Dynamic component loading optimization method

Abstract

The invention relates to the technical field of component loading optimization, and discloses a dynamic component loading optimization method, which comprises the following steps: collecting and identifying current real-time operation data to obtain a corresponding prediction demand category and an actual operation state; setting a plurality of component loading nodes based on the predicted demand category, generating a first component loading strategy according to all the component loading nodes, performing analogue simulation according to the first component loading strategy, and calculating a simulation demand coefficient and a simulation state coefficient according to a simulation result; calculating a simulation application coefficient of the first component loading strategy according to the simulation demand coefficient and the simulation state coefficient, and judging whether the first component loading strategy is optimized or not, if not, issuing a loading instruction according to the first component loading strategy, and if yes, generating a second component loading strategy, the component loading strategy is dynamically adjusted according to the requirement during operation, the user requirement is met, stable operation is achieved, and meanwhile the system response speed is increased.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of component loading optimization, in particular to a dynamic component loading optimization method. BACKGROUND

[0002] In the current process of enterprise digital transformation, the development demand for applications is increasing, and the existing application development platform can provide certain application construction capabilities, but there are obvious deficiencies in component loading and resource management: the traditional component loading method is mostly static loading, that is, all required components are loaded at one time when the application starts, regardless of whether the component is used in the current running scenario, which leads to slow application startup speed and high memory occupation, seriously affecting the system response speed; and lacks dynamic perception and adaptation ability to runtime demand, cannot dynamically adjust the component loading strategy, and is difficult to realize precise allocation and efficient use of component resources, thereby reducing the application development efficiency, increasing the operation and maintenance cost, and restricting the promotion speed of enterprise digital transformation. SUMMARY

[0003] To solve the above technical problems, the present application provides a dynamic component loading optimization method, which generates a first component loading strategy and performs simulation by identifying and predicting demand categories and actual running states, obtains a simulation application coefficient, judges whether to optimize according to the simulation application coefficient, dynamically adjusts the component loading strategy according to the runtime demand, meets the user demand and stable operation, and improves the system response speed.

[0004] In some embodiments of the present application, a dynamic component loading optimization method is provided, comprising: Collecting current real-time running data and identifying to obtain corresponding predicted demand categories and actual running states; Based on the predicted demand categories, a plurality of component loading nodes are set, and a first component loading strategy is generated according to all component loading nodes, simulation is performed according to the first component loading strategy, and a simulation demand coefficient and a simulation state coefficient are calculated according to the simulation result; According to the simulation demand coefficient and the simulation state coefficient, a simulation application coefficient of the first component loading strategy is calculated, and it is judged whether to optimize the first component loading strategy, if not, the loading instruction is issued according to the first component loading strategy, if yes, a second component loading strategy is generated.

[0005] In some embodiments of the present application, the corresponding predicted demand categories and actual running states are obtained, comprising: The current real-time running data is classified by type, and resource data set, behavior data set, state data set and usage data set are constructed according to the classification result; generate a first predicted resource demand and a corresponding first predicted confidence level according to the resource dataset and a pre-constructed dynamic load prediction model; generate a second predicted running demand and a corresponding second predicted confidence level according to the behavior dataset and a pre-constructed user intention prediction model; generate a third predicted running demand and a corresponding third predicted confidence level according to the usage dataset and a pre-constructed component dependency graph; determine a predicted demand type and a corresponding comprehensive predicted confidence level based on the first predicted running demand, the second predicted running demand and the third predicted running demand; obtain a historical running data package of the predicted demand type and generate a running evaluation value of the predicted demand type; set a component loading node number of the predicted demand type according to the running evaluation value; compare each real-time running data in the resource dataset with a corresponding preset resource data interval and generate an actual resource state according to a comparison result; compare each real-time running data in the state dataset with a corresponding preset state data interval and generate an actual business state according to a comparison result; generate an actual running state according to the actual resource state and the actual business state.

[0006] In some embodiments of the present application, a plurality of running demand evaluation indexes are pre-set; evaluate and analyze historical running data packages of the predicted demand type based on the plurality of running demand evaluation indexes to obtain running sub-evaluation values of the plurality of running demand evaluation indexes; generate a running evaluation value according to the running sub-evaluation values of the plurality of running demand evaluation indexes and weight coefficients of the corresponding running demand evaluation indexes; The calculation formula of the running evaluation value is: ; wherein Y1 is the running evaluation value, n is the running demand evaluation index, pi is the running sub-evaluation value of the i th running demand evaluation index, and ai is the weight coefficient of the i th running demand evaluation index; quantize each comparison result to obtain a plurality of first quantization values in the resource dataset and a plurality of second quantization values in the state dataset, wherein each first quantization value corresponds to a real-time running data in the resource dataset, and each second quantization value corresponds to a real-time running data in the state dataset; obtain the actual resource state according to the plurality of first quantization values and obtain the actual business state according to the plurality of second quantization values; The calculation formula of the actual running state is: ; wherein Y2 is an actual running state, z1 is a first conversion coefficient, b1 is a weight coefficient of an actual resource state, m1 is a number of real-time running data in a resource data set, is a first quantization value, is a weight coefficient of the first quantization value, z2 is a second conversion coefficient, b2 is a weight coefficient of an actual service state, m2 is a number of real-time running data in a state data set, is a second quantization value, is a weight coefficient of the second quantization value.

