request interface is detected, the resource scheduling system of the cross-domain resource scheduling platform obtains the type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through the resource perception system, and then determines the valid resource based on the type attribute information in the resource call request and the type attribute information of each candidate resource, and for each valid resource, calculates the resource quantity score according to the resource quantity attribute information, and the preset weight and user weight corresponding to each resource quantity attribute item, determines the target resource through the resource quantity score, and finally the resource request interface sends a resource start request to the computing power center corresponding to the target resource, so that the computing power center starts the target resource and links to the user end, realizes the scheduling and use of computing power resources across the wide area network, can perform user-level scheduling on resources of different architectures and different systems on different computing power centers, and the user does not need to understand the type attributes and resource quantity attributes of each computing power center, but only needs to select the type attribute of the resource, thereby ensuring the efficient use of computing power resources and improving resource utilization. At the same time, it can also meet user needs in a timely and effective manner.
[0128] The embodiment of the present invention further provides a cross-domain resource scheduling platform, which includes a resource request interface, a resource scheduling system and a resource perception system, wherein:
[0129] The resource perception system is used to obtain type attribute information and resource quantity attribute information of resources in each computing power center;
[0130] The resource request interface is used to obtain the resource call request sent by the user end, and send the resource call request to the resource scheduling system, obtain the target resource fed back by the resource scheduling system, and send a resource start request to the computing center corresponding to the target resource, so that the computing center starts the target resource and links to the user end;
[0131] The resource scheduling system is used to respond to the resource call request sent by the resource request interface, obtain the type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through the resource perception system, determine the valid resources among all the candidate resources based on the type attribute information in the resource call request and the type attribute information of each candidate resource, determine the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weight and user weight corresponding to each resource quantity attribute item, and determine the target resource based on all resource quantity scores, and feed the target resource back to the resource request interface.
[0132] In some implementations, the resource scheduling system is further configured to:
[0133] Based on the type attribute information in the resource call request and the type attribute information of each candidate resource, a type compliance vector for each candidate resource is constructed, wherein each element in the type compliance vector respectively indicates whether the platform type, computing architecture, intelligent computing architecture, storage architecture, operating system and software version of the candidate resource meet the resource call request; based on the type compliance vector of each candidate resource, valid resources are determined among all candidate resources.
[0134] In some implementations, the resource scheduling system is further configured to:
[0135] For each of the valid resources, for each resource quantity attribute item in the resource quantity attribute information, weighted calculation is performed on the total amount and the remaining amount of the resource quantity attribute item according to the preset weight and user weight corresponding to the resource quantity attribute item to obtain the score of the resource quantity attribute item; based on the scores of all resource quantity attribute items, the resource quantity score of the valid resource is determined.
[0136] In some implementations, the resource scheduling system is further configured to:
[0137] Determine whether all valid resources are derived from the same computing power center. If not, determine the computing power center corresponding to each valid resource as the computing power center to be determined; obtain the connection attribute information of each computing power center to be determined through the resource perception system, and determine the target computing power center based on the connection attribute information of each computing power center to be determined; if there are multiple valid resources in the target computing power center, then for each valid resource in the target computing power center, perform the steps of determining the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weights and user weights corresponding to each resource quantity attribute item.
[0138] In some implementations, the connection attribute information includes connectivity values, network delay, network bandwidth, network stability, and historical resource average scores, and the resource scheduling system is further configured to:
[0139] The computing power centers to be judged whose connectivity values are represented as disconnected are eliminated. For each remaining computing power center to be judged, the central connection score of the computing power center to be judged is determined based on the network delay, network bandwidth, network stability and historical resource average score of the computing power center to be judged; the computing power center to be judged with the largest central connection score is determined as the target computing power center.
[0140] In some implementations, the resource request interface is further used to:
[0141] Send a resource startup request to the computing power center corresponding to the target resource, and obtain the resource link usage information fed back by the computing power center; push the resource link usage information to the user end, and obtain the resource link request fed back by the user end, and in response to the resource link request, link the target resource to the user end.
