Method and device for identifying idle and wasted hardware and software resources of cloud platform users

Abstract

The application relates to a cloud platform user idle waste software and hardware resource behavior identification method and device, and belongs to the technical field of cloud computing. The method comprises the following steps: acquiring real-time data of a cloud platform user, including monitoring data of user login on the platform, user opening of a virtual machine, opening of software, license request, virtual machine runtime resource occupation, and client and cloud end traffic interaction; extracting feature data in the real-time data; taking the feature data of the user as input and inputting the feature data into a trained software and hardware resource waste identification model to obtain a user behavior identification result. The application constructs a feature model suitable for user idle connection identification, that is, a software and hardware resource waste identification model, can realize real-time monitoring of the waste of software and hardware resources by the user, automatically identifies the waste, has high identification accuracy, maximally avoids misidentification problems, and improves user experience.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of cloud computing, and particularly relates to a method and device for identifying idle and wasted software and hardware resource behaviors of a user of a cloud platform. BACKGROUND

[0002] On an oilfield exploration and development cloud platform, a large number of professional software are often deployed. In general, running of the professional software will occupy a large amount of hardware resources. In addition, since the software is generally commercial software, sharing and use are realized through purchase of a license and user request for a license server to improve the use efficiency of the license and staggered use. However, in actual situations, the use efficiency of the license and the hardware resources is not high, and there are many wasted behaviors, including: abnormal exit of a user; intentional occupation of a license by a user to ensure smooth progress of the business of the user; long-term occupation of a license by a user without work; and forgetting to close an opened connection and a virtual machine by a user, and the like. Since the resources are limited, it is very important to identify the above behaviors to improve the utilization rate of the software and hardware resources.

[0003] At present, the methods for identifying idle and wasted software and hardware resource behaviors of a user mainly include the following types:

[0004] (1) Setting a user connection time length and a license idle threshold value, reminding a user whether to continue using when the time is reached, and disconnecting if the user does not select to continue using.

[0005] (2) Monitoring keyboard and mouse message events of a user, and judging that a user system is in an idle state if there is no keyboard and mouse message event of the user in a period of time, and releasing software and hardware resources and closing a connection by a platform.

[0006] (3) Setting an allowed idle time and a function module not monitored for different software and different users, and not collecting a license even if it is idle.

[0007] Obviously, the above methods are too limited in setting, and each has a disadvantage, as follows:

[0008] For the method (1), a user is interrupted in work and the experience of the user is affected by the pop-up window asking the user.

[0009] For the method (2), if a user is doing a long-time calculation or rendering, the user does not need to always move the keyboard and the mouse, and even the user can not log in to the platform to continuously run a task in the background, and the task will be wrongly closed.

[0010] For method (3), the administrator needs to set a large number of rules according to different users and software, and the management is very cumbersome. Since the reconfiguration is set according to experience, it is difficult to set reasonable values for dozens of software and thousands of platform users. If the setting time is too long, a large amount of software and hardware resources will be wasted. If the setting time is relatively short, it will cause the user's task to be interrupted, or a pop-up window will remind the user to continue to increase the use time, resulting in a decrease in user experience, or the user will ignore the prompt, resulting in interruption of the user's normal use.

[0011] Taking a cloud platform of an oilfield research institute as an example, the cloud platform includes more than 20 professional software, which can be divided into CPU, memory type and GPU type according to the demand for hardware resources; and can be divided into computing type and graphics processing type according to the use function. After the user logs in to the cloud platform, the user first connects the virtual machine in the cloud, opens the professional software deployed in the virtual machine, and after the professional software requests the license server successfully, the corresponding work can be carried out. Since the user will appear in the use process, the browser is closed, the remote client is closed, the user forgets to close, and the malicious occupation of the license resource is not used, etc., which leads to that the platform license does not produce the economic benefits, the limited virtual machine resources are occupied, and the users with real needs are in waiting. However, the existing solutions cannot effectively and accurately identify the idle and waste of software and hardware resources of the user, and are easy to mistakenly disconnect the platform connection of the normal use user, affecting the user experience. SUMMARY

[0012] The purpose of the present application is to provide a method and device for identifying the idle and waste of software and hardware resources of a user of a cloud platform, to solve the problem that the existing method cannot accurately identify the idle and waste of software and hardware resources of the user, resulting in poor user experience.

