Resource recommendation method and device based on multi-cloud environment, equipment and storage medium

Abstract

The invention discloses a resource recommendation method and device based on a multi-cloud environment, equipment and a storage medium, relates to the technical field of resource management, and discloses a resource recommendation method based on the multi-cloud environment, and the method comprises the steps: obtaining resource information of a resource demand and a resource instance, and carrying out the feature extraction according to the resource information, and obtaining a resource feature vector; generating a target demand vector according to the resource demand and the resource feature vector; performing vector similarity retrieval according to the target demand vector to obtain a target candidate resource set; performing score calculation according to candidate resources in the target candidate resource set to obtain candidate resource scores; and sorting the candidate resources according to the candidate resource scores to generate a recommendation result list. The technical effects of accurately matching the user demand and the resource instance in the multi-cloud environment, efficiently recommending the resource meeting the user demand, and improving the resource utilization efficiency and the user experience are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of resource management, in particular to a resource recommendation method and device based on a multi-cloud environment, equipment and a storage medium. BACKGROUND

[0002] With the continuous development of cloud computing technology, major cloud service providers provide a variety of resource instances, including computing resources, storage resources and network resources. When deploying a business, users need to select the most suitable configuration combination from these heterogeneous cloud resources, taking into account not only basic capabilities such as computing performance, storage capacity, network bandwidth, but also service level, stability, cost structure and scalability. In the selection process, due to the different resource specifications, performance indicators and billing methods of each cloud service provider, the decision-making process faced by users becomes increasingly complex and difficult to control, especially in a multi-cloud deployment environment, resource selection not only involves multi-dimensional evaluation, but also requires efficient response.

[0003] Traditional resource recommendation methods mainly rely on rule screening or fixed scoring models. Rule screening methods filter resources through pre-set hard conditions, such as setting a minimum memory capacity or a specific number of CPU cores. The recommended results only meet the explicit conditions and lack a deep understanding of the demand details. Another method uses fixed weights to score resources and sorts them according to the total score of each indicator. This method can reflect the importance of indicators to some extent, but due to the fixed scoring weights, it is difficult to adapt to the diversification of user preferences and changes in scenarios.

[0004] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0005] The main purpose of the present application is to provide a resource recommendation method, device, equipment and storage medium based on a multi-cloud environment, aiming to solve the technical problem of difficult to balance the response speed and matching accuracy of resource recommendation in a multi-cloud environment.

[0006] To achieve the above purpose, the present application provides a resource recommendation method based on a multi-cloud environment, the method comprising: obtaining resource requirements and resource information of resource instances, and performing feature extraction according to the resource information to obtain a resource feature vector; generating a target demand vector according to the resource requirements and the resource feature vector; performing vector similarity retrieval according to the target demand vector to obtain a target candidate resource set; performing scoring calculation according to the candidate resources in the target candidate resource set to obtain candidate resource scores; The candidate resources are ranked according to the candidate resource scores, and a recommended result list is generated.

[0007] In an embodiment, the step of calculating scores according to the candidate resources in the target candidate resource set to obtain candidate resource scores comprises: obtaining a performance matching degree weight, a cost benefit weight, an availability weight, and an expansibility weight; calculating a performance matching degree score, a cost benefit score, an availability score, and an expansibility score of the candidate resources in the target candidate resource set according to the performance matching degree weight, the cost benefit weight, the availability weight, and the expansibility weight, respectively; calculating scores according to the performance matching degree score, the cost benefit score, the availability score, and the expansibility score to obtain candidate resource scores.

[0008] In an embodiment, the step of calculating a performance matching degree score, a cost benefit score, an availability score, and an expansibility score of the candidate resources in the target candidate resource set according to the performance matching degree weight, the cost benefit weight, the availability weight, and the expansibility weight comprises: determining a central processor weight, a memory weight, a storage weight, and a network weight according to the performance matching degree weight, and calculating the performance matching degree score of the candidate resources in the target candidate resource set according to the central processor weight, the memory weight, the storage weight, and the network weight; determining a budget weight and a cost performance weight according to the cost benefit weight, and calculating the cost benefit score of the candidate resources according to the budget weight and the cost performance weight; determining a reliability weight and a stability weight according to the availability weight, and calculating the availability score of the candidate resources according to the reliability weight and the stability weight; determining a vertical expansion weight and a horizontal expansion weight according to the expansibility weight, and calculating the expansibility score of the candidate resources according to the vertical expansion weight and the horizontal expansion weight.

[0009] In an embodiment, after the step of ranking the candidate resources according to the candidate resource scores to generate a recommended result list, the method further comprises: obtaining feedback information of the candidate resources in the recommended result list, the feedback information comprising selection behavior information and satisfaction feedback; analyzing deviation information of the feedback information and the candidate resource scores according to the selection behavior information and the satisfaction feedback; updating weight values of the performance matching degree weight, the cost benefit weight, the availability weight, and the expansibility weight according to the deviation information.

[0010] In an embodiment, the step of performing vector similarity retrieval according to the target demand vector to obtain a target candidate resource set comprises: obtaining a similarity threshold, a preset vector database, and a preset constraint condition; performing approximate nearest neighbor retrieval in the preset vector database according to the target demand vector to obtain an initial candidate resource set; selecting a similar candidate resource set with a similarity greater than the similarity threshold from the initial candidate resource set; regarding a resource instance in the similar candidate resource set satisfying the preset constraint condition as a target candidate resource, and determining a target candidate resource set according to the target candidate resource.

[0011] In an embodiment, the step of generating a target demand vector according to the resource demand and the resource feature vector comprises: obtaining computing capability demand information, storage demand information, network bandwidth demand information, budget limit demand information, and a business scenario type according to the resource demand; determining a preset demand mapping rule according to the business scenario type; performing standardization processing on the capability demand information, the storage demand information, the network bandwidth demand information, and the budget limit demand information according to the preset demand mapping rule to obtain a standardized demand parameter mapping; performing vector space isomorphism according to the standardized demand parameter mapping and the resource feature vector to obtain an isomorphic demand vector; performing normalization processing on the isomorphic demand vector to obtain a target demand vector.

[0012] In an embodiment, the step of performing feature extraction according to the resource information to obtain a resource feature vector comprises: obtaining computing capability information, memory information, storage information, network information, and a preset norm of the resource instance in the resource information; extracting basic features, cost features, availability features, and scalability features according to the computing capability information, the memory information, the storage information, and the network information; performing normalization processing on the basic features, the cost features, the availability features, and the scalability features to obtain a normalized resource vector; performing feature combination on the normalized resource vector to obtain a resource feature vector of a preset dimension, and performing normalization processing on the resource feature vector according to the preset norm to obtain a resource feature vector.

[0013] In addition, to achieve the above object, the application further provides a resource recommendation device based on a multi-cloud environment, which comprises: a resource acquisition module, configured to acquire resource demand and resource information of a resource instance, and perform feature extraction according to the resource information to obtain a resource feature vector; a demand generation module, configured to generate a target demand vector according to the resource demand and the resource feature vector; a resource candidate module, configured to perform vector similarity retrieval according to the target demand vector to obtain a target candidate resource set; a score calculation module, configured to perform score calculation according to candidate resources in the target candidate resource set to obtain candidate resource scores; a resource recommendation module, configured to sort the candidate resources according to the candidate resource scores to generate a recommendation result list.

[0014] In addition, to achieve the above object, the application further provides a resource recommendation device based on a multi-cloud environment, which comprises: a memory, a processor and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the resource recommendation method based on a multi-cloud environment as described above.

[0015] In addition, to achieve the above object, the application further provides a storage medium, which is a computer readable storage medium, and a computer program is stored on the storage medium, and the computer program is executed by a processor to implement the steps of the resource recommendation method based on a multi-cloud environment as described above.

[0016] In addition, to achieve the above object, the application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the resource recommendation method based on a multi-cloud environment as described above.

[0017] The one or more technical solutions provided by the application have at least the following technical effects: The technical means comprises: converting resource information of resource instances into resource feature vectors through feature extraction, combining resource requirements to generate target requirement vectors, and then screening a target candidate resource set through vector similarity retrieval, and generating a recommended result list through score calculation and sorting; in a multi-cloud environment, resource types are various, and demand and resource matching dimensions are multiple and complex; due to the lack of standardized feature representation and accurate matching mechanism, the existing technology often leads to inaccurate resource and demand matching, low recommendation efficiency and results that do not meet actual demand; by converting resource information and demand into vectors to achieve standardized representation, using vector similarity retrieval to achieve efficient screening, and combining score sorting to optimize the results, the problems of difficult resource and demand matching, low recommendation efficiency and inaccurate results in a multi-cloud environment are solved; compared with the prior art, the accuracy and efficiency of resource recommendation are improved, and a recommended result list that better meets resource demand can be generated. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings incorporated in and forming a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings required to be used in the embodiments or prior art description will be briefly introduced. Obviously, those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0020] Figure 1 A flowchart provided by the resource recommendation method based on a multi-cloud environment according to Embodiment 1 of the present application; Figure 2 A flowchart provided by the resource recommendation method based on a multi-cloud environment according to Embodiment 2 of the present application; Figure 3 A module structure diagram of the resource recommendation device based on a multi-cloud environment according to the embodiments of the present application; Figure 4 A device structure diagram of a hardware operating environment involved in the resource recommendation method based on a multi-cloud environment according to the embodiments of the present application.

