Target node selection method based on time sequence prediction model

Abstract

The invention relates to the technical field of cloud computing, and discloses a target node selection method based on a time sequence prediction model, and the method comprises the steps: obtaining historical monitoring data of a computing node; a Prophet time sequence prediction model is constructed; the Prophet time sequence prediction model comprises a trend sub-model, a periodicity sub-model and an irregular emergency sub-model; fitting the Prophet time sequence prediction model based on historical monitoring data to obtain a load prediction model and a migration cost prediction model; predicting the load condition of the next period based on the load prediction model to obtain a load prediction value; predicting the migration cost of the next period based on the migration cost prediction model to obtain a migration cost prediction value; and determining a migrated target node based on the load prediction value and the migration cost prediction value. According to the invention, the accuracy and efficiency of cloud resource prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing technology, and in particular to a target node selection method based on a time series prediction model. Background Art

[0002] As cloud computing technology continues to develop, many algorithms have been developed to allocate cloud resources flexibly and scalably. However, with the rapid development of cloud computing today, the scale of cloud resources is constantly increasing, and the allocation of cloud resources has become more complicated. From the demand side, the load and scale of the business are constantly changing; from the supply side, the resource utilization rate, fragmentation rate, and number of users are also constantly changing. At present, some methods that provide resources greater than the user's demand under high availability conditions can meet the changing load of any user. In the context of the current growing scale of cloud resources, a large amount of cloud resources will be wasted and lost, and the utility is low.

[0003] Related technologies have proposed some solutions to reduce losses, such as intelligently shutting down idle hosts in the platform, optimizing scheduling methods to effectively concentrate cloud resources, distinguishing resource states according to load conditions and optimizing switching states, etc. However, in the face of ever-changing supply and demand conditions, the above loss reduction solutions do not take into account on-demand scaling, and cannot flexibly schedule cloud resources, which is inefficient. In addition, the current resource scheduling strategy is often provided passively, and the load of cloud platform resources in complex scenarios cannot be predicted with insufficient accuracy. Summary of the invention

[0004] In view of this, the present invention provides a target node selection method based on a time series prediction model to improve the accuracy and efficiency of cloud resource prediction.

[0005] In a first aspect, the present invention provides a target node selection method based on a Prophet time series prediction model, the method comprising: obtaining historical monitoring data of computing nodes; constructing a Prophet time series prediction model; the Prophet time series prediction model comprises a trend sub-model, a periodic sub-model and an irregular emergency event sub-model; fitting the Prophet time series prediction model based on the historical monitoring data to obtain a load prediction model and a migration cost prediction model; predicting the load situation of the next period based on the load prediction model to obtain a load prediction value; predicting the migration cost of the next period based on the migration cost prediction model to obtain a migration cost prediction value; and determining the target node for migration based on the load prediction value and the migration cost prediction value.

[0006] In an optional implementation, obtaining historical monitoring data of a computing node includes: collecting monitoring data of the computing node according to a preset period; the monitoring data includes at least CPU usage and memory usage; and processing the monitoring data to obtain historical monitoring data.

[0007] In an optional embodiment, constructing a trend sub-model includes: determining the carrying capacity, the initial growth rate and the first offset parameter; identifying the change points of the historical monitoring data and defining a growth rate function that changes over time; adjusting the first offset parameter based on the change points to obtain the second offset parameter; constructing the trend sub-model based on the carrying capacity, the initial growth rate and the second offset parameter.

[0008] In an optional implementation, constructing a periodic sub-model includes: identifying periodic features of historical monitoring data; and constructing a periodic sub-model based on the periodic features.

[0009] In an optional implementation, constructing the irregular emergency event sub-model includes: identifying irregular emergency events in historical monitoring data; extracting event features of the irregular emergency events; and constructing the irregular emergency event sub-model based on the features.

[0010] In an optional implementation, the method further includes: collecting monitoring data for the next period and updating the Prophet time series prediction model.