[0007] In some embodiments of the present application, the first component loading strategy is generated according to all component loading nodes, comprising: generating a component reference library of a predicted demand category, wherein the component reference library comprises a plurality of to-be-loaded components sorted according to demand time; performing timing analysis based on the sorting result of the to-be-loaded components in the component reference library and a plurality of component loading nodes of the predicted demand category to obtain a plurality of loading sub-strategies of each component loading node, and generating a plurality of pending component loading strategies; performing load prediction according to the loading sub-strategy of each component loading node in each pending component loading strategy to obtain a predicted load value at the corresponding component loading node; sequentially constructing a predicted load value sequence corresponding to each pending component loading strategy, and calculating a first loading coefficient of the corresponding pending component loading strategy according to the predicted load value sequence; wherein each loading sub-strategy comprises a first type of loading component and a second type of loading component; performing running state prediction according to the loading sub-strategy of each component loading node in each pending component loading strategy to obtain a predicted running state at the corresponding component loading node; sequentially constructing a predicted running state sequence corresponding to each pending component loading strategy, and calculating a second loading coefficient of the corresponding pending component loading strategy according to the predicted running state sequence; calculating a comprehensive loading coefficient of the corresponding pending component loading strategy based on the first loading coefficient and the second loading coefficient; sorting the plurality of pending component loading strategies according to the comprehensive loading coefficient, and setting the pending component loading strategy ranked first as the first component loading strategy.

[0008] In some embodiments of the present application, a load value threshold is preset; comparing each predicted load value in the predicted load value sequence with the load value threshold, calculating the number of component loading nodes with a predicted load value less than the load value threshold, denoted as a first number, and calculating a first difference value of the predicted load value less than the load value threshold. calculate a number of component loading nodes with a predicted load value not less than a load value threshold, denoted as a second number, and calculate a second difference value of the predicted load value not less than the load value threshold; calculate a load fluctuation degree of the predicted load value sequence, and set a first compensation coefficient according to the load fluctuation degree; generate a first loading coefficient corresponding to the to-be-determined component loading strategy based on the first number and corresponding first difference values, the second number and corresponding second difference values, and the first compensation coefficient; pre-set a running state threshold; compare each predicted running state in the predicted running state sequence with the running state threshold, calculate a number of component loading nodes with a predicted running state less than the running state threshold, denoted as a third number, and calculate a third difference value of the predicted running state less than the running state threshold; calculate a number of component loading nodes with a predicted running state not less than the running state threshold, denoted as a fourth number, and calculate a fourth difference value of the predicted running state not less than the running state threshold; calculate a state fluctuation degree of the predicted running state sequence, and set a second compensation coefficient according to the state fluctuation degree; generate a second loading coefficient corresponding to the to-be-determined component loading strategy based on the third number, corresponding third difference values, the fourth number, corresponding fourth difference values, and the second compensation coefficient.

[0009] In some embodiments of the present application, wherein J1 is the first loading coefficient, u2 is the second number, u1 is the first number, w1 is the first compensation coefficient, is a first loading conversion coefficient, is the r1th first difference value, is a weight coefficient of the r1th component loading node, is the r2th second difference value, is a weight coefficient of the r2th component loading node, J2 is the second loading coefficient, u3 is the third number, u4 is the fourth number, w2 is the second compensation coefficient, is a second loading conversion coefficient, is the r3th third difference value, is a weight coefficient of the r3th component loading node, is the r4th fourth difference value, is a weight coefficient of the r4th component loading node.

[0010] In some embodiments of the present application, the method further comprises: ​​According to the comprehensive prediction credibility, the number of category inspection nodes is set, wherein each category inspection node is set before a corresponding component loading node, and a time interval between the category inspection node and the corresponding component loading node is set according to the comprehensive prediction credibility; Real-time operation data at the category inspection node is acquired, and real-time demand categories at the corresponding category inspection node are calculated; If the real-time demand categories are consistent with the predicted demand categories, the first component loading strategy is not adjusted; If the real-time demand categories are inconsistent with the predicted demand categories, the first component loading strategy is adjusted according to the real-time demand categories, and simulation is performed according to the adjusted first component loading strategy.

[0011] In some embodiments of the present application, simulation demand coefficients and simulation state coefficients are calculated according to simulation results, including: A virtual machine is constructed based on a current resource data set, a behavior data set, a state data set and a usage data set; Simulation collection nodes are set according to the comprehensive loading coefficient of the first component loading strategy; The first component loading strategy is simulated based on the virtual machine, simulation data is collected according to the simulation collection nodes, and a simulation data set at each simulation collection node is generated; The simulation data set includes a simulation demand data subset and a simulation state data subset; A plurality of standard demand data intervals at each simulation collection node are preset; Simulation data in the simulation demand data subset at each simulation collection node is compared with a standard demand data interval at the corresponding simulation collection node, and simulation demand sub-coefficients at the corresponding simulation collection node are generated according to a comparison result; Simulation demand sub-coefficients at each simulation collection node are sequentially generated, and a weighted mean value is processed according to a weight coefficient of the simulation collection node, to obtain a simulation demand coefficient; Simulation data in the simulation state data subset at each simulation collection node is compared with a corresponding standard state data interval, and simulation state sub-coefficients at the corresponding simulation collection node are generated according to a comparison result; Simulation state sub-coefficients at each simulation collection node are sequentially generated, and a mean value is processed, to obtain a simulation state coefficient.

[0012] In some embodiments of the present application, a simulation application coefficient of the first component loading strategy is calculated according to the simulation demand coefficient and the simulation state coefficient, and it is determined whether to optimize the first component loading strategy, including: The calculation formula of the simulation application coefficient is: ; Wherein, F is a simulation application coefficient, f1 is a simulation demand coefficient, f2 is a simulation state coefficient, c1 is a weight coefficient of the simulation demand coefficient, and c2 is a weight coefficient of the simulation state coefficient. Pre-set an application coefficient threshold value; When the simulation application coefficient is less than the application coefficient threshold value, it is determined to optimize the first component loading strategy, and a second component loading strategy is generated; When the simulation application coefficient is not less than the application coefficient threshold value, it is determined not to optimize the first component loading strategy, and a loading instruction is issued according to the first component loading strategy.