[0013] Based on the above purpose, a technical solution of a method for identifying the idle and waste of software and hardware resources of a user of a cloud platform is as follows:

[0014] Step 1: Obtain real-time data of a user of a cloud platform, including monitoring data of user login on the platform, user opening of a virtual machine, opening of software, license request, virtual machine running resource occupation, and client and cloud end traffic interaction, extract feature data in the real-time data, the feature data including N features related to idle and waste behavior of the user, N>3;

[0015] Step 2: input the feature data of the user as input into a trained software and hardware resource waste identification model to obtain a behavior identification result of the user; the software and hardware resource waste identification model is obtained by inputting the feature data marked with idle and waste behavior into a neural network for training.

[0016] Based on the above purpose, a cloud platform user idle waste hardware and software resource behavior identification device technical scheme is as follows:

[0017] Comprising a memory and a processor, and a computer program stored on the memory and running on the processor, the processor is coupled with the memory, the processor executes the computer program to realize the above-mentioned identification method.

[0018] The beneficial effects of the above two technical schemes are:

[0019] The identification method and device of the application, through the real-time data of the cloud platform user, extracts the feature data related to the user idle waste behavior, constructs a feature model suitable for user idle connection identification, that is, a hardware and software resource waste identification model, which is used for real-time monitoring of the user waste hardware and software resources and automatic identification, high identification accuracy, maximum avoidance of misidentification problems, and improved user experience.

[0020] Further, in step two, the determination step of the feature data marked with idle waste behavior is as follows:

[0021] (1) Obtain the historical data of each monitoring data, and extract the attribute data therein;

[0022] (2) Clean the attribute data;

[0023] (3) Data sampling, constructing sample data, dividing the sample data into idle time period and non-idle time period according to the size and duration of the interactive flow, and marking idle behavior and non-idle behavior according to the divided idle time period and non-idle time period.

[0024] Specifically, in step (1), the attribute data includes: user, software type used, monitoring time, whether the user is online on the platform, whether the user opens the application window, CPU, GPU, memory, disk I / O of the hardware resource where the user is located, software occupied CPU, software occupied GPU, software occupied memory, software occupied disk I / O, interactive flow, opening duration, system type, and whether it occupies a license.

[0025] Specifically, in step (2), the cleaning of the attribute data includes filling the missing values, removing duplicate data, and reducing redundant attributes.

[0026] Specifically, in step (3), the result division of the idle time period and the non-idle time period includes the following steps:

[0027] Count the number of user interaction traffic within each set time period, divide the time interval, and determine the frequency of each user interaction traffic, that is, the ratio of the number of times the interaction traffic belongs to the time interval to the total number of traffic statistics.

[0028] Based on the frequency distribution of user interaction traffic, a threshold P is determined. Time periods with less than P interaction traffic are marked as idle time periods; time periods with more than or equal to P interaction traffic are marked as non-idle time periods.

[0029] To mark idle behavior, the method also includes: when at least two consecutive periods are marked as idle time periods, and the time interval between these marking records exceeds a set time threshold T, the boundary time point between the marked idle time period and the non-idle time period is found backward from the current marked time period and recorded as an idle behavior, until all idle behaviors are marked.

[0030] Because the duration of idle behavior varies greatly when users connect and interact with the cloud platform using different types of software, using the same time threshold T for all of them would lead to discrimination errors. To avoid this problem, different time thresholds T need to be determined for different user software. The steps for determining the time threshold T include:

[0031] Using Δt = 0.5 min as the step size, statistics were collected at different time thresholds T. i The number of times an activity is identified as idle; based on a time threshold T. i A line graph is plotted with the number of idle behaviors as the horizontal and vertical axes. The absolute value of the slope of the line segment formed by adjacent thresholds is calculated. The average of the absolute values ​​of the slopes of the adjacent N (N≥3) line segments is used as the stability index S. The optimal threshold T is determined by the stability index S.