[0011] In a second aspect, the present invention provides a target node selection device based on a Prophet time series prediction model, the device comprising: a data acquisition module for acquiring historical monitoring data of computing nodes; a model construction module for constructing a Prophet time series prediction model; the Prophet time series prediction model comprises a trend sub-model, a periodic sub-model and an irregular emergency event sub-model; a model fitting module for fitting the Prophet time series prediction model based on historical monitoring data to obtain a load prediction model and a migration cost prediction model; a load prediction module for predicting the load situation of the next period based on the load prediction model to obtain a load prediction value; a migration prediction module for predicting the migration cost of the next period based on the migration cost prediction model to obtain a migration cost prediction value; and a node determination module for determining the target node for migration based on the load prediction value and the migration cost prediction value.

[0012] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to execute a target node selection method based on a Prophet time series prediction model according to the first aspect or any corresponding embodiment thereof.

[0013] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions are used to enable a computer to execute a target node selection method based on a Prophet time series prediction model according to the first aspect or any corresponding embodiment thereof.

[0014] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, wherein the computer instructions are used to enable a computer to execute a target node selection method based on a Prophet time series prediction model according to the first aspect or any corresponding embodiment thereof.

[0015] The technical solution provided by this application may have the following beneficial effects:

[0016] The present invention provides a target node selection method based on a Prophet time series prediction model. The historical monitoring data of the computing node is obtained to provide data support for the subsequent time series prediction model, reflecting the actual usage and change trend of cloud resources. A Prophet time series prediction model is constructed. The Prophet time series prediction model includes a trend sub-model, a periodic sub-model and an irregular emergency event sub-model, which respectively capture different characteristics in the time series. The trend sub-model represents the change trend of the historical monitoring data over time, the periodic sub-model represents the periodic change of the historical monitoring data, and the irregular emergency event sub-model represents the sudden and irregular change. The Prophet time series prediction model is fitted based on the historical monitoring data to obtain a load prediction model and a migration cost prediction model, so as to realize the prediction of the cloud resource load and migration cost, and provide a basis for the subsequent target node selection. The load situation of the next cycle is predicted based on the load prediction model to obtain a load prediction value; the migration cost of the next cycle is predicted based on the migration cost prediction model to obtain a migration cost prediction value; the target node for migration is determined based on the load prediction value and the migration cost prediction value.

[0017] The above solution improves the accuracy and efficiency of cloud resource prediction by obtaining historical monitoring data of computing nodes, building a Prophet time series prediction model, fitting the model, obtaining a load prediction model and a migration cost prediction model, and finally determining the target node for migration based on the predicted value. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flowchart of a target node selection method based on a Prophet time series prediction model according to an embodiment of the present invention;

[0020] Figure 2 Schematic diagram of the composition of the Prophet time series prediction model according to an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of the composition of a load prediction model according to an embodiment of the present invention;

[0022] Figure 4 is a schematic diagram of the composition of a migration cost prediction model according to an embodiment of the present invention;

[0023] Figure 5 is a system architecture diagram according to an exemplary embodiment;

[0024] Figure 6 is a flow chart of a target node selection method according to another embodiment of the present invention;

[0025] Figure 7 is a structural block diagram of a target node selection device based on a Prophet time series prediction model according to an embodiment of the present invention;

[0026] Figure 8 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0028] According to an embodiment of the present invention, an embodiment of a target node selection method based on a Prophet time series prediction model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0029] In this embodiment, a target node selection method based on the Prophet time series prediction model is provided. Figure 1is a flow chart of a target node selection method based on a Prophet time series prediction model according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0030] Step S101, obtaining historical monitoring data of computing nodes.

[0031] By setting corresponding monitoring items (such as CPU, memory, etc.), all computing nodes in the available zone of the cloud resource pool are monitored. Historical monitoring data is time series data, including the resource usage of each computing node in the past period of time, such as CPU usage, memory usage and other resource usage indicators.

[0032] Step S102, constructing a Prophet time series prediction model; the Prophet time series prediction model includes a trend sub-model, a periodicity sub-model and an irregular emergency event sub-model.

[0033] The Prophet model is a time series forecasting tool that decomposes the time series into four parts: trend, cycle, irregular sudden events, and error terms, allowing the model to more flexibly respond to complex and changing data. The trend sub-model represents the changing trend of historical monitoring data over time, the periodic sub-model represents the periodic changes of historical monitoring data, and the irregular sudden event sub-model represents sudden and irregular changes.