[0021] The object implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0022] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application, and are not used to limit the present application.

[0023] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings in the specification and specific embodiments.

[0024] The main solution of the embodiment of the application is: obtaining resource requirements and resource information of resource instances, and performing feature extraction according to the resource information to obtain a resource feature vector; generating a target demand vector according to the resource requirements and the resource feature vector; performing vector similarity retrieval according to the target demand vector to obtain a target candidate resource set; performing score calculation according to candidate resources in the target candidate resource set to obtain candidate resource scores; and sorting the candidate resources according to the candidate resource scores to generate a recommendation result list.

[0025] In the embodiment, for the convenience of description, the following describes an identification device for recommending resources based on a multi-cloud environment as an execution subject.

[0026] Since the existing technology cannot balance the response speed and matching accuracy of resource recommendation under a multi-cloud environment, the application provides a solution that converts resource information of resource instances into a resource feature vector through feature extraction, generates a target demand vector in combination with resource requirements, and then filters a target candidate resource set through vector similarity retrieval, and generates a recommendation result list through score calculation and sorting; in a multi-cloud environment, there are various types of resources, multiple and complex matching dimensions of requirements and resources, and the existing technology often lacks standardized feature representation and accurate matching mechanisms, resulting in inaccurate matching of resources and requirements, low recommendation efficiency, and results that do not meet actual requirements. By converting resource information and requirements into vectors to achieve standardized representation, efficiently filtering using vector similarity retrieval, and optimizing results through score sorting, the application solves the problems of difficult matching of resources and requirements, low recommendation efficiency, and inaccurate results under a multi-cloud environment; compared with the existing technology, the application improves the accuracy and efficiency of resource recommendation and can generate a recommendation result list that better meets resource requirements.

[0027] It should be noted that the execution subject of the embodiment can be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, or an electronic device capable of achieving the above functions, a resource recommendation device based on a multi-cloud environment, and the like. The following describes the embodiment and each of the following embodiments by taking a resource recommendation device based on a multi-cloud environment as an example.

[0028] Based on this, the embodiment of the application provides a resource recommendation method based on a multi-cloud environment, which is described below with reference to Figure 1 , Figure 1 The following is a flowchart of a first embodiment of the resource recommendation method based on a multi-cloud environment of the application.

[0029] In the embodiment, the resource recommendation method based on a multi-cloud environment includes steps S10-S50: In step S10, the resource demand and resource information of the resource instance are obtained, and feature extraction is performed according to the resource information to obtain a resource feature vector. It should be noted that the resource demand refers to the specific requirements of a user or an application program for the required resources in a multi-cloud environment. These requirements usually cover multiple dimensions, such as computing power requirements such as CPU core number, processor model, storage requirements such as storage capacity, storage type such as solid-state storage or mechanical storage, network requirements such as bandwidth size, delay upper limit, cost budget such as hourly or monthly expenditure upper limit, and reliability requirements such as service availability index, data backup frequency, etc. In this embodiment, the resource demand can be a literal description explicitly proposed by the user, or a parameterized requirement automatically generated by the system according to the application scenario.

[0030] In addition, the resource instance refers to a specific resource entity that actually exists, can be called or allocated in a multi-cloud environment. These entities come from different cloud service providers, such as Amazon Web Services (AWS), Microsoft Azure, Alibaba Cloud, etc. The types of resource instances include but are not limited to virtual server instances, cloud database instances such as relational database instances, non-relational database instances, object storage instances, container service instances, etc. Each resource instance has a unique identifier and specific running state.

[0031] In addition, the resource information is detailed data describing the properties and characteristics of the resource instance, which is used to represent the performance, configuration, state, etc. of the resource instance. Specifically, it includes the hardware configuration of the resource instance, such as CPU model and core number, memory capacity, disk I / O rate, software environment such as operating system type and version, pre-installed applications, service properties such as pay-as-you-go or annual or monthly package, availability percentage specified in the service level agreement, geographic location such as data center location, and historical performance data such as average response time in the past 7 days, resource usage peak, etc.

[0032] In addition, feature extraction refers to the process of screening, converting and extracting representative key features from resource information, which aims to convert unstructured or semi-structured resource information into structured feature data that can be used for calculation and comparison. In this process, the resource information needs to be preprocessed, such as converting the text description of the operating system type into numerical coding, standardizing continuous data, while eliminating redundant or irrelevant information such as vendor logo information unrelated to resource performance, and retaining features meaningful for resource matching.

[0033] Further, the resource feature vector is a multi-dimensional vector obtained after feature extraction and used to quantitatively represent the features of the resource instance, each dimension of which corresponds to a processed resource feature. For example, the resource feature vector of a virtual server instance can include the following dimensions: CPU core number, memory capacity, bandwidth size, price, etc., and the numerical values of each dimension together form a vector, so that the features of the resource instance can be represented and compared mathematically.

[0034] In a feasible implementation, in the initial stage of the system or when facing new users, the cold start problem caused by insufficient feedback data can be solved by initializing the weights (such as the weights of the performance, cost, and other dimensions) based on expert knowledge, providing a basis for recommendation; at the same time, a lightweight and rapid feedback mechanism is designed to collect the preference information of users on key factors (such as the emphasis on performance or cost) in the recommendation process; based on these rapid feedbacks, incremental weight adjustment is performed to gradually optimize the weight settings to fit the user needs; in this stage, a certain exploration rate needs to be maintained, and diversified resources are recommended to cover potential needs, and the exploration rate gradually decreases with the accumulation of user feedback data, realizing a smooth transition from dependence on initial rules to actual user preferences, and effectively solving the cold start problem.

[0035] In a feasible implementation, step S10 can include steps S11-S14: Step S11, obtaining the computing capability information, memory information, storage information, network information, and preset norm of the resource instance in the resource information; It should be noted that the computing capability information is information describing the computing performance of the resource instance in processing data and running tasks, mainly including the number of cores, the frequency in GHz, and the processor architecture of the Central Processing Unit (CPU), such as x86, ARM, floating-point operation capability (FLOPS), etc. It directly determines the speed and efficiency of the resource instance in processing complex tasks, for example, an 8-core CPU with a frequency of 3.2 GHz usually has higher computing capability than a 4-core CPU with a frequency of 2.5 GHz.

[0036] The memory information is the attribute information of the Random Access Memory (RAM) of the resource instance, including the memory capacity in GB or TB, the memory type such as Double Data Rate 4 (DDR4) or DDR5, the memory bandwidth in GB / s, and the memory delay in ns, etc. The memory information affects the ability of the resource instance to temporarily store and quickly access data, for example, an instance with 16 GB DDR5 memory is usually more fluent than an instance with 8 GB DDR4 memory when processing multiple tasks.

[0037] The storage information is characteristic information of a persistent storage device describing a resource instance, including storage capacity in GB or TB, storage type such as Solid State Drive (SSD), Hard Disk Drive (HDD), read-write speed such as Input / Output Operations Per Second (IOPS), continuous read-write rate in MB / s, and storage interface such as Serial Advanced Technology Attachment (SATA), Non-Volatile Memory Express (NVMe), etc. For example, in the storage information of a 1 TB NVMe SSD, the read-write speed can reach 3000 MB / s, which is much higher than that of a SATA HDD with the same capacity.

[0038] The network information is characteristic information of a resource instance in terms of network communication, including network bandwidth in Mbps or Gbps, network delay in ms, which refers to the time from data sending to receiving, network throughput in GB / s, which refers to the amount of data transmitted per unit time, supported network protocols such as Transmission Control Protocol / Internet Protocol (TCP / IP), User Datagram Protocol (UDP), and whether to support network functions such as load balancing, Virtual Private Network (VPN), etc. For example, a resource instance with network information of "10 Gbps bandwidth, delay ≤ 20 ms" usually has higher data transmission speed and stability than an instance with network information of "1 Gbps bandwidth, delay ≤ 50 ms".

[0039] The preset norm is a mathematical function preset for measuring the size of a vector, which is used in the embodiment to normalize the resource feature vector subsequently, to ensure the uniformity of the scale of the vector. Common preset norms include L1 norm, sum of absolute values of vector elements, L2 norm, square root of the sum of squares of vector elements, etc. For example, if the preset norm is L2 norm, the resource feature vector will be converted into a unit vector subsequently by using the norm, and the L2 norm of the vector is 1.

