Reserved instance reselling system and reserved instance reselling method

Abstract

Disclosed is a reserved instance reselling system and a reserved instance reselling method. Specifically, disclosed is a reserved instance reselling system, a reserved instance reselling method, and a recording medium for creating a profit by purchasing a reserved instance from a reserved instance providing system such as a cloud service system and reselling it to a client. The reserved instance reselling system according to an embodiment of the present invention resells a reserved instance purchased from a reserved instance providing system to a client, and includes: an instance usage amount prediction unit that predicts an instance usage amount of the client using a machine learning model learned in advance; a reserved instance purchase amount determination unit that determines a reserved instance purchase amount based on the instance usage amount; a reserved instance purchase unit that purchases the reserved instance based on the reserved instance purchase amount; and a reserved instance reselling unit that resells the purchased reserved instance to the client.

Description

Technical Field

[0001] This invention relates to a reserved instance resale system and method, and more specifically, to a reserved instance resale system and method for purchasing reserved instances from a reserved instance provisioning system such as a cloud service system and reselling them to a client. Background Technology

[0002] Cloud systems provide an environment where users can freely access desired services by connecting to virtual servers via terminals. Cloud systems offer a variety of services, including: services that lease storage space to individual users, allowing them to connect to the cloud system and store data in allocated space; infrastructure services such as basic computing environments or network services; platform services for computer use or solutions; and software services that allow users to use application software over the network. Cloud systems are used in various fields such as mobile applications, games, shopping malls, and social networking services. These cloud systems offer the following advantages: users can access resources without time or space constraints and can monitor resource usage in real time. Furthermore, cloud systems not only provide services that function similarly to existing physical servers but also manage the provided service resources in real time and provide search and categorization functions for these resources.

[0003] Customers pay a fee tailored by the cloud system as the cost of using its resources. For example, if a customer makes an on-demand request to the cloud system to use resources and pays an on-demand fee, the customer uses the cloud system's resources (instances) by paying a fee proportional to the time of use. More recently, cloud systems allow customers to purchase reserved instances, enabling them to use cloud system resources at a lower cost than the hourly on-demand fee. Reserved instances allow customers to reserve and use an instance for a specified period (e.g., 1 year, 3 years, etc.). For example, Amazon's EC2 reserved instances offer discounts of up to 62% compared to on-demand fees. Summary of the Invention

[0004] The purpose of this invention is to provide a reserved instance resale system, a reserved instance resale method, and a recording medium containing a program for executing the reserved instance resale method, which generate revenue by purchasing reserved instances from a reserved instance provisioning system such as a cloud service system and reselling them to clients.

[0005] Furthermore, the purpose of this invention is to provide a reserved instance resale system, a reserved instance resale method, and a recording medium that, in the process of reselling reserved instances to customers, calculates the appropriate purchase quantity of reserved instances by predicting the customer's instance usage, thereby maximizing the return on investment generated from the resale of reserved instances.

[0006] The reserved instance resale system of the present invention resells reserved instances to customers by purchasing them from a reserved instance providing system. The reserved instance resale system includes: an instance usage prediction unit that predicts the instance usage of the customer using a pre-learned machine learning model; a reserved instance purchase quantity determination unit that determines the reserved instance purchase quantity based on the instance usage; a reserved instance purchase unit that purchases the reserved instances based on the reserved instance purchase quantity; and an instance resale unit for reselling the purchased reserved instances to customers.

[0007] The reserved instance resale system of this invention may further include a learning unit, which extracts multiple variables, including usage patterns and number of customers, from learning data containing on-demand request data requested by customers in the past to generate a machine learning model based on the instance's operating system, instance type, region, and availability zone.

[0008] The learning department can generate the following models: a generalized linear regression model that predicts instance usage using usage patterns and customer numbers for each operating system, instance type, region, and availability zone as independent variables; a multivariate spline regression model that predicts instance usage by applying a smoothing function to the aforementioned independent variables; a long short-term memory model that predicts future instance usage patterns based on past instance usage patterns; and a multivariate regression model that predicts instance usage using deep learning based on the aforementioned independent variables.

[0009] The above instance usage prediction unit can predict the above instance usage based on at least one of the above generalized linear regression model, the above multivariate spline regression model, the above long short-term memory model, and the above multivariate regression model.

[0010] The aforementioned learning unit may include a generalized linear regression (GLR) learning unit for generating the aforementioned generalized linear regression model, a multivariate spline regression (MSR) learning unit for generating the aforementioned multivariate spline regression model, a long short-term memory (LSTM) learning unit for generating the aforementioned long short-term memory model, and a multivariate regression learning unit for generating the aforementioned multivariate regression (MR) model.

[0011] The instance usage prediction unit described above may include: an error comparison unit for comparing multiple prediction errors of multiple machine learning models, including the generalized linear regression model, the multivariate spline regression model, the long short-term memory model, and the multivariate regression model described above; a machine learning model selection unit for selecting one of the multiple machine learning models based on the comparison results of the multiple prediction errors of the multiple machine learning models described above; and an instance usage estimation unit for estimating the instance usage based on the machine learning model selected by the machine learning model selection unit. The machine learning model selection unit may select the machine learning model with the lowest root mean square error (RMSE) value from the multiple machine learning models described above.