[0034] Step S103, fitting the Prophet time series prediction model based on historical monitoring data to obtain a load prediction model and a migration cost prediction model.

[0035] The historical monitoring data is input into the Prophet time series prediction model, and the model is fitted to obtain the load prediction model and the migration cost prediction model. The load prediction model can predict the load of cloud resources in the future based on the historical monitoring data. The load prediction model considers the impact of factors such as trends, periodicity, and irregular emergencies on the load, thereby providing more accurate prediction results. The migration cost prediction model can predict the cost required to migrate a cloud host from one computing node to another, so as to select a suitable migration target node.

[0036] In an embodiment, Figure 2 As shown, the influencing factors mentioned in step S102: trend, cycle and irregular emergencies are accumulated to obtain a time series prediction model:

[0037] y(t)=g(t)+s(t)+h(t)+∈

[0038] Among them, ∈ is the error term, g(t), s(t), and h(t) are the trend sub-model, periodic sub-model, and irregular emergency event sub-model, respectively. The function is optimally fitted by using the L-BFGS method.

[0039] In order to obtain the load prediction model and migration cost prediction model respectively, as Figure 3 As shown in the figure, by fitting the model using the historical monitoring data of the computing nodes, we get the prediction models for CPU, memory and other monitoring data respectively, and then we set the weight values ​​empirically to get the final load prediction model:

[0040] y load (t) = λ1y cpu (t)+λ2y ram (t)+λ3y other (t)

[0041] like Figure 4 As shown in Figure 2, the same model is fitted with the migration cost score data of each node to obtain the node migration cost prediction model Y cost (t). The difference is that the data that changes over time is a vector Score cost = {CS1, CS2, ..., CS n}.

[0042] Step S104: predict the load condition of the next cycle based on the load prediction model to obtain a load prediction value.

[0043] The parameters used to predict the load are input into the load prediction model. The parameters include a timestamp (indicating the predicted time point or time period) and other external factors that may affect the load. In this embodiment, the load situation of the next time period is predicted, so the timestamp will point to the start time of the next period. After the model is calculated, the load prediction value of the next period is obtained, which indicates the predicted level of cloud resource load at the predicted time point or time period, reflecting the trend of load change. For example, the load prediction model predicts the next period T n+1 Load condition at time y load (T n+1 ).

[0044] Step S105 , predicting the migration cost of the next period based on the migration cost prediction model to obtain a migration cost prediction value.

[0045] First, the specific conditions for migration cost prediction are clarified, including the predicted time point or time period (i.e., the next cycle), the cloud resource attributes to be migrated (such as CPU, memory, etc.), and the characteristics of the target node. The predicted conditions are input into the migration cost prediction model, and the model will calculate the predicted cost required for resource migration in the next cycle under given conditions based on the rules and parameters learned internally. After the model calculation, a migration cost prediction value is obtained, which represents the estimated cost required to migrate cloud resources from one node to another at the predicted time point or time period.

[0046] In this embodiment, the migration cost model is used to predict the next period T n+1 The migration cost Y cost (T n+1 ), the values ​​in the migration cost vector are arranged in ascending order to obtain an ordered sequence

[0047] Step S106: determining a target node for migration based on the load prediction value and the migration cost prediction value.

[0048] The above steps have obtained the load forecast value and migration cost forecast value of the next cycle through the load forecast model and migration cost forecast model respectively. These forecast values ​​are calculated based on historical monitoring data and the Prophet time series forecast model, representing the future cloud resource usage and migration cost.

[0049] Analyze the load prediction value and migration cost prediction value to obtain the analysis results. Formulate a migration strategy based on the analysis results. It is necessary to consider multiple factors, including load balancing, migration cost, special attributes, etc. Specifically, select nodes with lower load as migration targets to balance the load pressure of each node and avoid overload. Select nodes with lower migration cost as migration targets to reduce resource loss and service interruption risks during the migration process. Consider the special attributes of the cloud host (such as affinity, NUMA, etc.) to ensure that the migrated cloud host can meet the specific operating environment and performance requirements.