[0032] The feature data used in training the software and hardware resource waste identification model, as well as the feature data collected and extracted in real time in step one, include: software opening time, effective limit time, total idle time, effective idle time percentage, number of idle times, total interaction traffic, average interaction traffic, interaction traffic fluctuation, CPU utilization fluctuation, GPU utilization fluctuation, memory utilization fluctuation, disk I / O utilization fluctuation, average GPU utilization, and average disk I / O speed. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method for identifying user idle and wasteful software and hardware resources in Embodiment 1 of the present invention;

[0034] Figure 2 This is an interactive traffic frequency distribution diagram in Embodiment 1 of the method of the present invention;

[0035] Figure 3is the broken line graph of the number of idle behaviors at different time thresholds in the method embodiment 1 of the present application;

[0036] Figure 4 is the identification model schematic diagram of the BP neural network in the method embodiment 1 of the present application;

[0037] Figure 5 is the identification method flow chart of the idle waste of software and hardware resources behaviors by the cloud platform business system in the method embodiment 2 of the present application. DETAILED DESCRIPTION

[0038] The specific implementation of the present application will be further described below in combination with the drawings.

[0039] Method embodiment 1:

[0040] The present embodiment proposes an identification method of the idle waste of software and hardware resources behaviors by the cloud platform user, and the arrangement flow is as shown in the figure Figure 1 The implementation idea of the method is as follows: the features of the idle connection and the closed connection of the user are extracted from the real-time use data and the collected historical data on the monitoring platform, the identification model for the waste use behaviors of the software and hardware resources of the user is constructed, the identification model is trained, the management of the permitted resources and the hardware resources is optimized according to the identification result of the identification model, and the use rate of the software and hardware resources is improved.

[0041] The specific steps of the identification method of the idle waste of software and hardware resources behaviors by the cloud platform user in the present embodiment are as follows:

[0042] Step one, obtaining the real-time data of the cloud platform user, and extracting the feature data therefrom.

[0043] The feature data in the present step is a plurality of features related to the idle waste behaviors of the user which are preset in advance, and the selection principle and the extraction process of the features will be specifically described in step two.

[0044] Step two, taking the feature data of the user as the input, inputting into the trained waste identification model of the software and hardware resources, obtaining the behavior identification result of the user, identifying the idle waste behaviors of the user, and performing corresponding processing.

[0045] Specifically, the feature data obtained in step one is inputted into the previously trained identification model, if the output value is 1, it is determined as the normal idle behavior, if the output value is 0, it is determined as the waste idle behavior. According to the identification result, prompt information is given to the user, if the prompt information is not responded by the user within a certain time, the user data is temporarily stored, the virtual machine is suspended, and the occupied permit is recycled.

[0046] In the present step, the establishment process of the waste identification model of the software and hardware resources is as follows:

[0047] S1. Collection and processing of data to obtain sample data, the specific steps are:

[0048] (1) Collect and organize the platform user login, user opening of the virtual machine, opening of the software, license request, monitoring data of virtual machine runtime resource occupation, monitoring of client and cloud traffic interaction, etc. Extract the feature data and time series data of the user's use of software and hardware resources on the cloud platform, organize and summarize these data, and write them into the database.

[0049] For example, 17 attribute data, i.e., feature data and time series data of software and hardware resources, are extracted from the collected and organized data, including: user, software type used, monitoring time, whether the user is online on the platform, user opening of the application window, CPU, GPU, memory, disk I / O of the hardware resource where the user is located, software occupied CPU, software occupied GPU, software occupied memory, software occupied disk I / O, interactive traffic, opening duration, system type, and whether a license is occupied.