[0040] In step S12, the base feature, the cost feature, the availability feature, and the scalability feature are extracted according to the computing power information, the memory information, the storage information, and the network information. It should be noted that the basic features are features reflecting the basic running capability of the resource instance extracted from the computing capability information, memory information, storage information and network information, and are directly related to the core performance of the resource. For example, the number of CPU cores, the main frequency extracted from the computing capability information; the memory capacity extracted from the memory information; the storage capacity, read and write speed extracted from the storage information; the bandwidth, delay extracted from the network information, etc. belong to the basic features, which are the basic guarantee for the normal operation of the resource instance.

[0041] The cost feature is a feature related to the use cost of the resource instance, which is extracted from the content related to charging in the resource information, including hourly rent, storage fee, such as monthly fee per GB, network traffic fee, such as data transmission fee per GB, whether to support pay-as-you-go or long-term package, such as discount information of annual or monthly package, etc. For example, the cost feature of a certain resource instance can be "2 yuan per hour, annual discount 8 times, and traffic fee 0.5 yuan per GB".

[0042] The availability feature is a feature describing the reliable running capability of the resource instance, which is extracted from the content related to stability and fault recovery in the resource information, including the availability percentage in the service level agreement, Service Level Agreement, SLA, such as 99.9% indicating that the allowed unavailable time per month is not more than 43 minutes, the fault automatic recovery time, such as 30 minutes, the data backup frequency, such as daily backup, whether to support multi-region redundant deployment, etc. For example, the availability feature of a certain resource instance is "availability 99.95%, fault recovery time ≤ 15 minutes, daily backup".

[0043] The scalability feature is a feature reflecting the capability of the resource instance to dynamically adjust performance according to demand, which is extracted from the content related to configuration adjustment in the resource information, including the maximum scalable core number of CPU, such as the current 4 cores scalable to 16 cores, the maximum scalable capacity of memory, such as the current 32 GB scalable to 128 GB, the maximum scalable capacity of storage, the time required for expansion, such as 5 minutes for online expansion, whether to support automatic expansion and contraction, automatically adjusting configuration according to load, etc. For example, the scalability feature of a certain resource instance is "CPU scalable to 32 cores, memory scalable to 256 GB, support automatic expansion and contraction, expansion time ≤ 3 minutes".

[0044] It can be understood that the basic features are extracted from the computing capability information, memory information, storage information and network information, such as CPU core number, memory capacity, storage read and write speed, bandwidth, etc. Then the cost feature is extracted from the charging content of the resource information, including rent, discount, etc. The availability feature is extracted from the stability and fault handling information of the resource, such as availability percentage, backup frequency, etc. The scalability feature is extracted from the configuration adjustment information of the resource, such as maximum expansion capacity, expansion time, etc.

[0045] Step S13, according to the basic features, the cost features, the availability features and the scalability features, normalization processing is performed to obtain a normalized resource vector; It should be noted that the normalized resource vector is a vector formed by combining the normalized basic features, cost features, availability features and scalability features in a predetermined order. Each dimension corresponds to a normalized feature, for example, the number of CPU cores in the basic features is normalized to 0.3, the memory capacity is normalized to 0.5, and the hourly rental cost in the cost features is normalized to 0.2, etc., which together form a multi-dimensional vector.

[0046] In addition, normalization processing is a process of converting features of different magnitudes and units to a unified scale, such as 0-1 or -1 to 1, with the purpose of eliminating the influence of magnitude differences between features on subsequent calculations.

[0047] It can be understood that for each sub-feature in the basic features, normalization is performed according to its value range; then the same normalization processing is performed on each sub-feature in the cost features, availability features and scalability features; finally, all normalized sub-features are combined in a predetermined order to form a normalized resource vector. In this embodiment, the Min-Max normalization method can be used to map the feature values to the [0, 1] interval.

[0048] Step S14, performing feature combination on the normalized resource vector to obtain a resource feature vector of a predetermined dimension, and performing normalization processing on the resource feature vector according to the predetermined norm to obtain a resource feature vector.

[0049] It should be noted that the predetermined dimension is the dimension number of the resource feature vector that is predetermined in advance, which is used to unify the structure of the vector for subsequent vector similarity calculation and model processing. For example, if the predetermined dimension is 30, then after feature combination, the resource feature vector needs to include 30 feature dimensions. If the dimension of the original normalized resource vector is insufficient, it is supplemented through feature combination; if it exceeds, the key features are retained.

[0050] Further, the resource feature vector is the final vector obtained after feature combination and predetermined norm normalization processing, which has a predetermined dimension and is normalized by a predetermined norm, such as L2 norm, so that the scale of the vector is unified, such as L2 norm being 1, which can be directly used for subsequent vector similarity retrieval and other processing.

[0051] It can be understood that the features in the normalized resource vector are combined to generate new cross features such as "bandwidth to delay ratio", "storage capacity to cost ratio", etc., so that the feature dimension reaches a preset dimension such as 30 dimensions; then the preset norm determined in step S11 such as L2 norm is used to normalize the vector, and the vector is converted into a unit vector with a norm of 1; and finally a resource feature vector available for subsequent processing is obtained.

[0052] In this embodiment, the original dimension of the normalized resource vector is 15, 15 new cross features are generated by feature combination to make the total dimension reach the preset 30 dimensions; and then the L2 norm is used to normalize the 30-dimensional vector, each element is divided by the L2 norm of the vector, and finally the resource feature vector is obtained. The normalized features are combined into a vector, for high-dimensional features, principal component analysis (PCA) method can be used for dimension reduction, and finally a 128-dimensional feature vector is generated. The vector structure is: 0-31 dimensions are computing performance features; 32-63 dimensions are storage performance features; 64-95 dimensions are network performance features; and 96-127 dimensions are cost and availability features. The normalization calculation formula is as follows: = V / ||V|| In the formula, V is the resource feature vector, i.e. the original vector, ||V|| is the preset norm of the vector, i.e. the L2 norm, is the resource feature vector.

[0053] Step S20, generating a target demand vector according to the resource demand and the resource feature vector; It should be noted that the target demand vector is a multi-dimensional vector generated by combining the resource demand and the resource feature vector, which is used to quantitatively represent the user demand, and the dimension is completely consistent with that of the resource feature vector, so as to perform subsequent vector similarity calculation. The vector is formed by converting each requirement in the resource demand into a feature value corresponding to the resource feature vector, for example, "4-core CPU" in the resource demand corresponds to the dimension of CPU core number in the resource feature vector, and "monthly rent not more than 500 yuan" corresponds to the feature value of the price dimension. For the feature dimension not mentioned in the resource demand, such as the operating system type, the common value of the dimension in the resource feature vector or the default value such as 0 indicating no specific requirement can be filled to ensure the integrity of the vector dimension.

[0054] It can be understood that the resource demand is parsed to determine the specific requirements contained therein, such as computing, storage, price, and the like; referring to the dimension structure of the resource feature vector, each requirement in the resource demand is converted into a corresponding feature value, and for a feature dimension not explicitly stated in the resource demand, a reasonable default value is set according to the actual scene, such as if the resource demand does not mention the operating system, the dimension can be set to 0, indicating no limitation on the operating system; the feature values are combined to form a target demand vector, ensuring that the dimensions are completely consistent with the resource feature vector.

[0055] In a feasible implementation, step S20 can include steps S21-S25: Step S21, obtaining computing capability demand information, storage demand information, network bandwidth demand information, budget limit demand information, and service scenario type according to the resource demand; It should be noted that the resource demand is a specific appeal of a user or an application for the required resources in a multi-cloud environment, covering multiple aspects such as function, performance, and cost, and is the basis for subsequent generation of a target demand vector.

[0056] The computing capability demand information is the computing performance requirement in the resource demand for processing data and running tasks, including the minimum or target value of the number of CPU cores, such as at least 8 cores, the requirement for CPU main frequency, such as no less than 3.0 GHz, whether GPU acceleration is needed, such as GPU support is needed for deep learning scenarios, and the concurrent processing capability requirement of the computing task, such as supporting 1000 task scheduling per second.

[0057] The storage demand information is the requirement in the resource demand for data storage, including the size of the storage capacity, such as at least 1 TB, the preference for the type of storage, such as requiring SSD to ensure high-speed reading and writing, the upper limit of data access delay, such as read delay not exceeding 5 ms, and whether distributed storage is needed, such as a distributed file system for big data processing.

[0058] The network bandwidth demand information is the requirement in the resource demand for data transmission capability, including the minimum bandwidth value, such as at least 10 Gbps, the upper limit of network delay, such as round-trip delay not exceeding 30 ms, the stability requirement of data transmission, such as packet loss rate not exceeding 0.1%, and whether to support a specific network protocol, such as UDP for real-time communication.