[0050] In this embodiment, in order to complete the cloud host migration attributes, meet special attributes (such as affinity, NUMA, etc.), and calculate the node resources not more than the target node quota multiplied by the over-allocation ratio, the migration cost score required for the maximum case is calculated as Score max-cost ,right Filter and keep the values ​​greater than Score max-cost Finally, select from the screening results from small to large.

[0051] In summary, this embodiment improves the accuracy and efficiency of cloud resource prediction by acquiring historical monitoring data of computing nodes, building a Prophet time series prediction model, fitting the model, obtaining a load prediction model and a migration cost prediction model, and finally determining the target node for migration based on the predicted value.

[0052] In an optional implementation, the above step S101 includes the following steps:

[0053] Step S1011, collecting monitoring data of computing nodes according to a preset period; the monitoring data at least includes CPU usage and memory usage.

[0054] Monitoring data is collected from each computing node in the cloud resource pool according to the preset time period. Monitoring data is the basis for subsequent analysis and prediction to ensure the accuracy of the prediction model.

[0055] The preset period is a pre-set time interval that determines the frequency of data collection. For example, the preset period can be set to every hour, every half hour, or every 15 minutes, etc., depending on the system's real-time requirements. The monitoring data includes at least resource usage indicators such as CPU usage and memory usage. Optionally, the monitoring data can also include disk I / O and network bandwidth. These indicators can fully reflect the resource usage and performance status of the computing nodes. According to the preset period, the corresponding monitoring data is automatically collected from each computing node.

[0056] Step S1012, processing the monitoring data to obtain historical monitoring data.

[0057] The data processing process can include multiple steps such as data cleaning, data conversion, and data normalization. Data cleaning aims to remove invalid or abnormal data points, such as missing values, duplicate values, or obviously erroneous data. Data conversion may involve converting data in different formats or units into a unified standard format for comparison and analysis. Data normalization is a process of scaling data so that it falls into a small specific interval (such as 0 to 1), which helps to eliminate the dimensional effects between different indicators and improve the predictive performance of the model. After data processing, the data obtained is historical monitoring data.

[0058] In an embodiment of the present invention, each attribute value of the cloud host type that allows migration of the cloud host in the historical monitoring data, such as CPU, memory and other specification attributes, is normalized and multiplied by its weight value, and the sum is obtained to obtain the resource allocation score (Allow Score, AS) of the cloud host, and the expression is as follows:

[0059]

[0060] Among them, AS flavorIndicates the value of the cloud host specification attribute w flavor is the weight of the attribute value, and the sum of the attribute values ​​of CPU and memory is equal to 1.

[0061] In historical monitoring, N cycles of data of the cloud host can be migrated. Based on this data, the actual load value of the cloud host is calculated. The historical data of the monitoring indicator is normalized and multiplied by the weight value. The sum is used to obtain the actual load score (Use Score, US) of the cloud host. The expression is as follows:

[0062]

[0063] Among them, US metric Indicates the actual load value of the cloud host's CPU and memory attributes w metric is the weight of the attribute value, and the sum of the actual load values ​​of the CPU and memory attributes is equal to 1.

[0064] The migration cost score (CS) of the cloud host is set as the sum of the resource allocation score and the actual load value:

[0065] CS=AS+US

[0066] The migration cost of the cloud host is also monitored for N cycles. After monitoring for N cycles, the time and historical monitoring data of the corresponding monitoring items are obtained from the monitor.

[0067] In an optional implementation, the process of constructing the trend sub-model in step S102 includes the following steps:

[0068] Step a11, determining the carrying capacity, the initial growth rate and the first offset parameter.

[0069] The carrying capacity represents the maximum value that the time series data can reach, that is, the upper limit of the system or resources. In this embodiment, it may represent the maximum value of indicators such as CPU utilization and memory utilization. The initial growth rate represents the growth rate of the time series data in the initial stage, reflecting the initial change trend of resource utilization. The first offset parameter is a constant term in the trend submodel, which is used to adjust the fit between the model and the actual data, and represents the offset of the time series data at the initial moment.

[0070] Step a12, identifying the change points of the historical monitoring data and defining a growth rate function that changes over time.