[0013] In some embodiments of the present application, the second component loading strategy is generated, comprising: Calculate the application coefficient difference value between the simulation application coefficient and the application coefficient threshold value; According to the application coefficient difference value, determine the optimization level, which includes the first optimization level, the second optimization level and the third optimization level; According to the optimization level, select the corresponding preset optimization model, and optimize and adjust the first component loading strategy to generate the second component loading strategy.

[0014] The dynamic component loading optimization method of the embodiments of the present application has the following beneficial effects compared with the prior art: By identifying the predicted demand category and the actual running state, the first component loading strategy is generated and simulated to obtain the simulation application coefficient, and it is determined whether to optimize according to the simulation application coefficient, so as to dynamically adjust the component loading strategy according to the runtime demand, meet the user demand and stable operation, and improve the system response speed. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a flowchart of a dynamic component loading optimization method in the embodiments of the present application. DETAILED DESCRIPTION

[0016] The specific embodiments of the present application will be further described in detail below in combination with the drawings and examples. The following examples are used to illustrate the present application, but not to limit the scope of the present application.

[0017] In the description of the present application, it needs to be understood that the terms "center", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0018] The terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise stated, the meaning of "a plurality of" is two or more.

[0019] In the description of the present application, it needs to be explained that, unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through intermediate medium, or the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0020] As shown in Figure 1 The dynamic component loading optimization method of the embodiment of the present application comprises: S101: collecting current real-time running data and identifying to obtain corresponding predicted demand category and actual running state; S102: setting a plurality of component loading nodes based on the predicted demand category, and generating a first component loading strategy according to all component loading nodes, simulating according to the first component loading strategy, and calculating simulation demand coefficient and simulation state coefficient according to the simulation result; S103: calculating the simulation application coefficient of the first component loading strategy according to the simulation demand coefficient and the simulation state coefficient, and judging whether to optimize the first component loading strategy, if not, issuing loading instructions according to the first component loading strategy, if yes, generating a second component loading strategy.

[0021] In some embodiments of the present application, the corresponding predicted demand category and the actual running state are obtained, comprising: The current real-time running data is classified by type, and resource data set, behavior data set, state data set and use data set are constructed according to the classification result; generate a first predicted resource requirement and a corresponding first predicted confidence level according to the resource dataset and a pre-constructed dynamic load prediction model; generate a second predicted operation requirement and a corresponding second predicted confidence level according to the behavior dataset and a pre-constructed user intention prediction model; generate a third predicted operation requirement and a corresponding third predicted confidence level according to the usage dataset and a pre-constructed component dependency graph; determine a predicted requirement type and a corresponding comprehensive predicted confidence level based on the first predicted operation requirement, the second predicted operation requirement and the third predicted operation requirement; obtain historical operation data of the predicted requirement type and generate an operation evaluation value of the predicted requirement type; set a component loading node number of the predicted requirement type according to the operation evaluation value; compare each real-time operation data in the resource dataset with a corresponding preset resource data interval and generate an actual resource state according to a comparison result; compare each real-time operation data in the state dataset with a corresponding preset state data interval and generate an actual business state according to a comparison result; generate an actual operation state according to the actual resource state and the actual business state.

[0022] In the embodiment, the resource dataset includes real-time usage of hardware resources such as CPU usage, memory application and network bandwidth, and can directly reflect the resource requirement state of the system.

[0023] In the embodiment, the behavior dataset includes user operation behavior, function module calling, page browsing path and the like, and can reflect the actual behavior requirement of the user.

[0024] In the embodiment, the state dataset includes real-time state of a business process, data processing progress, business rule execution and the like, and can reflect the requirement of the business layer.

[0025] In the embodiment, the usage dataset includes loading time, usage frequency and execution efficiency of the component, and can reflect the actual usage and dependency relationship of the component.

[0026] In the embodiment, the preset requirement duration is set in advance according to the predicted requirement type, and the component loading node number is set according to the historical running duration of the predicted requirement type. When the operation evaluation value is larger, the component loading node number is larger, that is, the component loading frequency is higher, so as to meet the stable operation requirement of the system under high load and the user satisfaction.

[0027] In the embodiment, the preset state data interval is constructed according to preset upper and lower limit values of hardware resources such as CPU usage, memory application, network bandwidth, etc. When in the preset resource state data interval, it is a normal state, and if not, it is an abnormal state.

[0028] In the embodiment, the preset business state data interval refers to a data interval that meets standards or requirements in aspects such as business processes, data processing, and business rules, and is used to evaluate whether the operation state of the business level is normal. If in the preset business state data interval, it is a normal state, and if not, it is an abnormal state. Through the above comparison results, the actual resource state and business state of the system can be accurately judged, thereby providing reliable data support for subsequent component loading strategies.

[0029] In some embodiments of the present application, a plurality of running demand evaluation indexes are preset; Based on the plurality of running demand evaluation indexes, the historical running data packets of the predicted demand category are evaluated and analyzed to obtain running sub-evaluation values of the plurality of running demand evaluation indexes; A running evaluation value is generated according to the running sub-evaluation values of the plurality of running demand evaluation indexes and the weight coefficients of the corresponding running demand evaluation indexes; The calculation formula of the running evaluation value is: ; Wherein, Y1 is the running evaluation value, n is the running demand evaluation index, pi is the running sub-evaluation value of the i-th running demand evaluation index, ai is the weight coefficient of the i-th running demand evaluation index; Each comparison result is quantified to obtain a plurality of first quantization values in the resource data set and a plurality of second quantization values in the state data set, wherein each first quantization value corresponds to a real-time running data in the resource data set, and each second quantization value corresponds to a real-time running data in the state data set; The actual resource state is obtained according to the plurality of first quantization values, and the actual business state is obtained according to the plurality of second quantization values; The calculation formula of the actual running state is: ; Wherein, Y2 is the actual running state, z1 is the first conversion coefficient, b1 is the weight coefficient of the actual resource state, m1 is the number of real-time running data in the resource data set, is the s1-th first quantization value, is the weight coefficient of the s1-th first quantization value, z2 is the second conversion coefficient, b2 is the weight coefficient of the actual business state, m2 is the number of real-time running data in the state data set, is the s2-th second quantization value, The weight coefficient of the s2th second quantization value.