[0059] The budget limit requirement information is a constraint on the use cost in the resource requirement, including a maximum expenditure limit per unit of time, such as 10,000 yuan per month, whether to accept a prepayment discount, such as accepting a package-year payment, a tolerance range of cost fluctuations, such as a price fluctuation of no more than 10% in a short period of time, and whether to prefer a cost-effective resource, such as selecting a low-price resource on the premise of meeting performance.

[0060] The business scenario type refers to a specific application scenario corresponding to the resource requirement. Different scenarios have different focuses on resources. Common types include an e-commerce promotion scenario, which requires high concurrency and high bandwidth, a financial transaction scenario, which requires high security and high availability, a big data analysis scenario, which requires large storage and strong computing power, a video live streaming scenario, which requires low latency and high bandwidth, and the like.

[0061] In step S22, a preset requirement mapping rule is determined according to the business scenario type. It should be noted that the preset requirement mapping rule is a rule set prepared in advance for different business scenario types, which is used to convert unstructured or semi-structured requirement information into standardized parameters. The rule set defines the correspondence between requirement indicators and standard parameters, weight allocation, and conversion methods according to the characteristics of the business scenario.

[0062] It can be understood that, according to the business scenario type, a corresponding rule is matched from a preset rule library. Then, it is verified whether the rule is applicable to the specific circumstances of the current resource requirement, such as whether it covers all indicators in the requirement. The preset requirement mapping rule corresponding to the business scenario type is determined as the basis for subsequent standardization processing.

[0063] In this embodiment, if the business scenario type is an e-commerce promotion scenario, the preset requirement mapping rule for the e-commerce promotion scenario is called from the rule library. The rule includes specific mapping relationships, such as a concurrency of ≥100,000 times / second corresponding to a CPU of ≥8 cores, and a peak traffic of ≥100 Mbps corresponding to a bandwidth of ≥20 Gbps.

[0064] The scenario-based rule basis can be provided for the standardization processing of the requirement information, ensuring that the converted parameters meet the actual requirements of the business scenario and improving the accuracy of requirement conversion.

[0065] In step S23, the capability requirement information, the storage requirement information, the network bandwidth requirement information, and the budget limit requirement information are standardized according to the preset requirement mapping rule, and a standardized requirement parameter mapping is obtained. It should be noted that the standardized demand parameter mapping is a standardized parameter set obtained by converting the computing power demand information, storage demand information, network bandwidth demand information, budget limit demand information through a preset demand mapping rule. It is presented in the form of key-value pairs, each key corresponds to a demand index, and the value is the standard parameter value of the index, and the parameter format matches the feature dimension of the resource feature vector.

[0066] It can be understood that the preset demand mapping rule is called, and the computing power demand information is converted into standardized computing parameters according to the rule, such as converting strong computing power into CPU core number≥8 cores and main frequency≥3.0GHz; then the storage demand information is converted into standardized storage parameters, such as converting high-speed storage into storage type=SSD and capacity≥1TB; then the network bandwidth demand information is converted into standardized network parameters, such as converting low delay and high bandwidth into bandwidth≥20Gbps and delay≤30ms”; then the budget limit demand information is converted into standardized cost parameters, such as converting budget limit into monthly expenditure≤8000 yuan; and these standardized parameters are integrated to form a standardized demand parameter mapping.

[0067] In step S24, vector space isomorphism is performed according to the standardized demand parameter mapping and the resource feature vector, and an isomorphic demand vector is obtained. It should be noted that vector space isomorphism refers to adjusting the structure and dimension of the standardized demand parameter mapping, so that the demand parameter corresponding to the vector and the resource feature vector are in the same vector space, that is, the number of dimensions is the same, and the meaning and order of each dimension are consistent, ensuring that the two can be mathematically operated, such as similarity calculation.

[0068] Further, the isomorphic demand vector is a demand vector obtained after vector space isomorphism processing, which is completely consistent with the dimension structure of the resource feature vector. Each dimension corresponds to the demand parameter of the same dimension in the resource feature vector, and the dimension not explicitly mentioned can be set to a default value according to the business scenario type, such as 0 or the average level if there is no special requirement.

[0069] It can be understood that the dimension structure of the resource feature vector is analyzed to determine the feature dimensions and order contained therein, such as CPU core number, memory capacity, storage capacity, bandwidth, price, availability, etc.; the standardized demand parameter mapping is adjusted according to the structure, and the dimensions not explicitly mentioned in the demand but contained in the resource feature vector are supplemented, such as the availability dimension in the resource feature vector, which is filled with the default value of the business scenario if not mentioned in the demand, which can be set to 99.9% in the e-commerce scenario; then the parameters in the standardized demand parameter mapping are arranged according to the dimension order of the resource feature vector to form a vector form; and finally an isomorphic demand vector isomorphic to the resource feature vector is obtained.

[0070] Step S25, normalize the isomorphic demand vector to obtain a target demand vector.

[0071] It can be understood that the same normalization method as the resource feature vector is adopted, such as min-max normalization, to determine the value range of each dimension parameter in the isomorphic demand vector, such as the CPU core number range 2-64 cores and the price range 1000-20000 yuan / month; the normalization calculation is performed on each dimension parameter of the isomorphic demand vector to convert it into a value between 0 and 1; and the normalized parameters are combined to obtain the target demand vector. The extracted demand parameters are mapped to a 128-dimensional demand vector through the same processing flow as the resource features. For parameters not explicitly specified by the user, default values or inferred values based on historical data are used to fill in.

[0072] Step S30, vector similarity retrieval is performed according to the target demand vector to obtain a target candidate resource set; It should be noted that vector similarity retrieval refers to the process of filtering out resource instances corresponding to resource feature vectors with higher similarity by calculating the similarity between the target demand vector and each resource feature vector. Common similarity calculation methods include cosine similarity and Euclidean distance, wherein the cosine similarity is used to measure the similarity in direction between two vectors, and the value closer to 1 indicates a higher similarity; the Euclidean distance is used to measure the straight-line distance between two vectors in space, and the smaller the value, the higher the similarity.

[0073] It can be understood that the target candidate resource set is a set of resource instances with higher similarity to the target demand vector obtained through vector similarity retrieval, and the resource instances in the set are more matched with the user's resource demand in terms of features. For example, if it is found through similarity retrieval that the cosine similarity between the resource feature vectors of 10 resource instances and the target demand vector is greater than 0.8, then the 10 resource instances collectively constitute the target candidate resource set, which is the basis for further scoring and sorting.

[0074] In a feasible implementation, step S30 can include steps S31-S34: Step S31, obtain a similarity threshold, a preset vector database, and a preset constraint condition; It should be noted that the similarity threshold is a critical value preset for judging whether the similarity between the target demand vector and the resource feature vector meets the standard, and its value range is usually between 0 and 1, and if the cosine similarity is used. When the similarity between two vectors is greater than the threshold, it indicates that the corresponding resource instance has a high matching degree with the resource demand and can be used as a candidate resource; otherwise, it is excluded. In this embodiment, the setting of the similarity threshold needs to be combined with the accuracy requirement of the business scenario.

[0075] Additionally, the preset vector database is a structured database storing resource feature vectors of all resource instances, which is optimized to support efficient vector retrieval operations such as approximate nearest neighbor search. The database contains the mapping relationship between resource feature vectors and corresponding resource instances, and can quickly return matching resource feature vectors and associated resource instance information according to vector similarity.

[0076] Additionally, the preset constraint condition is an additional condition for screening candidate resources in addition to vector similarity, which is set in advance based on business rules or user implicit requirements, including but not limited to cost constraints, performance constraints, reliability constraints, compliance constraints, etc.

[0077] Step S32, performing approximate nearest neighbor search in the preset vector database according to the target demand vector to obtain an initial candidate resource set; It should be noted that approximate nearest neighbor search is an efficient vector retrieval algorithm, which does not strictly calculate the similarity between the target vector and all vectors in the database, but quickly finds a batch of vectors most similar to the target vector through indexing technology, greatly improves the retrieval speed at the premise of sacrificing a little accuracy, and is suitable for large-scale vector databases.

[0078] It can be understood that approximate nearest neighbor search is performed using Milvus and other vector databases. The approximate nearest neighbor search calculation formula is as follows:

[0079] In the formula, is the number of returned most similar resources.

[0080] Further, the initial candidate resource set is a set of resource instances with the highest similarity to the target demand vector obtained by approximate nearest neighbor search, which has not been screened by the similarity threshold and the preset constraint condition, and is the basis for further screening.

[0081] Step S33, selecting a similar candidate resource set with a similarity greater than the similarity threshold from the initial candidate resource set; It should be noted that the similar candidate resource set is a set of resource instances with a similarity greater than the similarity threshold selected from the initial candidate resource set, which is screened only based on vector similarity to ensure that the matching degree of resources and demand meets the standard in the feature level.

[0082] It can be understood that the exact similarity, such as cosine similarity, of each resource instance in the initial candidate resource set to the target demand vector is calculated; then each similarity value is compared with a similarity threshold value, such as 0.8; then the resource instances with similarity values greater than the threshold value are screened out; and finally, these resource instances are summarized to form a similar candidate resource set.