[0012] The instance usage prediction unit may include: a first instance usage prediction unit that predicts instance usage based on the generalized linear regression model and outputs a first instance usage; a second instance usage prediction unit that predicts instance usage based on the multivariate spline regression model and outputs a second instance usage; a third instance usage prediction unit that predicts instance usage based on the long short-term memory model and outputs a third instance usage; a fourth instance usage prediction unit that predicts instance usage based on the multivariate regression model and outputs a fourth instance usage; an error analysis unit that calculates a weighted value of multiple machine learning models by analyzing multiple prediction errors of multiple machine learning models, including the generalized linear regression model, the multivariate spline regression model, the long short-term memory model, and the multivariate regression model; and an instance usage estimation unit that estimates the instance usage by adjusting the response rates of the first instance usage, the second instance usage, the third instance usage, and the fourth instance usage based on the weighted values ​​of the multiple machine learning models.

[0013] The reserved instance resale system of this invention may further include a limit value removal unit, which predicts the usage of the instance by removing data that exceeds the standard range from the data related to the instance request.

[0014] The above-mentioned reserved instance purchase quantity determination department can recommend the purchase quantity of the above-mentioned reserved instances based on operating system, instance type, region and availability zone, and calculate and provide the rate of return when purchasing reserved instances.

[0015] The reserved instance resale method of the present invention resells reserved instances to customers by purchasing them from a reserved instance providing system. The reserved instance resale method includes: a step in which an instance usage prediction unit predicts the instance usage of the customer using a pre-learned machine learning model; a step in which a reserved instance purchase quantity determination unit determines the reserved instance purchase quantity based on the instance usage; a step in which a reserved instance purchase unit purchases the reserved instance from the reserved instance providing system based on the reserved instance purchase quantity; and a step in which an instance resale unit resells the reserved instance to the customer.

[0016] The reserved instance resale method of this invention may further include the step of the learning department extracting multiple variables, including usage patterns and number of customers, from learning data containing on-demand request data requested by customers, based on the instance's operating system, instance type, region, and availability zone, to generate the above-mentioned machine learning model.

[0017] The steps for generating the above machine learning models may include generating the following models: a generalized linear regression model that predicts instance usage using usage patterns and customer numbers for each operating system, instance type, region, and availability zone as independent variables; a multivariate spline regression model that predicts instance usage by applying a smoothing function to the above independent variables; a long short-term memory model that predicts future instance usage patterns based on past instance usage patterns; and a multivariate regression model that predicts instance usage using deep learning based on the above independent variables.

[0018] The steps for predicting the usage of the above instances may include using at least one machine learning model, such as the generalized linear regression model, the multivariate spline regression model, the long short-term memory model, and the multivariate regression model, to predict the usage of the above instances.

[0019] The steps for predicting the above instance usage may include: an error comparison unit comparing multiple prediction errors of multiple machine learning models, including the above generalized linear regression model, the above multivariate spline regression model, the above long short-term memory model, and the above multivariate regression model; a machine learning model selection unit selecting one of the multiple machine learning models based on the comparison results of the multiple prediction errors of the multiple machine learning models; and an instance usage estimation unit estimating the above instance usage based on the machine learning model selected by the above machine learning model selection unit.

[0020] The steps for predicting the instance usage may include: a first instance usage prediction unit predicting the instance usage based on the generalized linear regression model and outputting a first instance usage; a second instance usage prediction unit predicting the instance usage based on the multivariate spline regression model and outputting a second instance usage; a third instance usage prediction unit predicting the instance usage based on the long short-term memory model and outputting a third instance usage; a fourth instance usage prediction unit predicting the instance usage based on the multivariate regression model and outputting a fourth instance usage; an error analysis unit calculating a weighted average of multiple machine learning models, including the generalized linear regression model, the multivariate spline regression model, the long short-term memory model, and the multivariate regression model, based on multiple prediction errors; and an instance usage estimation unit adjusting the response rates of the first instance usage, the second instance usage, the third instance usage, and the fourth instance usage based on the weighted averages of the multiple machine learning models to estimate the instance usage.

[0021] According to the present invention, a computer-readable recording medium is provided that records a program for performing the above-described reserved instance resale method.

[0022] According to embodiments of the present invention, a reserved instance resale system, a reserved instance resale method, and a recording medium containing a program for executing the reserved instance resale method are provided, which generate revenue by purchasing reserved instances from a reserved instance provisioning system such as a cloud service system and reselling them to clients.

[0023] This invention can estimate customer instance usage by using all of the generalized linear regression model, multivariate spline regression model, long short-term memory model, and multivariate regression model, and adjusting the response rate of each model.

[0024] This invention can predict instance usage in response to future customer instance requests even when no customer instance requests have been received. By utilizing multiple search engines that are weighted according to prediction errors, the accuracy of predicting future customer instance usage can be improved.