[0142] In some implementations, the cross-domain resource scheduling platform further includes a cloud storage module and a data transmission interface, wherein:
[0143] A data transmission interface is used to respond to the detection of a data transmission request, call the unified resource storage API to write the data to be transmitted into the corresponding resource, and create stream data based on the data to be transmitted; call the asynchronous task service to create a data upload task, and obtain the data to be transmitted from the stream data through the asynchronous task service, and call the cloud storage API to write the data to be transmitted into the cloud storage module.
[0144] In some implementations, the resource scheduling system is further configured to:
[0145] In response to detecting a task to be executed of the target resource, determine whether the task processing data corresponding to the task to be executed is cross-domain data, wherein the cross-domain data is data processed or obtained by resources of other computing centers different from the computing center of the target resource; if so, obtain the task processing data from the cloud storage module and send the task processing data to the target resource.
[0146] In some implementations, the resource scheduling system is further configured to:
[0147] Determine whether the task to be executed is a task that has been executed by resources of other computing power centers. If so, obtain the actual result data of the task to be executed during the execution of the task to be executed; obtain the historical result data of other computing power centers from the cloud storage module or the other computing power centers, and determine whether the actual result data and the historical result data meet the consistency condition; if not, display the overwrite request to the user terminal through the resource request interface, and obtain the overwrite confirmation instruction fed back by the user terminal; in response to the overwrite confirmation instruction, upload the actual result data to the cloud storage module, and overwrite the historical result data.
[0148] like Figure 6 As shown, Figure 6 It is a structural diagram of a cross-domain resource scheduling platform provided by an embodiment of the present invention, wherein the platform includes a data transmission interface, a resource request interface, a cloud storage module, a resource scheduling system, a resource perception system, a resource monitoring system, and a resource monitoring interface, a resource scheduling interface and a data synchronization interface set for each computing power center.
[0149] Among them, the data transmission interface is used to provide users with a unified data transmission service. Users only need to transmit once to transfer pre-calculated data to resources, and at the same time hide the transmission to the designated high-speed cloud storage. This process is asynchronous and transparent to users. The resource usage interface can receive user resource call requests and resource link requests, and schedule resources through the scheduling algorithm of the resource scheduling system. After the scheduling is completed, the resource start request is sent to the specific computing center. The computing center starts the target resource and returns the resource link and usage information, and waits to receive the user's resource link request, linking the target resource for the user to use. The request interface and the link interface are both standard RESTful APIs.
[0150] The resource scheduling system is the core system. It can obtain resource information through the resource perception system, calculate the score of effective resources through the scheduling algorithm, determine the target resource and obtain the API where the target resource is located, request to create resources, return the resource link and the used URL, authentication and other information to the interface, and wait for the user to connect.
[0151] The resource scheduling system can also call the resource scheduling interface to start or shut down connected resources, and call the data synchronization interface to start or shut down data transmission. The resource monitoring system can call the resource monitoring interface to monitor each computing center in real time. The cloud storage module can call the data synchronization interface to achieve unified data synchronization.
[0152] The method and platform provided by the embodiments of the present invention have the following technical effects:
[0153] 1. Ability to realize computing resource scheduling across wide area networks, such as user-level cross-domain resource request scheduling of CPU, GPU and other resources in different computing centers, different platforms and different architectures (such as X86 and ARM architectures);
[0154] 2. Based on the object storage on the cloud, a unified data transmission interface is provided. When users use the unified data transmission interface, from the user's perspective, the data is only transmitted to the computing center dispatched by the platform. In fact, the data transmission interface will automatically synchronize the data to the cloud object storage. The storage supports dynamic high-speed bandwidth network and adopts asynchronous transmission mode. Users cannot actually see the synchronization process.
[0155] 3. For all resource requests and uses, users do not need to care about which computing center the specific resources belong to. All requests and uses are automatically forwarded through the resource request interface. The forwarding process is transparent to users. Users can select the type and attributes of resources, such as application platform (HPC, cloud computing, intelligent computing, physical server, operating system, application software, etc.), resource architecture (X86, ARM, etc.), required total memory size, required number of nodes, required storage space size, required number of GPU cards, etc. After submitting the resource call request, the platform automatically allocates resources, creates a system account, and guides users to log in to the platform to use resources;
[0156] 4. Based on the monitoring system deployed in each computing power center, the monitoring data is automatically pulled and summarized at a fixed time to form a computing power resource usage database. The resource call system will compare, match and calculate the user's resource call request with the data in the database, obtain the resource scheduling interface with a high score, create user resources and provide a resource request interface;
[0157] 5. Use secure encrypted transmission channels to encrypt all transmitted data, and authorize the use of resources through user authorization to ensure that user-level data flows safely and reliably across different data centers and clouds.