[0083] In step S34, the resource instances in the similar candidate resource set that meet the preset constraint condition are taken as target candidate resources, and a target candidate resource set is determined according to the target candidate resources.

[0084] It should be noted that the target candidate resources are resource instances that are screened out from the similar candidate resource set and meet the preset constraint condition, which not only meet the vector similarity requirement, but also meet the additional business rules or user demand, and are the final candidate resources for scoring and sorting.

[0085] It can be understood that the detailed information of each resource instance in the similar candidate resource set is obtained; then whether these resource instances meet the preset constraint condition obtained in step S31 is checked one by one; then the resource instances that meet all the preset constraint conditions are determined as target candidate resources; and all the target candidate resources are summarized to form a target candidate resource set.

[0086] In step S40, scoring calculation is performed on the candidate resources in the target candidate resource set to obtain candidate resource scores. It can be understood that scoring calculation refers to a process of comprehensively scoring each candidate resource according to a preset evaluation index system. The evaluation index usually includes multiple dimensions, and each dimension is set with a corresponding weight according to its importance. The performance of the candidate resource in each dimension is converted into a score, and then the weights are combined for weighted summation to obtain the comprehensive score of each candidate resource.

[0087] Further, the candidate resource score is a score obtained by scoring calculation and used to represent the comprehensive advantages and disadvantages of the candidate resource. The score comprehensively considers the performance of the candidate resource in multiple evaluation indexes, and the higher the score, the more the candidate resource meets the overall demand of the user.

[0088] In addition, it can be understood that the evaluation index system for scoring calculation is determined, and the included indexes, such as performance matching degree, price reasonableness, service availability, historical failure rate, and the weights of the indexes, are specified; then for each candidate resource, the score of each evaluation index is calculated; the scores of each index are multiplied by the corresponding weights and summed to obtain the comprehensive score of each candidate resource, i.e., the candidate resource score.

[0089] In a feasible implementation, step S40 can include steps S41-S43: Step S41, obtain the performance matching degree weight, the cost benefit weight, the availability weight, and the scalability weight; It should be noted that the performance matching degree weight is a parameter that is pre-set and used to measure the importance of the performance matching degree score in the comprehensive score of the candidate resource, and the value range is 0 to 1. The size of the weight is determined according to the degree of dependence of the business scenario on performance, for example, in scenarios such as real-time transactions and big data analysis that have extremely high performance requirements, the performance matching degree weight will be set to be relatively high, such as 0.4; and in scenarios such as ordinary office work that have relatively low performance requirements, the weight will be relatively low, such as 0.2. Its role is to reflect the importance of performance matching in the comprehensive score, and to ensure that the score result is consistent with the focus of business needs.

[0090] The cost benefit weight is a parameter that measures the importance of the cost benefit score in the comprehensive score, and the value range is 0 to 1. The weight reflects the sensitivity of the business scenario to cost, for example, small and medium-sized enterprises have limited cost budget, and the cost benefit weight may be set to 0.3; and large enterprises pay more attention to performance, and the cost benefit weight may be set to 0.2. Through the weight, the relationship between cost and other factors can be balanced in the comprehensive score, and resources with low cost performance are avoided.

[0091] The availability weight is a parameter that measures the importance of the availability score in the comprehensive score, and the value range is 0 to 1. Availability refers to the ability of a resource to run continuously and stably, and for scenarios such as financial transactions and medical services that have extremely high stability requirements, the availability weight will be set to be relatively high, such as 0.3; for non-core business scenarios, the weight may be relatively low, such as 0.1. The weight ensures that high-availability resources receive due attention in the comprehensive score.

[0092] The scalability weight is a parameter that measures the importance of the scalability score in the comprehensive score, and the value range is 0 to 1. Scalability refers to the ability of a resource to dynamically adjust performance according to business growth, and for scenarios such as start-ups and emerging e-commerce that are rapidly expanding, the scalability weight may be set to 0.2; for scenarios with stable business, the weight may be set to 0.1. The weight enables the comprehensive score to reflect the long-term adaptation ability of the resource.

[0093] Step S42, calculate the performance matching degree score, the cost benefit score, the availability score, and the scalability score for the candidate resources in the target candidate resource set according to the performance matching degree weight, the cost benefit weight, the availability weight, and the scalability weight; It can be understood that the performance matching degree score is a score that measures the matching degree of the performance of a candidate resource with the performance requirement in the resource demand, and the value range is usually 0 to 10 points, with 10 points being a complete match. Its calculation basis includes the similarity of the performance characteristics of the candidate resource, such as computing power, memory, storage, and network, with the corresponding characteristics in the target demand vector, and the higher the similarity, the higher the score.

[0094] The cost-effectiveness score is a score for measuring the cost performance ratio of the candidate resource, with a value range of 0 to 10, which comprehensively considers the use cost and performance of the resource. When calculating, the cost of the candidate resource, such as rent, is compared with the average cost of resources with the same performance level. The more the cost is lower than the average value or the performance is higher than the cost of the same resource, the higher the score.

[0095] The availability score is a score for measuring the reliable operation ability of the candidate resource, with a value range of 0 to 10, which is calculated based on the service availability percentage, fault recovery time, backup mechanism and other availability characteristics of the resource. The higher the availability percentage, such as 99.99%, and the shorter the fault recovery time, such as 10 minutes, the higher the score.

[0096] The scalability score is a score for measuring the dynamic adjustment performance ability of the candidate resource, with a value range of 0 to 10, which is calculated according to the maximum expansion limit, expansion speed, and whether to support automatic expansion and contraction of the resource. The larger the maximum expansion limit, such as CPU can be expanded from 8 cores to 64 cores, and the faster the expansion speed, such as within 5 minutes, the higher the score.

[0097] It can be understood that for each candidate resource in the target candidate resource set, the similarity between its performance characteristics and the performance characteristics in the target demand vector is calculated, and the performance matching degree score is determined according to the similarity; then the cost of the candidate resource is compared with the average cost of the same performance resource, and the cost-effectiveness score is calculated; then the availability score is calculated according to the availability characteristics of the resource; and then the scalability score is calculated according to the scalability characteristics; the calculation of each score is combined with the corresponding weight.

[0098] In a feasible implementation, step S42 can include steps S421-S424: Step S421, determining the central processor weight, memory weight, storage weight, and network weight according to the performance matching degree weight, and calculating the performance matching degree score of the candidate resource in the target candidate resource set according to the central processor weight, the memory weight, the storage weight, and the network weight; It should be noted that the central processor weight is a sub-weight for measuring the importance of the central processor, CPU performance, in the overall performance matching evaluation, under the performance matching degree weight, with a value range of 0 to the performance matching degree weight, and its size is determined according to the degree of dependence of the business scenario on the CPU. For example, in a large data analysis scenario that requires a large amount of calculation, the central processor weight will be higher; and in a light office scenario, the weight will be lower. This weight, together with the memory weight, the storage weight, and the network weight, constitutes the subdivision of the performance matching degree weight, and the sum of the four is equal to the performance matching degree weight.

[0099] Memory weight is a sub-weight under performance matching degree weight, measuring the importance of memory performance, with a value range of 0 to performance matching degree weight, reflecting the dependence of business scenarios on memory capacity and read-write speed. For example, in scenarios running multiple tasks or large applications, the memory weight will be higher than the storage weight; while in simple data storage scenarios, the memory weight will be relatively low.

[0100] Storage weight is a sub-weight under performance matching degree weight, measuring the importance of storage performance such as capacity, type, and read-write speed, with a value range of 0 to performance matching degree weight, related to the demand of business scenarios for data storage and access. For example, in database scenarios requiring high-frequency read-write, the storage weight will be higher; while in archiving scenarios requiring only low-frequency storage, the weight will be lower.

[0101] Network weight is a sub-weight under performance matching degree weight, measuring the importance of network performance such as bandwidth and delay, with a value range of 0 to performance matching degree weight, depending on the requirements of business scenarios for data transmission speed and stability. For example, in scenarios such as video live streaming and real-time communication, the network weight will be significantly higher than other sub-weights; while in offline data processing scenarios, the network weight will be lower. The performance matching degree score calculation formula is as follows:

[0102] In the formula, , , , are the matching scores of central processors, memories, storages, and networks respectively, , , , are the corresponding central processor weight, memory weight, storage weight, and network weight respectively, is the performance matching degree score.

[0103] Step S422, according to the cost-effectiveness weight, determine the budget weight and the cost-effective weight, and according to the budget weight and the cost-effective weight, calculate the cost-effectiveness score of the candidate resource; It should be noted that the budget weight is a sub-weight under the cost-effectiveness weight, measuring the matching degree of the cost of the candidate resource and the budget limit in the resource demand, with a value range of 0 to the cost-effectiveness weight, and its size depends on the emphasis of business scenarios on budget compliance. For example, in the scenario of small and medium-sized enterprises with strict budget constraints, the budget weight will be higher than the cost-effective weight; while in scenarios focusing on long-term benefits, the budget weight may be lower.