[0050] Since these data are all runtime monitoring data of the cloud platform and features of software and hardware themselves, they are measurable known data, and the present application does not involve the data origin and acquisition process, which will not be described.

[0051] (2) The attribute data collected above is cleaned.

[0052] The specific data cleaning process includes filling of missing values, removal of duplicate data, and reduction of redundant attributes. The specific operation process is:

[0053] a. Discard the row data with more than 40% of the missing attribute values;

[0054] b. For missing category data, complete it by searching the database and management personnel;

[0055] c. For missing CPU, GPU, memory data, and traffic interaction data, fill them by moving average method according to time sequence;

[0056] d. Then, remove the duplicate data and reduce the redundant data in the data (the reduction is introduced here).

[0057] (3) Selectively extract the data processed in step (2), i.e., sampling, to construct sample data; mark the results of the sample data, and divide the results into idle time period and non-idle time period.

[0058] The sampling in this step refers to the equidistant sampling method, which reduces the amount of monitoring data, allows the sample data to accommodate longer time, and greatly simplifies the calculation.

[0059] For example, if the interactive traffic number and the fluctuation of CPU, GPU and memory are used to determine whether the user is in a real working state, the sampling of the interactive traffic number distribution is taken as an example, 7 working days in a month are randomly selected, then divided by 30s, and the size of the interactive traffic in each 30s is calculated.

[0060] Then, the time series data after sampling is divided into intervals, and a time interval frequency distribution table is determined, and the discrimination threshold for dividing the user working time and idle time is determined according to the table.

[0061] For example, the interactive traffic is counted by 30s, and the counted interactive traffic is divided into equal intervals according to a group distance of 1000 to determine the frequency distribution table as shown in Table 1.

[0062] Table 1

[0063]

[0064]

[0065] In Table 1, the interactive traffic number is the number of information exchanged between the user and the cloud platform within 30s, and the occupancy frequency is the ratio of the number of intervals to which the interactive traffic belongs to the total number of traffic statistics. For example, within a group distance (1 thousand interactive information), 30670 interactive times are generated, and the occupancy frequency (here, actually per group distance, 1 thousand information, i.e. frequency ratio group distance) is 30670 / 1000=30.67. Therefore, by Table 1, the regularity of the user interactive traffic distribution can be determined, and the threshold for dividing the normal working state and the system idle state is given, i.e. the threshold for dividing the idle time period and the non-idle time period is obtained, which is used to generate the 0,1 result label of the monitoring time series data, and then the idle behavior is divided and recognized.

[0066] In order to determine the division threshold and realize the recognition of the non-normal idle time period of the user, the normal system use time period and the idle time period need to be recognized from the original data, and then the non-normal idle time period is recognized from the idle time period.

[0067] In this step, according to the size and duration of the interactive traffic, whether it is an idle time is divided, so a kind of optimization method for determining the division threshold is provided, and the specific method is as follows:

[0068] a. Idle behavior division:

[0069] According to the frequency distribution of the user's interactive traffic, the discrimination threshold P of the suspected idle behavior is set.

[0070] For example, the monitoring data of the user using professional software on the cloud platform is stored in the database, and a connection usually has several thousand status record data. In the user monitoring data, if the interactive traffic is not 0, it indicates that the user is using a remote window, and if the interactive traffic monitoring data is 0, it indicates that the user has closed the remote connection window. After closing the remote connection window, the user can also perform a computing task in the background, or can be in an idle state. By attaching Figure 2 In the middle, the interactive traffic value is compared with the actual operation. When the interactive traffic is greater than 5000, it is normal use, and at this time, P = 5000 can be set as the discrimination threshold of the suspected idle behavior.

[0071] Therefore, when the interactive traffic number of 30s is less than the threshold P, the time period is counted as a suspected idle time period, and is marked with 0. Conversely, for the normal use time period, 1 is used for marking.