[0030] In the embodiment, the historical operation data packet includes a number of historical component quantities, types, historical loading frequencies, etc. corresponding to the loading of the predicted demand category, and the operation demand evaluation indexes include but are not limited to the component type, the component loading quantity in a preset period, the loading frequency, and the required resources when loading the component, etc.

[0031] In the embodiment, the data associated with each operation demand evaluation index in the historical operation data packet is screened out and compared with the preset reference data mapped by each operation demand evaluation index, and the operation sub-evaluation value is obtained according to the comparison result. For example, the preset reference data of the loading frequency evaluation index is A, when the associated data in the historical data packet is greater than A, the corresponding operation sub-evaluation value is greater, and vice versa.

[0032] In the embodiment, the value range of the first quantization value and the second quantization value is [-1, 1], when in the preset interval and closer to the middle value, it is closer to 1, when not in the preset interval and farther away from the upper limit value or the lower limit value, it is closer to -1.

[0033] In the embodiment, z1 refers to converting the sum of the first values into a value of the same dimension as the actual resource state, when the sum of the first values is greater, it means that the actual resource state is better, and vice versa, z2 refers to converting the sum of the second values into a value of the same dimension as the actual business state, when the sum of the second values is greater, it means that the actual business state is better, and vice versa, and the value range of the actual resource state and the actual business state is (0, 1).

[0034] In some embodiments of the application, the first component loading strategy is generated according to all component loading nodes, including: generating a component reference library of the predicted demand category, wherein the component reference library includes a number of to-be-loaded components sorted according to the demand time; performing timing analysis based on the sorting result of the to-be-loaded components in the component reference library and a number of component loading nodes of the predicted demand category, obtaining a number of loading sub-strategies of each component loading node, and generating a number of pending component loading strategies; performing load prediction according to the loading sub-strategy at each component loading node in each pending component loading strategy to obtain a predicted load value at the corresponding component loading node; sequentially constructing a predicted load value sequence corresponding to each pending component loading strategy, and calculating a first loading coefficient of the corresponding pending component loading strategy according to the predicted load value sequence; wherein each loading sub-strategy includes a first type of loading component and a second type of loading component. performing a running state prediction according to a loading sub-strategy at each component loading node in each to-be-determined component loading strategy, to obtain a predicted running state at the corresponding component loading node; sequentially constructing a predicted running state sequence corresponding to each to-be-determined component loading strategy, and calculating a second loading coefficient of the corresponding to-be-determined component loading strategy according to the predicted running state sequence; calculating a comprehensive loading coefficient of the corresponding to-be-determined component loading strategy based on the first loading coefficient and the second loading coefficient; sequentially sorting the to-be-determined component loading strategies according to the comprehensive loading coefficients, and setting the to-be-determined component loading strategy ranked first as the first component loading strategy.

[0035] In this embodiment, performing timing analysis means prioritizing the to-be-loaded components, obtaining a demand time ranking result according to the priority ranking, and assigning a corresponding loading sub-strategy to each loading node according to the ranking result and the timing relationship of the component loading nodes. In the assignment process, the loading order and loading time of the first type of loading components and the second type of loading components are considered to ensure the reasonable allocation and efficient use of system resources. At the same time, the timing analysis also considers factors such as system load and user behavior to dynamically adjust the component loading strategy to meet the real-time running needs of the system.

[0036] In this embodiment, the first type of loading component refers to the necessary component at the corresponding component loading node, and the second type of loading component refers to the non-necessary component at the corresponding component loading node. The first type of loading component in different loading sub-strategies of the same component loading node refers to the component that must be loaded at the current component loading node or the previous loading node, and the second type of loading component refers to the component that does not have to be loaded at the current component loading node or the previous loading node.

[0037] In this embodiment, the loading sub-strategies of adjacent component loading nodes are mutually influenced. For example, if the second type of loading component at the first component loading node includes the component of the next component loading node, then the first type of loading component at the second component loading node will be less, and the corresponding second type of loading component can be appropriately increased, thereby sequentially influencing the subsequent loading sub-strategies, and setting the sequentially set loading sub-strategies as a to-be-determined component loading strategy.

[0038] In some embodiments of the present application, a load value threshold is pre-set; comparing each predicted load value in the predicted load value sequence with the load value threshold, calculating the number of component loading nodes with a predicted load value less than the load value threshold, denoted as a first number, and calculating a first difference value of the predicted load value less than the load value threshold; calculating the number of component loading nodes with a predicted load value not less than the load value threshold, denoted as a second number, and calculating a second difference value of the predicted load value not less than the load value threshold; a load fluctuation degree of the sequence of predicted load values, and setting a first compensation coefficient according to the load fluctuation degree; generating a first loading coefficient corresponding to the to-be-determined component loading strategy based on the first quantity and corresponding first difference values, the second quantity and corresponding second difference values, and the first compensation coefficient; pre-setting a running state threshold; comparing each predicted running state in the sequence of predicted running states with the running state threshold, calculating a quantity of component loading nodes whose predicted running state is less than the running state threshold, denoted as a third quantity, and calculating a third difference value of the predicted running state less than the running state threshold; calculating a quantity of component loading nodes whose predicted running state is not less than the running state threshold, denoted as a fourth quantity, and calculating a fourth difference value of the predicted running state not less than the running state threshold; a state fluctuation degree of the sequence of predicted running states, and setting a second compensation coefficient according to the state fluctuation degree; generating a second loading coefficient corresponding to the to-be-determined component loading strategy based on the third quantity, corresponding third difference values, the fourth quantity, corresponding fourth difference values, and the second compensation coefficient.