[0104] The cost-effectiveness weight is a weight under the cost-effectiveness weight, which measures the importance of the performance-cost ratio of the candidate resource, and the value range is 0 to the cost-effectiveness weight, which reflects the degree of pursuit of the business scenario for the lowest cost under the premise of performance meeting or the optimal performance under the premise of fixed cost. For example, in a scenario that needs to balance cost and performance, the cost-effectiveness weight is set to be relatively high. The sum of the budget weight and the cost-effectiveness weight is equal to the cost-effectiveness weight. The cost-effectiveness score calculation formula is as follows:

[0105] In the formula, is the budget matching score, is the cost-effectiveness score, and are the budget weight and the cost-effectiveness weight, respectively, is the cost-effectiveness score.

[0106] Step S423, determining the reliability weight and the stability weight according to the availability weight, and calculating the availability score of the candidate resource according to the reliability weight and the stability weight; It should be noted that the reliability weight is a sub-weight under the availability weight, which measures the importance of the failure probability and the failure recovery capability of the candidate resource, and the value range is 0 to the availability weight, which is related to the demand of the business scenario for “few failures and fast recovery” of the resource. For example, in a financial transaction scenario that does not allow long interruptions, the reliability weight will be significantly higher than the stability weight.

[0107] The stability weight is a sub-weight under the availability weight, which measures the importance of the performance fluctuation degree of the candidate resource, such as CPU usage and bandwidth fluctuation, and the value range is 0 to the availability weight, which reflects the requirement of the business scenario for the “continuous stability” of the resource performance. For example, in a real-time data processing scenario, the stability weight will be higher to avoid the influence of performance fluctuation on data processing accuracy. The sum of the reliability weight and the stability weight is equal to the availability weight. The availability score calculation formula is as follows:

[0108] In the formula, is the reliability score, is the stability score, and are the reliability weight and the stability weight, respectively, is the availability score.

[0109] Step S424, determining the vertical expansion weight and the horizontal expansion weight according to the scalability weight, and calculating the scalability score of the candidate resource according to the vertical expansion weight and the horizontal expansion weight.

[0110] It should be noted that the vertical expansion weight is a sub-weight under the expansibility weight, which measures the importance of "upgrading hardware configuration" of the candidate resource, such as increasing the number of CPU cores or expanding the memory capacity, and the value range is 0 to the expansibility weight, which is related to the demand of the business scenario for "single-instance performance improvement". For example, in the scenario where business growth mainly reflects the improvement of single-task complexity, the vertical expansion weight will be higher.

[0111] The horizontal expansion weight is a sub-weight under the expansibility weight, which measures the importance of the ability of the candidate resource to increase the number of instances, such as expanding from 1 server to 10 servers, and the value range is 0 to the expansibility weight, which reflects the demand of the business scenario for "multi-instance parallel processing". For example, in the scenario where business growth reflects an increase in concurrency, the horizontal expansion weight will be higher. The sum of the vertical expansion weight and the horizontal expansion weight is equal to the expansibility weight. The expansibility score calculation formula is as follows:

[0112] In the formula, is the vertical expansion score, is the horizontal expansion score, and are the vertical expansion weight and the horizontal expansion weight, respectively, is the expansibility score.

[0113] Step S43, according to the performance matching degree score, the cost benefit score, the availability score and the expansibility score, the score calculation is carried out, and the candidate resource score is obtained.

[0114] It can be understood that the performance matching degree score, the cost benefit score, the availability score and the expansibility score of each candidate resource obtained by calculation are obtained. Then each score is multiplied by the corresponding weight obtained; then the four products are added to obtain the comprehensive score of the candidate resource, that is, the candidate resource score; the above calculation is repeated for all candidate resources in the target candidate resource set to obtain the candidate resource score of each resource. The candidate resource score calculation formula is as follows:

[0115] In the formula, , , and are the performance matching degree score, the cost benefit score, the availability score and the expansibility score, respectively, , , and are the performance matching degree weight, the cost benefit weight, the availability weight and the expansibility weight, respectively, is the candidate resource score.

[0116] Step S50, ranking the candidate resources according to the candidate resource scores to generate a recommended result list.

[0117] It should be noted that the recommended result list is an ordered list generated by ranking the candidate resources according to the candidate resource scores, and the candidate resources in the list are arranged from high to low according to the scores. The candidate resources arranged in the front represent better comprehensive performance and better meet the resource requirements of the user.

[0118] It can be understood that the candidate resource score of each candidate resource is obtained; then all candidate resources are ranked in order from high to low according to the scores, and if the scores of two candidate resources are the same, further ranking can be performed according to a preset secondary ranking rule; finally, the ranked candidate resources are arranged in the form of a list to generate a recommended result list, which can contain key information of each candidate resource for user reference.

[0119] In a feasible implementation, the complex business requirements of the user can be decomposed into multiple sub-requirements, each sub-requirement corresponds to a resource type, forming a sub-requirement set; for each sub-requirement in the sub-requirement set, a single resource recommendation algorithm is used to generate a corresponding candidate resource list, and then all feasible resource combinations are generated based on these candidate resources; when calculating the comprehensive score of the combination, in addition to the score of the single resource, the compatibility score between resources (such as the degree of docking and adaptation of different cloud vendors), the cross-cloud connection performance score (such as data transmission speed and delay), the overall architecture complexity score (such as the difficulty of deployment and maintenance), and the management convenience score (such as whether unified management is supported) need to be considered, and these factors are combined with the single resource score to calculate the comprehensive score; a multi-objective optimization method is used to balance performance (such as processing speed and stability), cost (such as total rent and operation and maintenance cost), and complexity (such as architecture simplicity), and the combination with the best comprehensive performance is selected from the feasible resource combinations; finally, an architecture topology diagram is generated for the resource combination to intuitively display the relationship between resources, data flow, and the distribution of each resource in different cloud environments.

[0120] The embodiment provides a resource recommendation method based on a multi-cloud environment. By constructing a resource feature vector and a target demand vector, and using vector similarity retrieval, multi-dimensional score calculation and sorting and other technical means, the technical problems of low resource matching efficiency and difficulty in accurately meeting complex user requirements in a multi-cloud environment are solved, and the beneficial effects of efficiently screening, accurately recommending resources meeting user requirements, and optimizing resource selection to improve user experience and resource utilization efficiency are achieved.

[0121] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as the above-mentioned first embodiment can be referred to the above introduction, and will not be repeated hereinafter. On this basis, please refer to Figure 2After step S50, the resource recommendation method based on a multi-cloud environment further includes steps S60-S80: In step S60, feedback information of the candidate resources in the recommendation result list is obtained, and the feedback information includes selection behavior information and satisfaction feedback. It should be noted that the feedback information is actual reaction data of the user on the candidate resources in the recommendation result list, which is used to evaluate the accuracy of the recommendation system and optimize subsequent recommendation strategies.

[0122] In addition, the selection behavior information is operation behavior data of the user on the candidate resources in the recommendation result list, including but not limited to the resource finally selected by the user, such as the resource clicked, ordered, deployed and used, the browsing path before selection, such as which resource details are viewed, the selection time consumption, the time from the display list to the confirmation of selection, whether to skip the recommendation list to search independently, etc. For example, the user selects the resource ranked third in the recommendation list instead of the resource ranked first, or the user does not select any recommended resource but manually searches for other resources, which all belong to the selection behavior information.

[0123] The satisfaction feedback is subjective evaluation data of the user on the selected resource, which is usually collected in the form of scoring, questionnaire or text evaluation, and includes the satisfaction on the performance, cost, availability and scalability of the resource.

[0124] In step S70, deviation information of the feedback information and the candidate resource score is analyzed according to the selection behavior information and the satisfaction feedback. It should be noted that the deviation information is difference data between the feedback information and the candidate resource score, which is used to quantify the degree of inconsistency between the system recommendation result and the actual preference of the user, including selection deviation and satisfaction deviation. The selection deviation represents the difference between the resource actually selected by the user and the high-score resource in the recommendation list. The satisfaction deviation represents the difference between the satisfaction score of the user on the selected resource and the candidate resource score of the resource.

[0125] It can be understood that by comparing the resource selected by the user with the candidate resource score and the difference of the sub-dimensions such as performance and cost of the resource ranked higher in the recommendation list, the selection deviation is determined, such as the cost-effectiveness score of the resource selected by the user is higher, but the performance score is slightly lower, which indicates that the system may underestimate the importance of cost. Then the satisfaction feedback is analyzed: the satisfaction of the user on each dimension, such as 5 stars for cost, is compared with the sub-dimension score corresponding to the resource, such as 8 points for cost-effectiveness, and the satisfaction deviation is calculated, such as the cost dimension satisfaction is higher than the score, which indicates that the cost-effectiveness weight may be too low. Finally, the selection deviation and the satisfaction deviation are integrated to form the quantitative deviation information, such as cost dimension deviation + 15% performance dimension deviation - 5%.