[0158] 6. Provide an efficient resource scheduling algorithm. Users only need to provide resource requirements without having to understand the resource types and sizes of each computing center.
[0159] In addition to the above methods and devices, an embodiment of the present invention may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the cross-domain resource scheduling method provided by any embodiment of the present invention.
[0160] It should be noted that the terms used in the present invention are only for describing specific embodiments, rather than limiting the scope of the present application. As shown in the present specification, unless the context clearly indicates an exception, the words "one", "a", "a kind of" and / or "the" do not specifically refer to the singular, but may also include the plural. The terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of further restrictions, the elements defined by the sentence "include one..." do not exclude the presence of other identical elements in the process, method or device including the elements.
[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.
Claims
1. A cross-domain resource scheduling method, It is characterized in that Applied to a cross-domain resource scheduling platform, the cross-domain resource scheduling platform includes a resource request interface, a resource scheduling system and a resource perception system, and the method includes: The resource scheduling system, in response to detecting the resource call request sent by the resource request interface, obtains type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through the resource perception system; The resource scheduling system determines a valid resource from all candidate resources based on the type attribute information in the resource call request and the type attribute information of each candidate resource, wherein the type attribute information describes the values of all type attribute items of the resource, and the resource scheduling system selects the candidate resource whose type attribute information satisfies the type attribute information in the resource call request from all candidate resources as the valid resource; The resource scheduling system determines the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weight and user weight corresponding to each resource quantity attribute item, and determines the target resource based on all resource quantity scores, wherein the resource quantity attribute information describes the values of all resource quantity attribute items of the resource, the preset weight is used to weight the total amount of the resource quantity attribute item, and the user weight is used to weight the remaining amount of the resource quantity attribute item; The resource request interface sends a resource start request to the computing center corresponding to the target resource, so that the computing center starts the target resource and links to the user end.
2. The cross-domain resource scheduling method according to claim 1, It is characterized in that The resource scheduling system determines a valid resource from all candidate resources based on the type attribute information in the resource call request and the type attribute information of each candidate resource, including: The resource scheduling system constructs a type compliance vector for each candidate resource based on the type attribute information in the resource call request and the type attribute information of each candidate resource, wherein each element in the type compliance vector respectively indicates whether the platform type, computing architecture, intelligent computing architecture, storage architecture, operating system, and software version of the candidate resource meet the resource call request; Based on the type compliance vector of each candidate resource, a valid resource is determined among all candidate resources.
3. The cross-domain resource scheduling method according to claim 1, It is characterized in that The resource scheduling system determines the resource quantity score of each effective resource according to the resource quantity attribute information of each effective resource, and the preset weight and user weight corresponding to each resource quantity attribute item, including: For each of the valid resources, the resource scheduling system performs weighted calculation on the total amount and the remaining amount of each resource amount attribute item in the resource amount attribute information according to the preset weight and the user weight corresponding to the resource amount attribute item, to obtain a score for the resource amount attribute item; The resource scheduling system adds up the scores of all resource quantity attribute items to determine the resource quantity score of the effective resource.
4. The cross-domain resource scheduling method according to claim 1, It is characterized in that Before the resource scheduling system determines the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource and the preset weight and user weight corresponding to each resource quantity attribute item, the system further includes: The resource scheduling system determines whether all valid resources are from the same computing center. If not, the computing center corresponding to each valid resource is determined as the computing center to be determined. The resource scheduling system obtains connection attribute information of each to-be-judged computing power center through the resource perception system, and determines the target computing power center based on the connection attribute information of each to-be-judged computing power center; If there are multiple valid resources in the target computing power center, the resource scheduling system will execute the steps of determining the resource quantity score of each valid resource in the target computing power center based on the resource quantity attribute information of each valid resource, and the preset weights and user weights corresponding to each resource quantity attribute item.