[0039] In some embodiments of the present application, ; ; wherein J1 is the first loading coefficient, u2 is the second quantity, u1 is the first quantity, w1 is the first compensation coefficient, is a first loading conversion coefficient, is the r1th first difference value, is a weight coefficient of the r1th component loading node, is the r2th second difference value, is a weight coefficient of the r2th component loading node, J2 is the second loading coefficient, u3 is the third quantity, u4 is the fourth quantity, w2 is the second compensation coefficient, is a second loading conversion coefficient, is the r3th third difference value, is a weight coefficient of the r3th component loading node, is the r4th fourth difference value, is a weight coefficient of the r4th component loading node.

[0040] In the embodiment, the preset load value threshold is determined based on system historical operation data, hardware resource performance indicators, and expected business demand, and the like. Specifically, the preset load rate threshold reflects the maximum load proportion that the system can bear in a normal operation state. When the actual load rate of the system approaches or exceeds the threshold, it indicates that the system may face performance bottlenecks or resource shortages, and appropriate component loading adjustment measures need to be taken, such as reducing the loading of unnecessary components, optimizing the execution order of components, or increasing system resources, to ensure that the system can continue to operate stably and efficiently.

[0041] In the embodiment, the preset operation state threshold refers to the minimum operation state that the system should achieve in terms of business and resources in a normal operation state. When the predicted operation state is lower than the preset operation state threshold, it means that the system may have deficiencies or potential problems in some aspects, such as blocked business processes, low data processing efficiency, or insufficient utilization of hardware resources. At this time, the system needs to adjust the component loading strategy according to the specific situation, for example, increase the loading of key components, optimize the interaction logic between components, or upgrade hardware resources, to improve the overall operation performance of the system and ensure that the system can continuously and stably meet business demands.

[0042] In the embodiment, the first loading conversion coefficient refers to converting the sum of the first difference and the sum of the second difference into a value of the same dimension as the loading coefficient. When the sum of the first difference is larger and the sum of the second difference is smaller, the first loading coefficient is larger, and vice versa. The second loading conversion coefficient is the same, that is, when the sum of the third difference is smaller and the sum of the fourth difference is larger, the second loading coefficient is larger, and vice versa. Details are not repeated here.

[0043] In the embodiment, the value range of the compensation coefficient is (0.8, 1.2). When the sequence fluctuation degree is smaller, the compensation coefficient is larger, and vice versa.

[0044] In the embodiment, by calculating the first loading coefficient and the second loading coefficient, the comprehensive loading coefficient of each pending component loading strategy is obtained. The calculation formula of the comprehensive loading coefficient is: Z = α × J1 + β × J2, where Z is the comprehensive loading coefficient, α is the weight coefficient of the first loading coefficient, β is the weight coefficient of the second loading coefficient, and α + β = 1. The first component loading strategy is selected according to the comprehensive loading coefficient, that is, the component loading strategy that can meet the system load, business state, and user demand, to improve the component loading efficiency.

[0045] In some embodiments of the present application, the method further comprises: The number of category verification nodes is set according to the comprehensive prediction credibility, wherein each category verification node is set before a corresponding component loading node, and the time interval between the category verification node and the corresponding component loading node is set according to the comprehensive prediction credibility; Real-time operation data at the category verification node is obtained, and real-time demand categories at the corresponding category verification node are calculated; If the real-time demand categories are consistent with the predicted demand categories, the first component loading strategy is not adjusted; If the real-time demand categories are inconsistent with the predicted demand categories, the first component loading strategy is adjusted according to the real-time demand categories, and simulation is performed according to the adjusted first component loading strategy.

[0046] In this embodiment, the greater the comprehensive prediction credibility, the fewer the number of category verification nodes, and vice versa. Each category verification node is set before a corresponding component loading node. For example, five category verification nodes are set, the first category verification node is set before the first component loading node, the second category verification node is set before the second component loading node, and so on. The greater the comprehensive prediction credibility, the shorter the time interval between the first category verification node and the first component loading node, and vice versa.

[0047] In this embodiment, by setting the category verification nodes and the time intervals, the predicted demand categories are re-verified at the category verification nodes. If the verification result is consistent with the predicted demand categories, the subsequent component loading process is continued. If the verification result is inconsistent with the predicted demand categories, real-time operation data is re-collected and the predicted demand categories are re-calculated to ensure the accuracy and efficiency of component loading. In addition, by dynamically adjusting the number of category verification nodes and the time intervals, the flexibility and adaptability of component loading can be further improved to meet the needs of different scenarios.

[0048] In some embodiments of the present application, simulation demand coefficients and simulation state coefficients are calculated according to simulation results, including: A virtual machine is constructed based on the current resource data set, the behavior data set, the state data set, and the usage data set; Simulation collection nodes are set according to the comprehensive loading coefficient of the first component loading strategy; The first component loading strategy is simulated based on the virtual machine, simulation data is collected according to the simulation collection nodes, and simulation data sets at each simulation collection node are generated; The simulation data sets include simulation demand data subsets and simulation state data subsets; A number of standard demand data intervals at each simulation collection node are pre-set; The simulation simulation data in the simulation demand data subset at each simulation collection node is compared with the standard demand data interval at the corresponding simulation collection node, and a simulation demand subcoefficient at the corresponding simulation collection node is generated according to a comparison result; The simulation demand subcoefficient at each simulation collection node is sequentially generated, and a weighted mean value processing is performed according to the weight coefficient of the simulation collection node, to obtain a simulation demand coefficient; The simulation simulation data in the simulation state data subset at each simulation collection node is compared with the corresponding standard state data interval, and a simulation state subcoefficient at the corresponding simulation collection node is generated according to a comparison result; The simulation state subcoefficient at each simulation collection node is sequentially generated, and a mean value processing is performed, to obtain a simulation state coefficient.