[0072] If the interactive traffic is continuously marked as 0, when the time interval between these marked records exceeds a certain threshold (denoted as T), the last marked 0, 1 boundary time point is found from the marked time period, and is recorded as an idle behavior; continue to find the interactive traffic marked as 0 state record as the start of the next idle behavior. For example, the interactive traffic marked as 0 state record is shown in Table 2.

[0073] Table 2

[0074] Time series index Behavior start time Behavior end time Idle behavior sequence 25 2020-1-2 8:00:00 2020-1-2 8:05:00 1 205 2020-1-2 8:15:00 2020-1-2 8:23:00 2 265 2020-1-2 9:20:00 2020-1-2 10:00:00 3 … … … …

[0075] In Table 2, the time sequence subscript of the interactive traffic marked as 0, the behavior start time, the behavior end time, and the idle behavior order are recorded.

[0076] Next, a method for finding the optimal time interval threshold T is given. The determination method of the optimal time interval threshold is as follows:

[0077] Considering that the operation behaviors of different software can be quite different, for example, for drawing software, because a large number of operations need to be performed on the window, the interactive traffic will be larger, and the duration of the idle behavior is generally shorter. For computing software, after loading data and starting computing, there is basically no need to operate on the window. If the same threshold is set for different software, there will be a large error. Therefore, for the idle behavior discrimination threshold T, the threshold needs to be optimized according to the actual situation of the software, which can solve the threshold difference problem caused by different software types. The specific operation method is as follows:

[0078] Take Δt = 0.5min as the step, count the number of times of being judged as idle behavior at different time thresholds T i Next, the number of times of being judged as idle behavior is counted at different time thresholds T Figure 3The absolute values of the slopes of the line segments formed by the adjacent thresholds are calculated, and then the average of the absolute values of the slopes of the adjacent N (N≥3) line segments is given, to give a smoothness index S, and then the optimal threshold is given by the smoothness index. The specific operation method is as follows:

[0079] From 0.5 min, different thresholds are constructed according to a step of 0.5, such as 0.5 min, 1 min, 1.5 min, …, and so on. The number of times of idle condition of the statistical software is counted, as shown in Table 3 below:

[0080] Table 3

[0081] Threshold (min) 0.5 1 1.5 … 9.0 9.5 10 … Idle count 528 264 120 23 15 12 …

[0082] The above data is plotted into a line graph, as shown in FIG. 2. Figure 3 As can be seen from FIG. 2, the absolute values of the slopes of the curves in the graph gradually decrease from large to small, and then gradually increase from small to large, and finally decrease to 0. Among them, the absolute values of the slopes of the curves in the threshold range of 3.5-7.5 are larger than those in the threshold range of 2.5-3.5. Moreover, in a certain threshold interval, the absolute values of the slopes change relatively smoothly, which indicates that the habit of idle behavior tends to be stable, so taking the threshold in this interval can ensure that the normal operation of the user will not be mistakenly recognized as idle time, and idle time will not be missed.

[0083] Based on the above method of selecting the optimal threshold T, the absolute values of the slopes of the line segments formed by the adjacent thresholds are calculated, and then the average of the absolute values of the slopes of the adjacent N=5 line segments is taken as an index of whether the line is smooth, denoted as S. If there is and only one index S<5 (for example, when the index S is calculated by taking 5 adjacent line segments, 6 adjacent points are involved, that is, the data in Table 3 is grouped into 6 groups to complete the calculation, and each group calculates a specific S, which can be large or small, and there can be only one S<5, and the others are greater than 5), the smaller threshold of the third line segment corresponding to the index is taken as the candidate optimal threshold (N>=3, N is an odd number, the middle line segment is selected, N is an even number, the line segment with smaller slope is selected from the middle two line segments). If there are multiple candidate optimal thresholds, the minimum value is selected as the optimal threshold T. According to this method, each professional software will determine a different time threshold through calculation.

[0084] The above method has the advantages that it can efficiently identify non-normal idle time periods in a large amount of monitoring sample data set, so as to eliminate obviously normal use time periods and obtain a modeling data sample set.