[0025] According to embodiments of the present invention, a reserved instance resale system, a reserved instance resale method, and a recording medium are provided, which calculate the appropriate purchase quantity of reserved instances by predicting the instance usage of customers during the process of reselling reserved instances to customers, thereby maximizing the yield generated from the resale of reserved instances.

[0026] This invention can minimize the loss range when reselling reserved instances to customers. Attached Figure Description

[0027] Figure 1 This is a conceptual diagram illustrating the revenue creation model of the reserved instance resale system in an embodiment of the present invention.

[0028] Figure 2 This is a conceptual diagram illustrating the revenue generated by the reserved instance resale system according to an embodiment of the present invention, which is equivalent to the difference between the on-demand instance sales revenue and the reserved instance payment fee.

[0029] Figure 3 This is a structural diagram of the reserved instance resale system according to an embodiment of the present invention.

[0030] Figure 4 This is a flowchart of the reserved instance resale method according to an embodiment of the present invention.

[0031] Figure 5 This is a structural diagram of the learning unit and the instance usage prediction unit constituting a reserved instance resale system according to an embodiment of the present invention.

[0032] Figure 6 A flowchart illustrating the process of predicting instance usage according to an embodiment of the present invention is provided.

[0033] Figure 7 This is a structural diagram illustrating the learning unit and instance usage prediction unit constituting a reserved instance resale system according to another embodiment of the present invention.

[0034] Figure 8 A flowchart illustrating the process of predicting instance usage according to another embodiment of the present invention. Detailed Implementation

[0035] The advantages and features of the invention, as well as methods of implementing them, can be clearly understood by referring to the following detailed description of embodiments and accompanying drawings. However, the invention is not limited to the embodiments disclosed below, but can be implemented in various different forms. These embodiments are provided merely to complete the disclosure of the invention and to fully inform those skilled in the art of the scope of the invention, which is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0036] In this specification, when a part is said to "comprise" a component, unless specifically stated otherwise, this does not mean that other components are excluded, but may also include other components. The term "part" as used in this specification refers to a unit that processes at least one function or operation, and may, for example, represent software, a field-programmable gate array (FPGA), or a hardware component. The functionality provided by the "part" may be distributed among multiple components for execution or combined with other additional components. The "part" in this specification is not necessarily limited to software or hardware, but may be configured to reside in an addressable storage medium and may include one or more processors (microprocessors) for processing data stored in the storage medium. Embodiments of the invention will now be described in detail with reference to the accompanying drawings.

[0037] Regarding instance usage by multiple customers, the reserved instance resale system of this invention utilizes a pre-learned machine learning model and artificial intelligence to predict customer instance usage. Based on the predicted instance usage, the reserved instance resale system automatically determines the number of reserved instances to purchase from a reserved instance provider system, such as a cloud service system, and resells the purchased reserved instances to multiple customers.

[0038] According to embodiments of the present invention, the reserved instance resale system can maximize the revenue generated through instance resale by purchasing an appropriate number of reserved instances and reselling them to multiple customers. Furthermore, the reserved instance resale system can offer more instance discounts to multiple customers by increasing instance resale revenue, expand white label partners, prevent the churn of existing customers such as strategic clients, and increase revenue by attracting new customers.

[0039] Furthermore, according to embodiments of the present invention, cloud service providers can increase sales revenue by increasing the sales volume of reserved instances and can stably ensure profitability by ensuring instance utilization during the contract period. Moreover, from the customer's perspective, various instance discounts can be obtained through the reserved instance resale system, and instances such as virtual servers provided by the cloud service provider can be used for a lower fee than on-demand fees.

[0040] Figure 1 This is a conceptual diagram illustrating the revenue creation model of the reserved instance resale system in an embodiment of the present invention.

[0041] Reference Figure 1 The reserved instance resale system 20 generates revenue by purchasing discounted reserved instances from one or more cloud service systems 10 and reselling those instances to one or more customers 30. Cloud service system 10 is a system that provides reserved instances and may include cloud providers such as Amazon Web Services (AWS) and Azure.

[0042] The reserved instance resale system 20 can pay for the purchase of reserved instances from the cloud service system 10 and receive on-demand fees for customer 30's on-demand requests. In this case, the reserved instance resale system 20 can generate revenue equivalent to the difference between the on-demand fees paid by customer 30 and the reserved instance purchase fees paid to the cloud service system 10.

[0043] The reserved instance resale system 20 may have wired / wireless communication interfaces for data communication between the cloud service system 10 and the customer 30. Communication interfaces may include, but are not limited to, local area networks (LANs), integrated services digital networks (ISDNs), code division multiple access (CDMAs), wideband code division multiple access (WCDMAs), global system for mobile communication (GSMs), long term evolution (LTEs), world interoperability for microwave access (WIMAXs), wireless local area networks (WLANs), wide area networks (WANs), and wireless fidelity (WIFIs).