[0049] In this embodiment, the standard state data interval includes a preset demand data interval and a preset state data interval, which are selected according to the data type of the simulation simulation data, i.e., resource demand data or service state data.

[0050] In this embodiment, the standard demand data interval is constructed according to the minimum demand data and the maximum demand data corresponding to the user demand satisfaction guarantee at each simulation collection node.

[0051] In this embodiment, when the simulation simulation data in the simulation demand data subset is in the standard demand data interval and closer to the upper limit value, the simulation demand subcoefficient is larger and the corresponding value range is [0, 1]; when the simulation simulation data is not in the standard demand data interval and farther away from the lower limit value, the simulation demand subcoefficient is smaller and the corresponding value range is [-1, 0].

[0052] In this embodiment, when the simulation simulation data in the simulation state data subset is in the standard state data interval and closer to the intermediate value, the simulation state subcoefficient is larger and the corresponding value range is [0, 1]; when the simulation simulation data is not in the standard demand data interval and farther away from the boundary value, the simulation state subcoefficient is smaller and the corresponding value range is [-1, 0].

[0053] In this embodiment, the simulation collection node is set by comprehensively loading the coefficient, and when the comprehensive loading coefficient is larger, the number of the set simulation collection nodes is smaller, and vice versa.

[0054] In this embodiment, by setting a plurality of simulation collection nodes and calculating the simulation demand coefficient and the simulation state coefficient at each simulation collection node, it is determined whether the first component loading strategy meets the demand, the deficiencies of the first component loading strategy are found and adjusted in time, the component loading efficiency is improved, the loading strategy is adjusted in real time according to the running demand, the memory occupation is reduced, and the system reaction speed is improved.

[0055] In some embodiments of the present application, the simulation application coefficient of the first component loading strategy is calculated according to the simulation demand coefficient and the simulation state coefficient, and it is determined whether to optimize the first component loading strategy, comprising: The calculation formula of the simulation application coefficient is: ; Wherein, F is the simulation application coefficient, f1 is the simulation demand coefficient, f2 is the simulation state coefficient, c1 is the weight coefficient of the simulation demand coefficient, and c2 is the weight coefficient of the simulation state coefficient. The application coefficient threshold is preset. When the simulation application coefficient is less than the application coefficient threshold, it is determined to optimize the first component loading strategy, and a second component loading strategy is generated. When the simulation application coefficient is not less than the application coefficient threshold, it is determined not to optimize the first component loading strategy, and the loading instruction is issued according to the first component loading strategy.

[0056] In the present embodiment, c1+c2=1.

[0057] In the present embodiment, the simulation application coefficient is calculated to evaluate the feasibility and effectiveness of the first component loading strategy in application, so as to ensure the efficient operation of the system. The larger the simulation application coefficient is, the better the performance of the first component loading strategy in application is, and the more it can meet the real-time operation requirements of the system. On the contrary, the smaller the simulation application coefficient is, the worse the performance of the first component loading strategy in application is, and it needs to be adjusted and optimized.

[0058] In the present embodiment, the application coefficient threshold refers to the minimum requirement to meet the stable operation of the system and the state of resource demand. When the simulation application coefficient is lower than the threshold, it indicates that the current first component loading strategy may not meet the system requirements, and therefore needs to be optimized and adjusted.

[0059] In the present embodiment, the simulation simulation and evaluation of the first component loading strategy can further improve the accuracy and efficiency of component loading, and ensure the reasonable allocation and efficient use of system resources.

[0060] In some embodiments of the present application, the second component loading strategy is generated, comprising: The application coefficient difference between the simulation application coefficient and the application coefficient threshold is calculated. The optimization level is determined according to the application coefficient difference, and the optimization level includes the first optimization level, the second optimization level and the third optimization level. The corresponding preset optimization model is selected according to the optimization level, and the first component loading strategy is optimized and adjusted to generate the second component loading strategy.

[0061] In the embodiment, the first optimization level refers to fine-tuning the loading sub-strategies at the partial acquisition nodes, screening the loading sub-strategies at the simulation acquisition nodes with a simulation demand sub-coefficient less than a preset demand sub-coefficient threshold or a simulation state sub-coefficient less than a preset state sub-coefficient threshold, and inputting the loading sub-strategies into a preset optimization model of the first optimization level to obtain an optimized strategy, and optimizing and adjusting the first component loading strategy to generate a second component loading strategy. The preset optimization model corresponding to the first optimization level is constructed according to the optimized strategy when the historical demand sub-coefficient or the historical state sub-coefficient is less than the corresponding coefficient threshold.

[0062] In the embodiment, the second optimization level refers to adjusting the loading sub-strategies at the multiple acquisition nodes by a large amplitude, re-distributing the loading order and the loading time of the first type of loading components and the second type of loading components, and inputting the loading sub-strategies into a preset optimization model of the second optimization level to obtain an optimized strategy, and optimizing and adjusting the first component loading strategy to generate a second component loading strategy. The preset optimization model corresponding to the second optimization level is constructed according to the historical component loading strategy that meets the current predicted demand category, so as to select the minimum adjustment strategy of the current first component loading strategy.