[0085] b. Pseudo-idle behavior exclusion:

[0086] For the software that performs background calculation or because of network problems, the interactive traffic is small, which is misidentified as idle behavior by the previous step, and needs to be excluded by other data. The specific exclusion method is:

[0087] The CPU, GPU, memory, and disk I / O utilization of the software are monitored. The average utilization of the software CPU and GPU is more than 50%, and the memory range is greater than 15% of the memory occupied by the software during the idle behavior time period. The idle behavior is excluded.

[0088] S2. According to the sample data obtained above, determine the input indicators of the software and hardware resource waste identification model. In this embodiment, 17 input indicators are established, which are as follows:

[0089] Table 4

[0090]

[0091]

[0092] From the data analysis library, the newly generated data is cleaned by day, and the input indicator calculation is performed according to the contents in Table 4. After the calculation is completed, the feature selection is performed, the features with high correlation are removed, and then stored in the feature table. The data in the feature table is used for model training. The data in the feature table is shown in Table 5.

[0093] Table 5

[0094]

[0095] (Continued)

[0096]

[0097] For the samples in the above feature table, the attributes (i.e. input indicators) in columns 5 to 19 are used as the input of the BP neural network, and the result of the user's resource idle behavior (i.e. idle waste behavior) is used as the output. The BP neural network is used for modeling and training to obtain the software and hardware resource waste identification model.

[0098] The hidden layer of the identification model uses the Relu function as the activation function, and the output layer uses the sigmod function as the activation function. After training, the node number of the two hidden layers is 17 and 9 respectively, and the training effect is good. The identification model based on the BP neural network is shown in the accompanying Figure 4 .

[0099] In this embodiment, the extracted feature types are more, which can effectively improve the training accuracy of the model. Of course, if the training difficulty, training speed and time of the model training are considered, a few feature data can also be selected, for example, 1-2 time class features, 1-2 flow class features and 1-2 fluctuation class features.

[0100] The identification method of the application, by the change data of the traffic interaction between the platform remote desktop end and the cloud virtual machine, CPU, GPU, memory utilization, constructs a feature model suitable for user idle connection identification, that is, a software and hardware resource waste identification model, which is used for real-time monitoring of user waste of software and hardware resources to automatically identify, to solve the problem that the traditional method is not accurate in identifying user waste of software and hardware resources, is not friendly to users, and the optimization effect is not obvious; At the same time, the management of cloud platform professional software license resources and hardware resources is optimized, and the utilization rate of software and hardware resources is improved.

[0101] Method embodiment 2:

[0102] The embodiment provides an identification method for user idle waste of software and hardware resources of a cloud platform, and the method flow is as shown in Figure 5 The system is provided with a data analysis library and a business library, wherein the data analysis library is used for acquiring user monitoring data collected by the cloud platform business system, and performing data processing, including idle behavior preprocessing (for reference to step S1 in method embodiment 1), other feature construction (for reference to step S2 in method embodiment 1), etc., and outputting the processed feature data for BP neural network training of the cloud platform business system.

[0103] Figure 5 In order to further improve the identification accuracy of the identification module, the identification model needs to be optimized, for example, the business library is used for storing historical judgment data of the identification model, and the historical judgment data and the judgment result of the waste idle behavior this time are combined as sample data, and then the BP neural network is trained to obtain an optimized identification module.

[0104] Device embodiment:

[0105] The embodiment provides an identification device for user idle waste of software and hardware resources of a cloud platform, which comprises a memory and a processor, and a computer program stored in the memory and running on the processor, the processor is coupled with the memory, and the processor is used for running the program instructions stored in the memory to realize the identification method in method embodiments 1 or 2. Since the method is described in method instances 1 and 2, it is clear and complete, and the embodiment will not be repeated.

[0106] That is, the method in the above method embodiments should be understood as computer program instructions implementing the flow of the master / slave robot control method. These computer program instructions can be provided to a processor (such as a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices, etc.) so that the execution of these instructions by the processor produces the functions specified by the flow of the above method.