[0044] Figure 2 This is a conceptual diagram illustrating the revenue generated by the reserved instance resale system according to an embodiment of the present invention, equivalent to the difference between the on-demand instance sales revenue and the reserved instance payment fee. Figure 2 In this context, x represents the number of reserved instances purchased (x > 0), α represents the instance price (α > 0), θ represents the actual number of instances used (θ ≥ 0), and β represents the discount rate for reserved instances (0 < β < 1). α(1-β)x represents the reserved instance purchase fee payable to the cloud service system. αθ represents the on-demand instance sales fee received from the customer. If the on-demand sales fee is greater than the reserved instance purchase fee, it corresponds to a revenue range.

[0045] Conversely, if the cost of purchasing reserved instances is greater than the cost of selling on demand, that is, if the amount of reserved instances purchased x is greater than 1 / (1-β) times the actual amount of instances used θ, then it falls within the loss range.

[0046] Therefore, in order to maximize the benefits obtained by reselling reserved instances to customers, it is necessary to accurately predict the instance usage (on-demand request quantity) of multiple customers and calculate the reserved instance purchase quantity based on the predicted instance usage to purchase an appropriate number of reserved instances from the cloud service system.

[0047] The reserved instance resale system accurately predicts instance usage to purchase instances and provides on-demand instances that correspond to customer requests, thereby maximizing revenue while minimizing the loss range.

[0048] The reserved instance resale system of this invention can predict the instance usage of multiple customers through automatic analysis using machine learning models and artificial intelligence (AI) methods, and provide reserved instance purchase recommendations.

[0049] Figure 3 This is a structural diagram of the reserved instance resale system according to an embodiment of the present invention. Figure 4 This is a flowchart of the reserved instance resale method according to an embodiment of the present invention.

[0050] Reference Figure 3 and Figure 4 In order to predict instance usage, the reserved instance resale system 20 may include a learning data collection unit 210, a learning unit 220, a machine learning model 230, a data collection unit 240, a limit value removal unit 250, an instance usage prediction unit 260, a reserved instance purchase quantity determination unit 270, a reserved instance purchase unit 280, an instance resale unit 290, and an instance management unit 300.

[0051] The instance usage prediction unit 260 can use a machine learning model 230 related to instance usage to predict instance usage for multiple customers.

[0052] Machine learning model 230 can model the relationship between multiple independent variables, such as each instance's operating system (OS), each region, each instance type, each availability zone customer's instance usage pattern, and the number of customers, and a dependent variable that includes the instance usage of multiple customers.

[0053] The instance operating system includes, but is not limited to, Windows / Linux, etc. Instance types can include various combinations of central processing unit (CPU), memory, storage, and network capacity, such as Amazon EC2's T2, C4, M4, M5, and C5. Instance regions have multiple isolated locations called availability zones, such as, in the case of Amazon EC2, US East (Northern Visayas), US East (Ohio), US West (Northern California), US West (Oregon), Canada (Central), EU (Frankfurt), EU (London), EU (Paris), Asia Pacific (Seoul), Asia Pacific (Tokyo), South America (São Paulo), etc., but are not limited to these.

[0054] A learning data collection unit 210 and a learning unit 220 can be provided to generate a machine learning model 230 related to instance usage. The learning data collection unit 210 can collect learning data including information such as the instance usage patterns and number of customers for each operating system, region, instance type, and availability zone from multiple past customers. In this embodiment, the learning data may include past on-demand request data for instances from multiple customers.

[0055] The Learning Unit 220 can extract multiple variables from the learning data, including each instance type, operating system, region, usage pattern and number of customers in each availability zone, instance usage, etc., to generate one or more machine learning models. Figure 4 (Step S10). The machine learning model can be provided as an instance usage prediction model that learns instance usage through machine learning or deep learning.

[0056] In this embodiment, the learning unit 220 can generate at least one machine learning model among the following: Generalized Linear Regression (GLR), Multivariate Spline Regression (MSR), Long Short Term Memory (LSTM), and Multivariate Regression (MR). However, this is merely an example, and machine learning models not listed may also be generated.

[0057] The Generalized Linear Regression (GLR) model is a common multivariate generalized linear regression analysis model. It is a machine learning model used to predict instance usage using each instance type, each availability zone, usage pattern for each instance type, and number of customers as independent variables. The GLR model can identify meaningful variables among many variables through AIC stepwise regression and can apply variables in the form of multiple linear relationships in the process of predicting instance usage.

[0058] Multivariate spline regression (MSR) is a non-linear regression model that overcomes the limitations of linear models that assume linearity in real-world problems. It's a machine learning model used to predict instance usage by applying a smoothing function to independent variables. The variables used in MSR are the same as those used in generalized linear regression, but unlike generalized linear regression where each variable's relationship is applied linearly, MSR treats the pattern of each variable as a function and applies a smoothing function to each. Among many variables, meaningful variables can be identified through an ANOVA test. The smoothness (k_i) of each variable can be calculated using K-fold cross-validation to find the k_i with the minimum error.

[0059] Long Short-Term Memory (LSTM) models are deep learning machine learning models that predict future instance usage based on past instance usage patterns. This model aims to overcome the shortcomings of Recurrent Neural Networks (RNNs), specifically the reduced learning ability due to gradient descent during backward pass computation. This is achieved by adding cell states to the hidden states of the RNN to predict instance usage. An LSTM model can be implemented, for example, as a model that predicts the next unit of time based on patterns from the past 60 units (60 input layers, 2 hidden layers, 1 output layer).