[0063] In the embodiment, the third optimization level refers to re-determining the predicted demand category and comprehensively re-planning the loading sub-strategies at all acquisition nodes, re-setting the loading nodes and the loading order, and inputting the loading sub-strategies into a preset optimization model of the third optimization level to obtain a brand-new loading strategy as a second component loading strategy. The preset optimization model corresponding to the third optimization level is constructed according to the historical component loading strategy that meets the user demand and the state demand of all demand categories.

[0064] In the embodiment, the preset optimization model includes but is not limited to an optimization model based on machine learning, an optimization model based on deep learning, an optimization model based on genetic algorithm, etc., and is selected according to the optimization level and the actual demand.

[0065] The above only describes the preferred embodiments of the present application. It should be noted that those skilled in the art can make several improvements and replacements without departing from the technical principles of the present application, and these improvements and replacements should also be considered as the protection scope of the present application.

Claims

1. A dynamic component loading optimization method, characterized in that, include: Collect and identify the current real-time operational data to obtain the corresponding predicted demand category and actual operational status; Multiple component loading nodes are set based on the predicted demand categories, and a first component loading strategy is generated based on all component loading nodes. Simulation is performed according to the first component loading strategy, and simulation demand coefficients and simulation state coefficients are calculated based on the simulation results. The simulation application coefficients of the first component loading strategy are calculated based on the simulation demand coefficients and simulation state coefficients. It is then determined whether the first component loading strategy should be optimized. If not, a loading command is issued according to the first component loading strategy. If so, a second component loading strategy is generated.

2. The dynamic component loading optimization method as described in claim 1, characterized in that, The corresponding predicted demand categories and actual operating status are obtained, including: The current real-time running data is categorized by type, and resource datasets, behavior datasets, status datasets, and usage datasets are constructed based on the categorization results. The first predicted resource demand and the corresponding first predicted confidence level are generated based on the resource dataset and the pre-built dynamic load prediction model. The second prediction run requirements and the corresponding second prediction credibility are generated based on the behavioral dataset and the pre-built user intent prediction model. The third prediction is generated based on the dataset and a pre-built component dependency graph, along with the prediction runtime requirements and the corresponding third prediction confidence level. The forecast demand type and the corresponding comprehensive forecast reliability are determined based on the first forecast operation demand, the second forecast operation demand, and the third forecast operation demand. Obtain historical runtime data packages for the predicted demand type and generate runtime evaluation values ​​for the predicted demand type; The number of component loading nodes for the predicted demand type is set based on the operational evaluation value; Each real-time running data in the resource dataset is compared with the corresponding preset resource data range, and the actual resource status is generated based on the comparison results. Each real-time running data in the status dataset is compared with the corresponding preset status data range, and the actual business status is generated based on the comparison results. The actual operating status is generated based on the actual resource status and the actual business status.

3. The dynamic component loading optimization method as described in claim 2, characterized in that, Several operational requirement evaluation indicators are pre-defined; Based on several operational demand evaluation indicators, the historical operational data packets of the predicted demand categories are evaluated and analyzed to obtain operational sub-evaluation values ​​of several operational demand evaluation indicators. The operation evaluation value is generated based on the operation sub-evaluation values ​​of several operation demand evaluation indicators and the corresponding weight coefficients of the operation demand evaluation indicators. The formula for calculating the operational evaluation value is as follows: ； Where Y1 is the operation evaluation value, n is the operation demand evaluation index, pi is the operation sub-evaluation value of the i-th operation demand evaluation index, and ai is the weight coefficient of the i-th operation demand evaluation index. Each comparison result is quantized to obtain several first quantized values ​​in the resource dataset and several second quantized values ​​in the status dataset. Each first quantized value corresponds to a real-time running data in the resource dataset, and each second quantized value corresponds to a real-time running data in the status dataset. The actual resource status is obtained based on several first quantitative values, and the actual business status is obtained based on several second quantitative values. The formula for calculating the actual operating state is as follows: ； Where Y2 represents the actual operating status, z1 is the first conversion coefficient, b1 is the weighting coefficient of the actual resource status, and m1 is the number of real-time operating data in the resource dataset. For the s1th first quantization value, Z1 is the weighting coefficient for the s1th first quantization value, Z2 is the second conversion coefficient, B2 is the weighting coefficient for the actual business state, and M2 is the number of real-time running data in the state dataset. For the s2th second quantization value, The weighting coefficient for the second quantized value of the s2th time.

4. The dynamic component loading optimization method as described in claim 3, characterized in that, The first component loading strategy is generated based on all component loading nodes, including: Generate a component reference library for predicting demand categories, wherein the component reference library includes several components to be loaded, sorted by demand time; Based on the sorting results of the components to be loaded from the component reference library and the timing analysis of several component loading nodes of the predicted demand categories, several loading sub-strategies for each component loading node are obtained, and several undetermined component loading strategies are generated. Load prediction is performed according to the loading sub-strategy at each component loading node in each undetermined component loading strategy, and the predicted load value at the corresponding component loading node is obtained. Construct a sequence of predicted load values ​​for each pending component loading strategy in sequence, and calculate the first loading coefficient of the corresponding pending component loading strategy based on the predicted load value sequence; Each loading sub-strategy includes a first type of loading component and a second type of loading component; The running status is predicted according to the loading sub-strategy at each component loading node in each undetermined component loading strategy, and the predicted running status at the corresponding component loading node is obtained. The predicted running state sequence corresponding to the loading strategy of each undetermined component is constructed sequentially, and the second loading coefficient of the corresponding undetermined component loading strategy is calculated based on the predicted running state sequence. The comprehensive loading coefficient of the corresponding undetermined component loading strategy is calculated based on the first loading coefficient and the second loading coefficient. The loading strategies for several undetermined components are sorted according to the comprehensive loading coefficient, and the loading strategy of the undetermined component ranked first is set as the first component loading strategy.