5. The cross-domain resource scheduling method according to claim 4, It is characterized in that The connection attribute information includes connectivity value, network delay, network bandwidth, network stability and historical resource average score. Based on the connection attribute information of each computing power center to be judged, the target computing power center is determined, including: Eliminate the computing power centers whose connectivity values are represented as disconnected, and for each of the remaining computing power centers to be judged, determine the central connection score of the computing power center to be judged based on the network delay, network bandwidth, network stability and historical resource average score of the computing power center to be judged; The computing power center to be judged with the largest center connection score is determined as the target computing power center.
6. The cross-domain resource scheduling method according to claim 1, It is characterized in that The resource request interface sends a resource start request to the computing center corresponding to the target resource, so that the computing center starts the target resource and links to the user end, including: The resource request interface sends a resource start request to the computing center corresponding to the target resource, and obtains resource link usage information fed back by the computing center; The resource request interface pushes the resource link usage information to the user end, obtains the resource link request fed back by the user end, and links the target resource to the user end in response to the resource link request.
7. The cross-domain resource scheduling method according to claim 1, It is characterized in that The cross-domain resource scheduling platform also includes a cloud storage module and a data transmission interface, and the method also includes: In response to detecting a data transmission request, the data transmission interface calls the unified resource storage API to write the data to be transmitted into the corresponding resource, and creates stream data according to the data to be transmitted; The data transmission interface calls an asynchronous task service to create a data upload task, obtains the data to be transmitted from the stream data through the asynchronous task service, and calls a cloud storage API to write the data to be transmitted into the cloud storage module.
8. The cross-domain resource scheduling method according to claim 7, It is characterized in that The method further comprises: The resource scheduling system, in response to detecting the to-be-executed task of the target resource, determines whether the task processing data corresponding to the to-be-executed task is cross-domain data, wherein the cross-domain data is data processed or obtained by resources of other computing centers different from the computing center of the target resource; If so, the resource scheduling system obtains the task processing data from the cloud storage module and sends the task processing data to the target resource.
9. The cross-domain resource scheduling method according to claim 8, It is characterized in that After sending the task processing data to the target resource, the method further includes: The resource scheduling system determines whether the task to be executed is a task that has been executed by resources of other computing centers. If so, during the execution of the task to be executed, the actual result data of the task to be executed is obtained; The resource scheduling system obtains the historical result data of other computing centers from the cloud storage module or the other computing centers, and determines whether the actual result data and the historical result data meet the consistency condition, wherein determining whether the actual result data and the historical result data meet the consistency condition specifically includes: calculating the similarity between the two, and if the similarity is greater than a preset threshold, determining that the consistency condition is met; If not, the resource scheduling system displays the coverage request to the user terminal through the resource request interface, and obtains the coverage confirmation instruction fed back by the user terminal; In response to the overwrite confirmation instruction, the resource scheduling system uploads the actual result data to the cloud storage module and overwrites the historical result data.
10. A cross-domain resource scheduling platform, It is characterized in that The cross-domain resource scheduling platform includes a resource request interface, a resource scheduling system and a resource perception system, wherein: The resource perception system is used to obtain type attribute information and resource quantity attribute information of resources in each computing power center; The resource request interface is used to obtain the resource call request sent by the user end, and send the resource call request to the resource scheduling system, obtain the target resource fed back by the resource scheduling system, and send a resource start request to the computing center corresponding to the target resource, so that the computing center starts the target resource and links to the user end; The resource scheduling system is used to respond to the resource call request sent by the resource request interface, obtain the type attribute information and resource quantity attribute information of each candidate resource corresponding to the user through the resource perception system, determine the valid resource from all the candidate resources based on the type attribute information in the resource call request and the type attribute information of each candidate resource, determine the resource quantity score of each valid resource according to the resource quantity attribute information of each valid resource, and the preset weight and user weight corresponding to each resource quantity attribute item, and determine the target resource based on all resource quantity scores, and feed the target resource back to the resource request interface, wherein the type attribute information describes the values of all type attribute items of the resource, and the resource scheduling system screens out the candidate resource whose type attribute information meets the type attribute information in the resource call request from all the candidate resources as the valid resource; the resource quantity attribute information describes the values of all resource quantity attribute items of the resource, the preset weight is used to weight the total amount of the resource quantity attribute item, and the user weight is used to weight the remaining amount of the resource quantity attribute item.