[0060] Multiple regression (MR) models are predictive models used by deep learning to predict instance usage based on independent variables such as generalized linear regression. The Multiple Regression (MR) model (Keras-D / L) is a model that uses deep learning in the Keras environment to predict generalized linear regression, with the same variable information as the generalized linear regression model.

[0061] The instance usage prediction unit 260 can predict the future instance usage of multiple customers based on at least one of the following machine learning models: generalized linear regression (GLR), multiple spline regression (MSR), long short-term memory (LSTM), and multiple regression (MR).

[0062] The instance usage prediction unit 260 can predict the future instance usage of multiple customers based on information collected by the data collection unit 240, such as each operating system (OS), each region, each availability zone, instance usage patterns of each instance type, and number of customers.

[0063] The instance usage prediction unit 260 predicts instance usage in response to future customer instance requests, even when no customer instance requests have been received. It calculates instance usage based on various machine learning models from the learning unit and analyzes the prediction errors for multiple machine learning models. Furthermore, the instance usage prediction unit 260 estimates instance usage by assigning weights to a machine learning model based on the prediction errors or by adjusting the response rate of each machine learning model.

[0064] The instance usage prediction unit 260 selects a machine learning model based on prediction error and weighting, or uses multiple machine learning models and adjusts the response rate to estimate instance usage.

[0065] To more accurately predict instance usage by removing error factors, the limit removal unit 250 removes limit value data that exceeds the standard range from the data collected by the data collection unit 240. The instance usage prediction unit 260 can predict instance usage for each operating system, each region, each availability zone, and each instance type based on multiple independent variables of the data from which the limit value data has been removed. Figure 4 Step S20).

[0066] The reserved instance purchase quantity determination unit 270 can determine the reserved instance purchase quantity based on the instance usage predicted by the instance usage prediction unit 260. Figure 4 (Step S30). In this embodiment, the reserved instance purchase quantity determination unit 270 can determine the reserved instance purchase quantity that maximizes the return on investment based on the predicted instance usage of multiple customers using a maximum likelihood estimation (MLE) algorithm. Furthermore, the reserved instance purchase quantity determination unit 270 recommends the purchase quantity of reserved instances for applicable payer accounts based on operating system, region, availability zone, and instance type, and calculates and provides the return on investment when purchasing reserved instances.

[0067] The reserved instance purchase quantity determination unit 270 can calculate the reserved instance purchase quantity x by considering multiple variables such as instance price α, reserved instance discount rate β, and predicted instance usage θ. The reserved instance purchase quantity determination unit 270 can calculate the reserved instance purchase quantity based on a maximum likelihood estimation algorithm. The likelihood (likelihood function) L(θ|x) is proportional to the probability p(x|θ) of sample x occurring when θ is assumed, and represents the reasonableness of the estimate of parameter θ given sample x.

[0068] The reserved instance purchase quantity determination unit 270 can determine the reserved instance purchase quantity by applying different conditions to the maximum likelihood estimation model based on the utilization at each time point. The reserved instance purchase quantity determination unit 270 can calculate the reserved instance purchase quantity by using extreme values ​​to remove induced bias using percentages such as 50%, 75%, 90%, and 95%. For example, the reserved instance purchase quantity determination unit 270 can use the maximum likelihood estimation algorithm to find the reserved instance purchase quantity x that yields the maximum return y based on the following equation.

[0069]

[0070] The reserved instance purchase unit 280 can purchase reserved instances from the cloud service system 10 based on the reserved instance purchase quantity determined by the reserved instance purchase quantity determination unit 270, according to the operating system, region, availability zone, and instance type. Figure 4 Step S40).

[0071] Instance resale unit 290 can resell reserved instances purchased from cloud service system 10 to customers based on their on-demand requests. Figure 4 (Step S50). Instance resale department 290 can offer various discounts to resell instances based on the type of customer (e.g., white-label partners, strategic customers, new customers, etc.).

[0072] In this embodiment, the instance resale unit 290 can request payment of the on-demand amount from customers, and in order to attract new customers and prevent the churn of strategic customers, it can request payment of the on-demand amount with applicable discounts from new customers or strategic customers. Furthermore, to prevent the churn of strategic customers, it can also offer instance discounts to strategic customers, such as offering convertible reserved instances (RIs) that can change the price of a Standard Reserved Instance (Standard RI) to an instance family with a 40% discount based on the on-demand fee.

[0073] The Instance Management Unit 300 can manage the purchase details and sales details of reserved instances for each operating system, region, availability zone, and instance type, as well as information on instances available for sale to customers. The information managed by the Instance Management Unit 300 can be fed back to the Learning Data Collection Unit 210 and the Data Collection Unit 240, and used to generate machine learning models for determining the next reserved instance purchase volume and predicting instance usage. The Instance Management Unit 300 may include functionality for supporting and managing white labels (WL).