[0107] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application rather than limit it, although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that: the specific embodiments of the present application can still be modified or replaced by equivalents without departing from the spirit and scope of the present application, any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered within the protection scope of the claims of the present application.

Claims

1. A method for identifying idle and waste hardware and software resource behaviors of a cloud platform user, characterized in that, Comprising the following steps: Step one, obtaining the real-time data of the cloud platform user, including the monitoring data of the user login on the platform and the user opening the virtual machine, opening the software, license request, virtual machine runtime resource occupation, client and cloud traffic interaction, extracting the feature data in the real-time data; The feature data includes software opening duration and effective limit duration, total idle duration, effective idle duration ratio, idle times, total interaction traffic, average interaction traffic, interaction traffic fluctuation, CPU utilization fluctuation, GPU utilization fluctuation, memory utilization fluctuation, disk I / O utilization fluctuation, GPU utilization average, disk I / O speed average; Step two, input the feature data of the user as input into the trained software and hardware resource waste identification model to obtain the behavior identification result of the user; The software and hardware resource waste identification model is obtained by inputting the feature data marked with idle waste behavior into the neural network for training; The determination steps of the feature data marked with idle waste behavior are as follows: (1) Obtain the historical data of each monitoring data and extract the attribute data therefrom; The attribute data includes user and software type, monitoring time, whether the user is online on the platform, whether the user opens the application window, CPU, GPU, memory, disk I / O of the hardware resource where the user is located, CPU occupied by the software, GPU occupied by the software, memory occupied by the software, disk I / O occupied by the software, interaction traffic, opening duration, system type, whether the license is occupied; (2) Clean the attribute data; (3) Sample the data, construct sample data, divide the sample data into idle time period and non-idle time period according to the size and duration of the interaction traffic, and mark the idle behavior and non-idle behavior according to the divided idle time period and non-idle time period. 2.The cloud platform user idle waste hardware and software resource behavior identification method according to claim 1, characterized in that, The cleaning of the attribute data in step (2) includes filling the missing values, removing the duplicate data, and reducing the redundant attributes. 3.The cloud platform user idle waste hardware and software resource behavior identification method according to claim 1, characterized in that, In step (3), the result division of the idle time period and the non-idle time period includes the following steps: Statistical user interaction traffic number in each set time, divide the time interval, determine the occupation frequency of each user interaction traffic number, that is, the ratio of the number of time intervals to which the interaction traffic belongs to the total number of traffic statistics; According to the distribution of the occupation frequency of the user interaction traffic number, determine the threshold P, for the time period with interaction traffic number less than P, mark it as an idle time period; For the time period with interaction traffic number not less than P, mark it as a non-idle time period.

4. The cloud platform user idle and waste hardware and software resource behavior identification method of claim 3, wherein, Further comprising: When at least two consecutive time periods are marked as idle time periods, and the time interval of these marking records exceeds the set time threshold T, find out the dividing time points marked as idle time period and non-idle time period from the current marking time period, and record it as an idle behavior, until all the idle behaviors are marked.

5. The method of claim 4, wherein the cloud platform user idle waste of hardware and software resources behavior is identified by, The determination steps of the time threshold T include: Taking Δt=0.5 min as a step, the number of times of idle behavior determined at different time thresholds Ti is counted; a broken line graph is drawn with the time threshold Ti and the number of times of idle behavior as the horizontal and vertical coordinates, the absolute value of the slope of the line segment formed by adjacent thresholds is calculated, and the average value of the absolute values of the slopes of the adjacent N line segments is taken as a stability index S, and the optimal threshold T is determined through the stability index S, and N≥3.

6. A device for identifying idle and wasted hardware and software resources behavior of a cloud platform user, comprising a memory and a processor, and a computer program stored on the memory and running on the processor, the processor being coupled to the memory, characterized in that, The processor implements the method for cloud platform user idle waste of software and hardware resources behavior according to any one of claims 1-5 when executing the computer program.

Images