[0071] Time series data will mutate at certain moments, and these mutation points are called change points. Identifying change points and defining a growth rate function that changes over time can more accurately capture nonlinear trends in the data. Change points in historical monitoring data are identified through algorithms (such as the change point detection algorithm that comes with the Prophet model). After identifying the change points, a growth rate function that changes over time needs to be defined.

[0072] Step a13: adjusting the first offset parameter based on the change point to obtain a second offset parameter.

[0073] According to the position of the change point and the change of the growth rate function, the first offset parameter is adjusted to obtain the second offset parameter.

[0074] Step a14, constructing a trend sub-model based on the carrying capacity, the initial growth rate and the second offset parameter.

[0075] Substituting the above parameters into the mathematical expression of the trend sub-model, we get the final trend sub-model. The trend sub-model can capture the long-term trend and mutations at the change points in the time series data, providing a basis for subsequent analysis and prediction.

[0076] In this embodiment, a trend sub-model is constructed. For the monitoring items of cloud resources, a saturation growth model is selected, which is as follows:

[0077]

[0078] Among them, C is the carrying capacity, k is the growth rate, and m is the offset parameter.

[0079] In the actual cloud resource usage scenario, the above monitoring indicators are not static, the time series trend is also changing all the time, and some special changes occur at a specific moment, which requires the use of change point detection. Assume there are s change points, and the change point timestamp is s j , each change point will have a growth rate change δ j . Assuming the initial growth rate is k0, the growth rate over time is:

[0080]

[0081] where a j (t) is the indicator function:

[0082]

[0083] After the growth rate is determined, the function m(t) of the offset parameter m over time must also be adjusted accordingly to achieve the boundary point necklace. The offset adjustment at the change point yields:

[0084] m(t)=m0+a j (t) T γ

[0085] Among them, γ is the segment boundary:

[0086]

[0087] After converting the original invariant terms k and m in the formula into the above-mentioned function considering the time-varying characteristics, the trend model obtained is as follows:

[0088]

[0089] Among them, C(t) needs to be adjusted accordingly according to the scale of cloud resources.

[0090] In an optional implementation, the process of constructing the periodic sub-model in step S102 includes the following steps:

[0091] Step b11, identifying the periodic characteristics of historical monitoring data.

[0092] Periodicity may be manifested as regular fluctuations in resource usage over time periods such as days, weeks, months, or years. For example, some services may have lower loads on weekends or holidays, but higher loads during peak hours on weekdays.

[0093] Step b12, constructing a periodic sub-model based on the periodic characteristics.

[0094] The periodic submodel is used to identify periodic fluctuations in time series data, so as to more accurately predict future data changes. Mathematical tools such as Fourier series are used to construct periodic submodels based on the identified periodic features. The time period P is added as a parameter in the model to reflect the impact of different time periods on the data. By adjusting the parameters N and β of the Fourier series, the model can better fit the periodic fluctuations in historical monitoring data.

[0095] In this embodiment, a periodic sub-model is constructed to take time series into account. Cloud resource usage scenarios generally change over time, such as days, weeks, months, years, etc. The expression constructed using Fourier series is as follows:

[0096]

[0097] Where O is the time period in days, N is based on experience, N is 10 for a year and 3 for a week, and the parameter of the Fourier series is β = [a1, b1, ..., a N , b N ] T , use this parameter to construct a periodic vector matrix, that is, s(t) = X(t)β. And initialize the parameter β: Normal(0,σ 2), σ is a variable. The larger the value, the stronger the seasonal effect; on the contrary, the smaller the value, the weaker the seasonal periodicity.

[0098] In an optional implementation, the process of constructing the irregular emergency event sub-model in step S102 includes the following steps:

[0099] Step c11, identifying irregular emergencies in historical monitoring data.

[0100] Analyze historical monitoring data to identify data points that significantly deviate from normal trends, representing the occurrence of irregular emergencies.

[0101] Step c12, extracting event features of irregular sudden events.

[0102] Event characteristics may include the time of occurrence, duration, scope of impact (such as which resources are affected), degree of impact (such as the fluctuation range of resource utilization), etc.

[0103] Step c13, constructing an irregular emergency event sub-model based on the features.