[0074] Figure 5 This is a structural diagram of the learning unit and the instance usage prediction unit constituting a reserved instance resale system according to an embodiment of the present invention. Figure 6 A flowchart illustrating a process for predicting instance usage according to an embodiment of the present invention is provided. (Refer to...) Figure 5 and Figure 6 The learning unit 220 can generate multiple different machine learning models for predicting instance usage.

[0075] In an embodiment, the learning unit 220 may include a generalized linear regression learning unit 222 for generating a generalized linear regression (GLR) model, a multivariate spline regression learning unit 224 for generating a multivariate spline regression (MSR) model, a long short-term memory learning unit 226 for generating a long short-term memory (LSTM) model, and a multivariate regression learning unit 228 for generating a multivariate regression (MR) model.

[0076] The instance usage prediction unit 260 may include an error comparison unit 261, a machine learning model selection unit 262, and an instance usage estimation unit 263.

[0077] The error comparison unit 261 can compare multiple prediction errors of multiple machine learning models, including the generalized linear regression (GLR) model 232, the multiple spline regression (MSR) model 234, the long short-term memory (LSTM) model 236, and the multiple regression (MR) model 238. Figure 6 Step S21). The machine learning model selection unit 262 can select one machine learning model from multiple machine learning models based on the comparison results of multiple prediction errors of multiple machine learning models. Figure 6 Step S22). In an embodiment, the machine learning model selection unit 262 may, for example, calculate the root mean square error value of each machine learning model according to the following mathematical formula and then select the machine learning model with the smallest root mean square error value.

[0078]

[0079] The instance usage estimation unit 263 can estimate instance usage based on one or more machine learning models selected by the machine learning model selection unit 262. Figure 6 Step S23). In an embodiment, the instance usage estimation unit 263 may predict instance usage based on a machine learning model with the minimum root mean square error value, or based on more than one machine learning model with a root mean square error value less than the standard value.

[0080] In an embodiment, when estimating instance usage based on multiple machine learning models, the instance usage estimation unit 263 may perform statistical processing (e.g., arithmetic / geometric mean, median, etc.) on the multiple instance usages estimated based on the multiple machine learning models to estimate the instance usage for each operating system, each region, each availability zone, and each instance type.

[0081] Figure 7 This is a structural diagram illustrating the learning unit and instance usage prediction unit constituting a reserved instance resale system according to another embodiment of the present invention. Figure 8 A flowchart illustrating a process for predicting instance usage according to another embodiment of the present invention is provided. (Refer to...) Figure 7 and Figure 8 The instance usage prediction unit 260 can output multiple instance usage prediction units based on multiple different machine learning models. Figure 8 Step S24).

[0082] The instance usage prediction unit 260 may include a first instance usage prediction unit 264, a second instance usage prediction unit 265, a third instance usage prediction unit 266, a fourth instance usage prediction unit 267, an error analysis unit 268, and an instance usage estimation unit 269.

[0083] The first instance usage prediction unit 264 can predict instance usage based on the generalized linear regression (GRL) model 232 and output the first instance usage. The second instance usage prediction unit 265 can predict instance usage based on the multiple spline regression (MSR) model 234 and output the second instance usage. The third instance usage prediction unit 266 can predict instance usage based on the long short-term memory (LSTM) model 236 and output the third instance usage. The fourth instance usage prediction unit 267 can predict instance usage based on the multiple regression (MR) model 238 and output the fourth instance usage.

[0084] Error analysis unit 268 can analyze multiple prediction errors of multiple machine learning models, including generalized linear regression (GRL) model 232, multiple spline regression (MSR) model 234, long short-term memory (LSTM) model 236, and multiple regression (MR) model 238, to calculate the weighted value of multiple machine learning models. Figure 8 Step S25).

[0085] The instance usage estimation unit 269 can estimate instance usage by adjusting the response rates of the first instance usage, the second instance usage, the third instance usage, and the fourth instance usage based on the weighted values ​​of multiple machine learning models calculated by the error analysis unit 268. Figure 8 Step S26). In an embodiment, the greater the prediction error (e.g., root mean square error), the larger the weighting value assigned to the machine learning model by the instance usage estimation unit 269 to estimate the instance usage.

[0086] Each machine learning model can predict instance usage for multiple different future periods (e.g., 1 month, 3 months, 6 months, etc.). If the prediction shows an upward trend, the monthly reserved instance purchase amount can be recommended by reflecting the predicted value in a 1-month unit (minimum purchase period). Conversely, if the prediction shows a downward trend, the reserved instance purchase amount can be recommended by reflecting the predicted value in a 6-month unit (maximum prediction period).

[0087] The above embodiments can be implemented using hardware components, software components, and / or combinations of hardware and software components. For example, the apparatus, methods, and components described in the embodiments can be implemented using one or more general-purpose or special-purpose computers, such as processors, controllers, arithmetic logic units (ALUs), digital signal processors, microcomputers, field-programmable gate arrays (FPGAs), programmable logic units (PLUs), microprocessors, or any other means capable of executing and responding to instructions.