[0126] In this embodiment, the user selects a resource with a score of 8.5, ranked No. 3, with a cost-effectiveness score of 8.5, higher than the resource ranked No. 1 with a cost-effectiveness score of 7.8; the user's satisfaction with the cost is 5 stars, significantly higher than the expected value corresponding to the cost-effectiveness score of the resource, 8.5 corresponding to 4 stars, so the deviation information is a positive deviation of 12% in the cost-effectiveness dimension.

[0127] In step S80, the weight values of the performance matching degree weight, the cost-effectiveness weight, the availability weight, and the scalability weight are updated according to the deviation information.

[0128] It can be understood that based on the deviation information, the weight dimension that needs to be adjusted is determined, such as the positive deviation in the cost-effectiveness dimension, indicating that the cost-effectiveness weight may be too low; then a weight adjustment strategy is developed according to the quantitative degree of the deviation, such as 12%, positive deviation, the user's satisfaction is higher than the score, and the corresponding weight is increased, negative deviation, the user's satisfaction is lower than the score, and the corresponding weight is decreased, and the adjustment amplitude is positively correlated with the deviation size, such as 12% deviation corresponding to weight increase of 0.05; then the specific values of the performance matching degree weight, the cost-effectiveness weight, the availability weight, and the scalability weight are updated to ensure that the sum of the four weights after adjustment is still 1, such as the original cost-effectiveness weight 0.2, increased to 0.25, and the performance matching degree weight 0.4 decreased by 0.05 to 0.35, keeping the total sum 1; finally, the updated weights are stored in the weight configuration library for subsequent score calculation in the recommendation process.

[0129] In this embodiment, based on the deviation information of the positive deviation of 12% in the cost-effectiveness dimension, the cost-effectiveness weight is adjusted from 0.2 to 0.25, the performance matching degree weight is adjusted from 0.4 to 0.35, the availability weight 0.3 and the scalability weight 0.1 remain unchanged, and the total weight is still 1.

[0130] The user feedback can be converted into specific weight adjustment, making the subsequent score calculation more in line with the actual preferences of the user, realizing self-optimization of the recommendation system, and improving the long-term recommendation accuracy.

[0131] The embodiment provides a resource recommendation method based on a multi-cloud environment, which solves the technical problem of insufficient recommendation accuracy in the recommendation system caused by the mismatch between the user preferences and the system preset weights by means of acquiring user feedback information, analyzing the deviation between the feedback information and the score of the candidate resource, and dynamically adjusting the weight values, and achieves the beneficial effects of optimizing the recommendation strategy according to the actual feedback of the user, improving the recommendation accuracy and the user satisfaction.

[0132] It should be noted that the above examples are only used for understanding the present application and do not constitute a limitation on the resource recommendation method based on a multi-cloud environment of the present application, and more forms of simple transformation based on this technical concept are within the protection scope of the present application.

[0133] The application also provides a resource recommendation device based on a multi-cloud environment, which refers to Figure 3 The resource recommendation device based on the multi-cloud environment comprises: A resource acquisition module 10 is configured to acquire resource demand and resource information of a resource instance, perform feature extraction according to the resource information, and obtain a resource feature vector. A demand generation module 20 is configured to generate a target demand vector according to the resource demand and the resource feature vector. A resource candidate module 30 is configured to perform vector similarity retrieval according to the target demand vector, and obtain a target candidate resource set. A score calculation module 40 is configured to perform score calculation according to a candidate resource in the target candidate resource set, and obtain a candidate resource score. A resource recommendation module 50 is configured to sort the candidate resource according to the candidate resource score, and generate a recommendation result list.

[0134] The resource recommendation device based on the multi-cloud environment provided by the application adopts the resource recommendation method based on the multi-cloud environment in the above embodiment, and can solve the technical problem that it is difficult to balance the response speed and matching accuracy of resource recommendation in a multi-cloud environment. Compared with the prior art, the resource recommendation device based on the multi-cloud environment provided by the application has the same beneficial effects as the resource recommendation method based on the multi-cloud environment provided by the above embodiment, and other technical features in the resource recommendation device based on the multi-cloud environment are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0135] In an embodiment, the score calculation module 40 is further configured to acquire a performance matching degree weight, a cost-benefit weight, an availability weight, and an expansibility weight; calculate a performance matching degree score, a cost-benefit score, an availability score, and an expansibility score for a candidate resource in the target candidate resource set according to the performance matching degree weight, the cost-benefit weight, the availability weight, and the expansibility weight; and perform score calculation according to the performance matching degree score, the cost-benefit score, the availability score, and the expansibility score, to obtain a candidate resource score.

[0136] In an embodiment, the score calculation module 40 is further configured to determine a central processor weight, a memory weight, a storage weight, and a network weight according to the performance matching degree weight, and calculate a performance matching degree score of a candidate resource in the target candidate resource set according to the central processor weight, the memory weight, the storage weight, and the network weight; determine a budget weight and a cost performance weight according to the cost performance weight, and calculate a cost performance score of the candidate resource according to the budget weight and the cost performance weight; determine a reliability weight and a stability weight according to the availability weight, and calculate an availability score of the candidate resource according to the reliability weight and the stability weight; determine a vertical expansion weight and a horizontal expansion weight according to the expansibility weight, and calculate an expansibility score of the candidate resource according to the vertical expansion weight and the horizontal expansion weight.

[0137] In an embodiment, the resource recommendation module 50 is further configured to acquire feedback information of a candidate resource in the recommendation result list, the feedback information including selection behavior information and satisfaction feedback; analyze deviation information of the feedback information and the candidate resource score according to the selection behavior information and the satisfaction feedback; and update weight values of the performance matching degree weight, the cost performance weight, the availability weight, and the expansibility weight according to the deviation information.

[0138] In an embodiment, the resource candidate module 30 is further configured to acquire a similarity threshold, a preset vector database, and a preset constraint condition; perform approximate nearest neighbor search in the preset vector database according to the target demand vector to obtain an initial candidate resource set; select a similar candidate resource set with a similarity greater than the similarity threshold from the initial candidate resource set; take a resource instance in the similar candidate resource set that satisfies the preset constraint condition as a target candidate resource, and determine a target candidate resource set according to the target candidate resource.

[0139] In an embodiment, the demand generation module 20 is further configured to determine a preset demand mapping rule according to the business scenario type; perform standardization processing on the capability demand information, the storage demand information, the network bandwidth demand information, and the budget limit demand information according to the preset demand mapping rule to obtain a standardized demand parameter mapping; perform vector space isomorphism on the standardized demand parameter mapping and the resource feature vector to obtain an isomorphic demand vector; and perform normalization processing on the isomorphic demand vector to obtain a target demand vector.

[0140] In an embodiment, the resource obtaining module 10 is further configured to obtain computing capability information, memory information, storage information, network information and a preset norm of the resource instance in the resource information; extract basic features, cost features, availability features and scalability features according to the computing capability information, the memory information, the storage information and the network information; perform normalization processing according to the basic features, the cost features, the availability features and the scalability features to obtain a normalized resource vector; perform feature combination on the normalized resource vector to obtain a resource feature vector of a preset dimension, and perform normalization processing on the resource feature vector according to the preset norm to obtain a resource feature vector.

[0141] The application provides a resource recommendation device based on a multi-cloud environment, which comprises at least one processor and a memory in communication connection with the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the resource recommendation method based on the multi-cloud environment in the above embodiment one.

[0142] Reference will be made to the accompanying drawings Figure 4 which shows a structural schematic diagram of the resource recommendation device based on the multi-cloud environment suitable for being used to implement the embodiments of the application. The resource recommendation device based on the multi-cloud environment in the embodiments of the application can include but is not limited to mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistant, PAD (Portable Application Description, PMP (Portable Media Player, car terminals (such as car navigation terminals, etc.), and fixed terminals such as digital TVs, desktop computers, etc. Figure 4 The resource recommendation device based on the multi-cloud environment shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the application.

[0143] As Figure 4As shown, the multi-cloud environment based resource recommendation device can include a processing apparatus 1001 (e.g., a central processor, a graphics processor, etc.) that can perform various appropriate actions and processes according to programs stored in a ROM (Read Only Memory) 1002 or programs loaded from a storage apparatus 1003 into a RAM (Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the multi-cloud environment based resource recommendation device are also stored. The processing apparatus 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: input apparatuses 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output apparatuses 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage apparatus 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication apparatus 1009. The communication apparatus 1009 can allow the multi-cloud environment based resource recommendation device to perform wireless or wired communication with other devices to exchange data. Although the multi-cloud environment based resource recommendation device having various systems is shown in the figure, it should be understood that all the shown systems are not required to be implemented or possessed. More or less systems can be alternatively implemented or possessed.