[0104] Use machine learning or statistical methods to build an irregular emergency event sub-model based on the extracted event features. Add parameters that can reflect event characteristics, such as the probability of event occurrence, impact, etc. to the model. By training and adjusting the model parameters, it can accurately predict irregular emergencies that may occur in the future.

[0105] In this embodiment, irregular emergencies in the past are used to predict irregular emergencies in the future. The model of irregular emergencies h(t)=Z(t)k is adopted. The parameter k is initialized to Normal(0, v 2 ). v is a variable. Similarly, the larger the value, the stronger the irregular sudden effect; on the contrary, the smaller the value, the weaker the irregular sudden effect.

[0106] In an optional embodiment, the method further includes:

[0107] Collect monitoring data for the next period and update the Prophet time series prediction model.

[0108] After determining the target node for migration, select the corresponding target node for migration, and record the monitoring data of the next cycle again, and pass it to the collector in the prediction system as input. Perform data fitting again before the next cycle arrives to continuously improve the model accuracy.

[0109] Figure 5 is a system architecture diagram according to an exemplary embodiment, a target node selection method based on a Prophet time series prediction model of the present invention is applied to the system, such as Figure 5 As shown in the figure, the collector obtains data from the cloud platform and then transmits the data to the distributor. The distributor transmits data such as time, CPU and RAM to the historical monitoring database. The system also includes a load condition module, a migration cost module and a target node selection module. In order to analyze historical data, the system will store historical data in the database. Through the historical migration cost module and historical error term analysis, the system can evaluate the migration cost of historical data and the accuracy and reliability of the data. The entire data processing and monitoring system is uniformly managed and scheduled by the scheduling system.

[0110] The following is another embodiment of the present invention. Figure 6 As shown,

[0111] First, prepare a cloud platform with at least 20 computing nodes. Create multiple cloud hosts, evenly distribute them on each node, and simulate the usual cloud platform resource allocation (ensuring an allocation rate of about 80%). Perform business simulations on each cloud host to ensure the usage of cloud platform resources. Set corresponding tasks, create and delete cloud hosts at random times, simulate the usage of cloud platform customers, set dynamic resource scheduling methods (centralized or balanced), adjust the resources on the computing nodes accordingly, trigger corresponding alarms, and collect N cycles of monitoring data that change over time.

[0112] Secondly, the monitoring data of N cycles are fitted according to the load prediction model and the migration cost prediction model of the present invention.

[0113] Finally, the fitted time series prediction model is added to make corresponding predictions for the load and migration cost of the next cycle. The corresponding nodes are selected through the set target node selection method, and hot migration operations are performed. After the migration is completed, the monitoring information of the cycle is passed to the collector of the model, and the model is fitted again to continuously improve the accuracy.

[0114] In this embodiment, a target node selection device based on the Prophet time series prediction model is also provided, which is used to implement the above embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0115] This embodiment provides a target node selection device based on the Prophet time series prediction model. Figure 7 As shown, including:

[0116] Data acquisition module 701, used to acquire historical monitoring data of computing nodes;

[0117] Model building module 702, used to build a Prophet time series prediction model; the Prophet time series prediction model includes a trend sub-model, a periodicity sub-model and an irregular emergency event sub-model;

[0118] A model fitting module 703 is used to fit the Prophet time series prediction model based on historical monitoring data to obtain a load prediction model and a migration cost prediction model;

[0119] The load prediction module 704 is used to predict the load condition of the next cycle based on the load prediction model to obtain a load prediction value;

[0120] A migration prediction module 705 is used to predict the migration cost of the next period based on the migration cost prediction model to obtain a migration cost prediction value;

[0121] The node determination module 706 is used to determine the target node for migration based on the load prediction value and the migration cost prediction value.

[0122] In an optional implementation, the data acquisition module 701 includes:

[0123] A first data acquisition unit is used to collect monitoring data of the computing node according to a preset period; the monitoring data includes at least CPU usage and memory usage;

[0124] The first data acquisition unit is used to process the monitoring data to obtain historical monitoring data.