[0088] The processing device can run an operating system and one or more software applications running on the operating system. Furthermore, the processing device accesses, stores, manipulates, processes, and generates data in response to the execution of the software. For ease of understanding, although sometimes described as using a single processing device, those skilled in the art will understand that the processing device may include multiple processing elements and / or various types of processing elements.

[0089] For example, the processing device may include multiple processors or a processor and a controller. Other processing configurations, such as a parallel processor, may also be employed. Software may include computer programs, code, instructions, or combinations thereof, and the processing device may be configured to operate as needed or to command the processing device independently or collectively.

[0090] In order to be interpreted by or to provide instructions or data to a processing device, software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium or device, or transmitted signal wave. Software may be distributed across networked computer systems and stored or executed in a distributed manner. Software and data may be stored on more than one computer-readable recording medium.

[0091] The method according to the embodiments can be implemented by recording program instructions executable by various computer units on a computer-readable recording medium. The computer-readable medium may contain program instructions, data files, data structures, etc., individually or in combination. The program instructions recorded on the medium may be instructions specifically designed and configured for the embodiments or instructions known and usable by those skilled in the art of computer software.

[0092] Examples of computer-readable recording media include: magnetic media such as hard disks, floppy disks, and magnetic tapes; optical media such as CD-ROMs and DVDs; and hardware devices specifically configured for storing and executing program instructions, such as read-only memory (ROM), random access memory (RAM), and flash memory. Examples of program instructions include not only machine language code, such as code compiled by a compiler, but also high-level language code that can be executed by a computer using an interpreter. The aforementioned hardware devices can be configured to function as more than one software module to perform the operations of the embodiments, and vice versa.

[0093] As described above, although the embodiments have been illustrated with specific examples and accompanying drawings, those skilled in the art can make various modifications and variations based on this description. For example, even if the described techniques are performed in a different order than the described methods, and / or the components of the described systems, structures, devices, circuits, etc., are combined or integrated in a different form than the described methods, or are replaced or substituted by other components or equivalent technical solutions, suitable results can still be achieved. Therefore, other examples, other embodiments, and technical solutions equivalent to the scope of the invention claims also fall within the scope of the invention claims described below.

Claims

1. A reserved instance resale system, which purchases reserved instances from a reserved instance provider system and resells them to customers, characterized in that, include: The Learning Department generates machine learning models by analyzing data; The instance usage prediction department uses a machine learning model pre-learned by the learning department to predict the instance usage of customers who are to be sold reserved instances. The reserved instance purchase quantity determination department determines the reserved instance purchase quantity based on the above instance usage quantity; The reserved instance purchasing department purchases the reserved instances according to the above-mentioned reserved instance purchase volume; as well as The instance resale department is responsible for reselling the aforementioned reserved instances to customers. The usage prediction unit mentioned above includes: The error analysis unit calculates a weighted average of the prediction errors of multiple machine learning models; and The instance usage estimation department estimates the customer's instance usage by adjusting the response rate of the instance usage of the aforementioned machine learning models based on a weighted average of the aforementioned machine learning models. Specifically, the reserved instance purchase quantity determination unit calculates the aforementioned reserved instance purchase quantity in order to maximize the revenue corresponding to the difference between the reserved instance purchase fee paid to the cloud service system and the on-demand instance sales amount paid by the customer. The reserved instance purchase fee paid by the reserved instance purchase department to the cloud service system includes the aforementioned reserved instance purchase quantity, instance price, and discount rate. The reservation instance purchase quantity determination process is based on the instance usage of multiple customers. It uses a maximum likelihood estimation algorithm to remove the limiting value of induced bias, thereby determining the reservation instance purchase quantity that maximizes the return on investment. If the predicted usage of the aforementioned instances shows an upward trend, the reserved instance purchase quantity determination department recommends additional reserved instance purchase quantities by reflecting the predicted usage of the aforementioned instances within the minimum purchase period. If the predicted usage of the aforementioned instances shows a downward trend, the reserved instance purchase quantity determination unit recommends the additional reserved instance purchase quantity by reflecting the predicted usage of the aforementioned instances during the maximum prediction period. The aforementioned learning department extracts multiple variables, including usage patterns and the number of customers, from learning data containing on-demand request data from past customer requests, categorized by instance operating system, instance type, region, and availability zone to generate machine learning models. The learning department generates the following models: a generalized linear regression model that predicts instance usage using usage patterns and customer numbers for each operating system, instance type, region, and availability zone as independent variables; a multivariate spline regression model that predicts instance usage by applying a smoothing function to the above independent variables; a long short-term memory model that predicts future instance usage patterns based on past instance usage patterns; and a multivariate regression model that predicts instance usage using deep learning based on the above independent variables.

2. The reserved instance resale system according to claim 1, characterized in that, The aforementioned error analysis unit analyzes the prediction error of at least one machine learning model among the generalized linear regression model, multivariate spline regression model, long short-term memory model, and multivariate regression model generated by the aforementioned learning unit.