5. The dynamic component loading optimization method as described in claim 4, characterized in that, Preset load threshold; Compare each predicted load value in the predicted load value sequence with the load value threshold, calculate the number of component loading nodes whose predicted load value is less than the load value threshold, and record it as the first number. Also calculate the first difference between the predicted load value and the load value threshold. Calculate the number of component loading nodes whose predicted load value is not less than the load value threshold, denoted as the second number, and calculate the second difference between the predicted load value and the load value threshold. Calculate the load fluctuation degree of the predicted load value sequence, and set the first compensation coefficient according to the load fluctuation degree; The first loading coefficient is generated based on the first quantity and the corresponding first difference, the second quantity and the corresponding second difference, and the first compensation coefficient to generate the first loading coefficient of the loading strategy for the undetermined component. Pre-set operating status thresholds; Each predicted running state in the predicted running state sequence is compared with the running state threshold. The number of component loading nodes whose predicted running state is less than the running state threshold is calculated and recorded as the third number. The third difference between the predicted running state and the running state threshold is also calculated. Calculate the number of component loading nodes whose predicted operating state is not less than the operating state threshold, denoted as the fourth number, and calculate the fourth difference whose predicted operating state is not less than the operating state threshold. Calculate the degree of state fluctuation in the predicted operating state sequence, and set a second compensation coefficient according to the degree of state fluctuation; The second loading coefficient is generated based on the third quantity, the corresponding third difference, the fourth quantity, the corresponding fourth difference, and the second compensation coefficient to generate the second loading coefficient corresponding to the loading strategy of the undetermined component.

6. The dynamic component loading optimization method as described in claim 5, characterized in that, ； ； Where J1 is the first loading coefficient, u2 is the second quantity, u1 is the first quantity, and w1 is the first compensation coefficient. The first loading conversion factor, For the r1th first difference, The weight coefficient of the r1th component loading node, For the r2th second difference, J2 is the weight coefficient for the r2th component loading node, u3 is the third quantity, u4 is the fourth quantity, and w2 is the second compensation coefficient. For the second loading conversion factor, For the r3rd third difference, The weight coefficient for the r3rd component loading node. For the r-4th fourth difference, The weight coefficient is the loading node of the r4th component.

7. The dynamic component loading optimization method as described in claim 6, characterized in that, Also includes: The number of category inspection nodes is set according to the overall prediction confidence level, wherein each category inspection node is set before the corresponding component loading node, and the time interval between the category inspection node and the corresponding component loading node is set according to the overall prediction confidence level. Obtain real-time operational data at the category inspection node and calculate the real-time demand category at the corresponding category inspection node; If the real-time demand category matches the predicted demand category, the first component loading strategy will not be adjusted. If the real-time demand category is inconsistent with the predicted demand category, the loading strategy of the first component is adjusted according to the real-time demand category, and the simulation is performed according to the adjusted loading strategy of the first component.

8. The dynamic component loading optimization method as described in claim 7, characterized in that, Calculate the simulation demand coefficients and simulation state coefficients based on the simulation results, including: Based on the current resource dataset, behavior dataset, state dataset, and virtual machine built using the dataset; The simulation acquisition nodes are set according to the comprehensive loading coefficient of the first component loading strategy; The loading strategy of the first component is simulated based on a virtual machine. Simulation data is collected according to the simulation acquisition nodes, and simulation datasets are generated at each simulation acquisition node. The simulation dataset includes a subset of simulation requirement data and a subset of simulation state data. Pre-define several standard required data ranges at each simulation acquisition node; The simulation data in the subset of simulation requirements data at each simulation acquisition node is compared with the standard requirement data range at the corresponding simulation acquisition node, and the simulation requirement sub-coefficient at the corresponding simulation acquisition node is generated based on the comparison results. Simulation demand sub-coefficients are generated sequentially at each simulation acquisition node, and weighted averages are applied according to the weight coefficients of the simulation acquisition nodes to obtain the simulation demand coefficients. The simulation data in the subset of simulation state data at each simulation acquisition node is compared with the corresponding standard state data range, and the simulation state sub-coefficients at the corresponding simulation acquisition node are generated based on the comparison results. Simulation state sub-coefficients are generated sequentially at each simulation acquisition node and then averaged to obtain the simulation state coefficients.

9. The dynamic component loading optimization method as described in claim 8, characterized in that, Calculate the simulation application coefficients of the first component loading strategy based on the simulation demand coefficients and simulation state coefficients, and determine whether to optimize the first component loading strategy, including: The formula for calculating the simulation application coefficient is as follows: ； Where F is the simulation application coefficient, f1 is the simulation demand coefficient, f2 is the simulation state coefficient, c1 is the weighting coefficient of the simulation demand coefficient, and c2 is the weighting coefficient of the simulation state coefficient. Pre-set application coefficient thresholds; When the simulation application coefficient is less than the application coefficient threshold, it is determined that the loading strategy of the first component should be optimized and the loading strategy of the second component should be generated. When the simulation application coefficient is not less than the application coefficient threshold, it is determined that the first component loading strategy will not be optimized, and a loading command will be issued according to the first component loading strategy.

10. The dynamic component loading optimization method as described in claim 9, characterized in that, The second component loading strategy is generated, including: Calculate the difference between the simulation application coefficient and the application coefficient threshold; The optimization level is determined based on the difference in application coefficients, and the optimization level includes a first optimization level, a second optimization level, and a third optimization level. Select the corresponding preset optimization model based on the optimization level, optimize and adjust the first component loading strategy, and generate the second component loading strategy.

Images