[0144] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program codes for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through the communication apparatus, or installed from the storage apparatus 1003, or installed from the ROM 1002. When the computer program is executed by the processing apparatus 1001, the above-mentioned functions defined in the methods of the embodiments disclosed in the present application are performed.

[0145] The multi-cloud environment based resource recommendation device provided in the present application adopts the multi-cloud environment based resource recommendation method in the above-mentioned embodiments, and can solve the technical problem that it is difficult to balance the response speed and matching accuracy of resource recommendation in a multi-cloud environment. Compared with the prior art, the multi-cloud environment based resource recommendation device provided in the present application has the same beneficial effects as the multi-cloud environment based resource recommendation method provided in the above-mentioned embodiments, and other technical features in the multi-cloud environment based resource recommendation device are the same as the features disclosed in the previous embodiment method, which will not be described here.

[0146] It should be understood that various parts of the present application can be realized in hardware, software, firmware, or a combination thereof. In the above description of embodiments, specific functional, structural, material or characteristic features are combined in a manner known per se; however, each feature can also be implemented individually or in a different combination.

[0147] The above description is merely illustrative of the application and is not intended to limit the scope of the application. Any variations and modifications that can be made by those skilled in the art without departing from the spirit of the application are intended to be included in the scope of the application. The scope of the application is defined by the appended claims.

[0148] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e., a computer program) for performing the resource recommendation method based on a multi-cloud environment in the above-described embodiments.

[0149] The computer readable storage medium provided by the present application may, for example, be a U disk, but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more conductive wires, a portable computer disk, a hard disk, a RAM (Random Access Memory), a ROM (Read Only Memory), an Erasable Programmable Read Only Memory (EPROM), an optical fiber, a CD-ROM (CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electrical wire, an optical cable, an RF (Radio Frequency), or any suitable combination of the above.

[0150] The above computer readable storage medium can be included in the resource recommendation device based on a multi-cloud environment; or can exist separately and not be assembled into the resource recommendation device based on a multi-cloud environment.

[0151] The computer readable storage medium described above carries one or more programs, when the one or more programs are executed by the resource recommendation device based on a multi-cloud environment, the resource recommendation device based on a multi-cloud environment: obtaining resource requirements, resource information of resource instances, and performing feature extraction according to the resource information to obtain a resource feature vector; generating a target demand vector according to the resource requirements and the resource feature vector; performing vector similarity retrieval according to the target demand vector to obtain a target candidate resource set; performing score calculation according to the candidate resources in the target candidate resource set to obtain a candidate resource score; and sorting the candidate resources according to the candidate resource score to generate a recommendation result list.

[0152] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0153] The flow diagrams and the block diagrams in the drawings are illustrations of possible architectures, functions, and operations for systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flow diagrams or block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may

[0154] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. In some cases, the name of the module does not constitute a limitation on the module itself.

[0155] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e., a computer program) for executing the resource recommendation method based on a multi-cloud environment. The computer readable program instructions can solve the technical problem of difficultly balancing the response speed and matching accuracy of resource recommendation in a multi-cloud environment. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the resource recommendation method based on a multi-cloud environment provided by the above embodiments, and will not be described here.

[0156] The present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the resource recommendation method based on a multi-cloud environment are implemented.

[0157] The computer program product provided by the present application can solve the technical problem of difficultly balancing the response speed and matching accuracy of resource recommendation in a multi-cloud environment. Compared with the prior art, the computer program product provided by the present application has the same beneficial effects as the resource recommendation method based on a multi-cloud environment provided by the above embodiments, and will not be described here.

[0158] The above only describes some embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation, direct / indirect application in other related technical fields based on the technical concept of the present application, and the content of the specification and drawings are included in the patent protection scope of the present application.

Claims

1. A resource recommendation method based on a multi-cloud environment, characterized in that, The method includes: Obtain resource information of resource requirements and resource instances, and extract features based on the resource information to obtain resource feature vectors; Generate a target demand vector based on the resource requirements and the resource feature vector; Based on the target demand vector, a vector similarity retrieval is performed to obtain a set of target candidate resources; The candidate resource scores are calculated based on the candidate resources in the target candidate resource set. The candidate resources are sorted according to their scores to generate a list of recommended results.

2. The method as described in claim 1, characterized in that, The step of calculating the candidate resource score based on the candidate resources in the target candidate resource set includes: Obtain the performance matching weight, cost-effectiveness weight, availability weight, and scalability weight; Based on the performance matching weight, the cost-effectiveness weight, the availability weight, and the scalability weight, a performance matching score, a cost-effectiveness score, an availability score, and a scalability score are calculated for each candidate resource in the target candidate resource set. The candidate resource score is obtained by calculating the score based on the performance matching score, the cost-effectiveness score, the availability score, and the scalability score.

3. The method as described in claim 2, characterized in that, The step of calculating the performance matching score, cost-effectiveness score, availability score, and scalability score for the candidate resources in the target candidate resource set based on the performance matching score, cost-effectiveness score, availability score, and scalability score respectively includes: The CPU weight, memory weight, storage weight, and network weight are determined based on the performance matching weight, and the performance matching score of the candidate resources in the target candidate resource set is calculated based on the CPU weight, memory weight, storage weight, and network weight. The budget weight and cost-effectiveness weight are determined based on the cost-benefit weight, and the cost-benefit score of the candidate resource is calculated based on the budget weight and the cost-effectiveness weight. The reliability weight and stability weight are determined based on the availability weight, and the availability score of the candidate resource is calculated based on the reliability weight and the stability weight. The vertical expansion weight and the horizontal expansion weight are determined based on the expansion weight, and the expansion score of the candidate resource is calculated based on the vertical expansion weight and the horizontal expansion weight.

4. The method as described in claim 1, characterized in that, After the step of sorting the candidate resources according to the candidate resource scores and generating a recommendation result list, the method further includes: Obtain feedback information on candidate resources in the recommendation results list, including selection behavior information and satisfaction feedback; Based on the selection behavior information and the satisfaction feedback, analyze the deviation information between the feedback information and the candidate resource score; The weight values ​​of performance matching weight, cost-effectiveness weight, availability weight, and scalability weight are updated based on the deviation information.

5. The method as described in claim 1, characterized in that, The step of performing vector similarity retrieval based on the target demand vector to obtain the target candidate resource set includes: Obtain the similarity threshold, preset vector database, and preset constraints; Based on the target demand vector, an approximate nearest neighbor search is performed in the preset vector database to obtain an initial candidate resource set; Select a set of similar candidate resources from the initial candidate resource set whose similarity is greater than the similarity threshold; Resource instances that satisfy the preset constraints in the similar candidate resource set are taken as target candidate resources, and a target candidate resource set is determined based on the target candidate resources.

6. The method as described in claim 1, characterized in that, The step of generating the target demand vector based on the resource demand and the resource feature vector includes: Based on the resource requirements, obtain information on computing power requirements, storage requirements, network bandwidth requirements, budget constraints, and business scenario types; Determine the preset requirement mapping rules based on the business scenario type; The capacity requirement information, storage requirement information, network bandwidth requirement information, and budget constraint requirement information are standardized according to the preset requirement mapping rules to obtain a standardized requirement parameter mapping. Based on the standardized demand parameter mapping and the resource feature vector, a vector space isomorphism is performed to obtain an isomorphic demand vector; The isomorphic demand vector is normalized to obtain the target demand vector.

7. The method as described in claim 1, characterized in that, The steps for extracting features based on the resource information to obtain resource feature vectors include: Obtain the computing power information, memory information, storage information, network information, and preset norm of the resource instance in the resource information; Based on the computing power information, memory information, storage information, and network information, basic features, cost features, availability features, and scalability features are extracted. The normalized resource vector is obtained by normalizing the basic characteristics, cost characteristics, availability characteristics, and scalability characteristics. The normalized resource vector is combined with features to obtain a resource feature vector of a preset dimension, and the resource feature vector is normalized according to the preset norm to obtain a resource feature vector.

8. A resource recommendation device based on a multi-cloud environment, characterized in that, The device includes: The resource acquisition module is used to acquire resource information of resource requirements and resource instances, and to extract features based on the resource information to obtain resource feature vectors; The demand generation module is used to generate a target demand vector based on the resource demand and the resource feature vector; The resource candidate module is used to perform vector similarity retrieval based on the target demand vector to obtain a set of target candidate resources. The scoring calculation module is used to calculate the scores of candidate resources based on the candidate resources in the target candidate resource set, and obtain the candidate resource scores. The resource recommendation module is used to sort the candidate resources according to their scores and generate a list of recommended results.

9. A resource recommendation device based on a multi-cloud environment, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the resource recommendation method based on a multi-cloud environment as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the resource recommendation method based on a multi-cloud environment as described in any one of claims 1 to 7.

Images