3. The reserved instance resale system according to claim 2, characterized in that, The usage prediction unit mentioned above also includes: The first instance usage prediction unit predicts the usage of the above instance based on the above generalized linear regression model and outputs the first instance usage. The second instance usage prediction unit predicts the usage of the above instance based on the above multivariate spline regression model and outputs the usage of the second instance. The third instance usage prediction unit predicts the instance usage based on the aforementioned long short-term memory model and outputs the third instance usage; and The fourth instance usage prediction unit predicts the usage of the above instances based on the aforementioned multiple regression model and outputs the usage of the fourth instance. The instance usage estimation unit estimates the instance usage by adjusting the response rates of the first instance usage, the second instance usage, the third instance usage, and the fourth instance usage based on the weighted values ​​of multiple machine learning models.

4. The reserved instance resale system according to claim 1, characterized in that, It also includes a limit removal unit, which is used to remove data that exceeds the standard range from the data related to instance requests in order to predict the usage of the aforementioned instances.

5. The reserved instance resale system according to claim 1, characterized in that, The above-mentioned reserved instance purchase quantity determination department recommends the purchase quantity of the above-mentioned reserved instances based on the operating system, instance type, region, and availability zone, and calculates and provides the rate of return when purchasing reserved instances.

6. A method for reselling reserved instances, characterized in that, by purchasing reserved instances from a reserved instance provisioning system and reselling them to customers, include: Regarding the instance usage of customers who are to sell reserved instances, the instance usage forecasting department uses a pre-learned machine learning model to forecast the instance usage of the aforementioned customers. The reserved instance purchase quantity determination department determines the reserved instance purchase quantity based on the above instance usage quantity; The steps for the reserved instance purchasing department to purchase the reserved instances from the reserved instance providing system based on the reserved instance purchase quantity; as well as The instance resale department will resell the above-mentioned reserved instances to customers. The steps for predicting the usage of the above instances include: The steps involve calculating a weighted average of multiple prediction errors from multiple machine learning models; and The step of estimating the instance usage by adjusting the response rate of the instance usage of multiple machine learning models based on the weighted values ​​of multiple machine learning models described above. Specifically, the number of reserved instances purchased is calculated to maximize the revenue corresponding to the difference between the reserved instance purchase fees paid to the cloud service system and the on-demand instance sales revenue paid by the customer. The reserved instance purchase fee paid by the reserved instance purchase department to the cloud service system includes the aforementioned reserved instance purchase quantity, instance price, and discount rate. The reservation instance purchase quantity determination process is based on the instance usage of multiple customers. It uses a maximum likelihood estimation algorithm to remove the limiting value of induced bias, thereby determining the reservation instance purchase quantity that maximizes the return on investment. If the predicted usage of the aforementioned instances shows an upward trend, the reserved instance purchase quantity determination department recommends additional reserved instance purchase quantities by reflecting the predicted usage of the aforementioned instances within the minimum purchase period. If the predicted usage of the aforementioned instances shows a downward trend, the reserved instance purchase quantity determination unit recommends the additional reserved instance purchase quantity by reflecting the predicted usage of the aforementioned instances during the maximum prediction period. The reserved instance resale method also includes a step whereby the learning department extracts multiple variables, including usage patterns and the number of customers, from learning data containing on-demand request data from past customer requests, based on the instance's operating system, instance type, region, and availability zone, to generate a machine learning model. The steps for generating the above machine learning models include generating the following models: a generalized linear regression model that predicts instance usage using usage patterns and customer numbers for each operating system, instance type, region, and availability zone as independent variables; a multivariate spline regression model that predicts instance usage by applying a smoothing function to the above independent variables; a long short-term memory model that predicts future instance usage patterns based on past instance usage patterns; and a multivariate regression model that predicts instance usage using deep learning based on the above independent variables.

7. The method for reselling reserved instances according to claim 6, characterized in that, The aforementioned machine learning models include generalized linear regression models, multivariate spline regression models, long short-term memory models, and multivariate regression models.

8. The method for reselling reserved instances according to claim 6, characterized in that, The steps for predicting the usage of the above instances also include: The first instance usage prediction unit predicts the instance usage based on the above generalized linear regression model and outputs the first instance usage. The second instance usage prediction unit predicts the usage of the above instance based on the above multivariate spline regression model and outputs the usage of the second instance. The third instance usage prediction unit predicts the instance usage based on the above long short-term memory model and outputs the third instance usage. The fourth instance usage prediction unit predicts the usage of the above instances based on the above multiple regression model and outputs the usage of the fourth instance; and The instance usage estimation unit estimates the instance usage by adjusting the response rates of the first instance usage, the second instance usage, the third instance usage, and the fourth instance usage based on the weighted values ​​of multiple machine learning models.

9. The method for reselling reserved instances according to claim 6, characterized in that, The steps for predicting instance usage also include removing data that is outside the standard range from the data related to instance requests.

10. The method for reselling reserved instances according to claim 6, characterized in that, In the steps of determining the number of reserved instances to purchase mentioned above Recommend the number of reserved instances to purchase based on operating system, instance type, region, and availability zone, and calculate and provide the rate of return when purchasing reserved instances.

Images