[0125] In an optional implementation, the model building module 702 includes:

[0126] a first trend sub-model unit for determining carrying capacity, initial growth rate, and first offset parameter;

[0127] The second trend sub-model unit is used to identify the change points of the historical monitoring data and define the growth rate function that changes over time;

[0128] A third trend sub-model unit is used to adjust the first offset parameter based on the change point to obtain a second offset parameter;

[0129] The fourth trend sub-model unit is used to construct a trend sub-model based on the carrying capacity, the initial growth rate and the second offset parameter.

[0130] In an optional implementation, the model building module 702 further includes:

[0131] A first periodic sub-model unit, used to identify periodic features of historical monitoring data;

[0132] The second periodic sub-model unit is used to construct a periodic sub-model based on the periodic characteristics.

[0133] In an optional implementation, the model building module 702 further includes:

[0134] A first irregular emergency event sub-model unit is used to identify irregular emergency events in historical monitoring data;

[0135] The second irregular emergency event sub-model unit is used to extract event features of irregular emergency events;

[0136] The third irregular emergency event sub-model unit is used to construct an irregular emergency event sub-model based on the features.

[0137] In an optional embodiment, the device further comprises:

[0138] The model updating unit is used to collect monitoring data for the next period and update the Prophet time series prediction model.

[0139] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0140] In this embodiment, a target node selection device based on the Prophet time series prediction model is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0141] The embodiment of the present invention also provides a computer device having the above Figure 7 A target node selection device based on the Prophet time series prediction model is shown.

[0142] See also Figure 8 , Figure 8 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 8As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 8 A processor 10 is taken as an example.

[0143] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0144] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0145] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0146] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0147] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 8 The example of connecting through bus is taken in the following.

[0148] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0149] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0150] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0151] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A target node selection method based on the Prophet time series prediction model, characterized in that: The method comprises: Get historical monitoring data of computing nodes; Constructing a Prophet time series prediction model; the Prophet time series prediction model includes a trend sub-model, a periodicity sub-model and an irregular emergency event sub-model; Fitting the Prophet time series prediction model based on the historical monitoring data to obtain a load prediction model and a migration cost prediction model; Predicting the load situation of the next cycle based on the load prediction model to obtain a load prediction value; Predicting the migration cost of the next cycle based on the migration cost prediction model to obtain a migration cost prediction value; A target node for migration is determined based on the load prediction value and the migration cost prediction value.

2. The method according to claim 1, characterized in that The obtaining of historical monitoring data of the computing node includes: Collect monitoring data of the computing node according to a preset period; the monitoring data includes at least CPU usage and memory usage; The monitoring data is processed to obtain the historical monitoring data.

3. The method according to claim 1, characterized in that Constructing the trend sub-model includes: Determine the carrying capacity, initial growth rate and first deflection parameters; Identifying change points of the historical monitoring data and defining a growth rate function that changes over time; Adjust the first offset parameter based on the change point to obtain a second offset parameter; The trend sub-model is constructed based on the carrying capacity, the initial growth rate and the second offset parameter.

4. The method according to claim 1, characterized in that: Constructing the periodic sub-model includes: Identifying periodic characteristics of the historical monitoring data; Based on the periodic characteristics, a periodic sub-model is constructed.

5. The method according to claim 1, characterized in that: Constructing the irregular emergency event sub-model includes: Identifying irregular emergencies in the historical monitoring data; Extracting event features of the irregular emergency events; Based on the characteristics, the irregular emergency event sub-model is constructed.

6. The method according to claim 1, characterized in that The method further comprises: The monitoring data for the next period is collected and the Prophet time series prediction model is updated.

7. A target node selection device based on the Prophet time series prediction model, characterized in that: The device comprises: A data acquisition module is used to obtain historical monitoring data of computing nodes; A model building module is used to build a Prophet time series prediction model; the Prophet time series prediction model includes a trend sub-model, a periodicity sub-model and an irregular emergency event sub-model; A model fitting module, used to fit the Prophet time series prediction model based on the historical monitoring data to obtain a load prediction model and a migration cost prediction model; A load prediction module, used to predict the load condition of the next cycle based on the load prediction model to obtain a load prediction value; A migration prediction module, used to predict the migration cost of the next period based on the migration cost prediction model to obtain a migration cost prediction value; The node determination module is used to determine the target node of the migration based on the load prediction value and the migration cost prediction value.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the method according to any one of claims 1 to 6.

Images