A load balancing method for key management system based on AI

Abstract

The present invention relates to the field of key management, and discloses an AI-based load balancing method for a key management system, comprising first collecting load information of each node in the key management system and storing it as historical data, wherein the load information includes CPU usage, memory occupancy and network bandwidth; then processing the historical data based on a pre-trained neural network model to obtain a load evaluation coefficient of each node in the key management system, and obtaining a load trend coefficient according to a change in the load evaluation coefficient within a monitoring cycle, and predicting the load trend of each node in the key management system within the next monitoring cycle; then identifying hot nodes among each node in the key management system according to the predicted load trend result, and judging whether the hot nodes have potential performance bottlenecks; finally adjusting the load balancing strategy according to the results of the hot nodes and the nodes for judging whether there are potential performance bottlenecks.

Description

Technical Field

[0001] The present invention relates to the field of key management, and in particular to an AI-based key management system load balancing method. Background Art

[0002] In the digital age, with the surge in data volume and the increase in the risk of data leakage, the importance of key management system has become increasingly prominent. It is not only the key to protecting static data, but also the basis for ensuring the security of data during transmission.

[0003] The key management system plays a vital role in data security and is the basis for ensuring the security of encryption algorithms. The key management system is responsible for the generation, storage, distribution, use, update and destruction of keys throughout their life cycle to ensure the security and compliance of keys. The key management system protects user privacy and corporate assets by protecting keys from unauthorized access and leakage, maintaining the confidentiality, integrity and availability of data.

[0004] In order to improve system availability and optimize resource utilization, existing key management systems also adopt load balancing strategies. The purpose is: through load balancing, the key management system can distribute the workload to multiple nodes, so that even if a node fails, other nodes can continue to provide services, thereby ensuring the high availability of the system. At the same time, load balancing can effectively avoid single point overload, ensure that the computing resources and data storage in the system are evenly utilized, and improve overall resource utilization.

[0005] However, there are still some problems in actual use. For example, although load balancing is designed to optimize resource utilization, improper configuration may cause certain nodes to become bottlenecks, affecting overall performance. Summary of the invention

[0006] The purpose of the present invention is to provide an AI-based key management system load balancing method to solve the above technical problems.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A key management system load balancing method based on AI, comprising:

[0009] S1. Collect the load information of each node in the key management system and store it as historical data, wherein the load information includes CPU usage, memory occupancy and network bandwidth;

[0010] S2. Process the historical data based on the pre-trained neural network model to obtain the load assessment coefficient of each node in the key management system, and obtain the load trend coefficient according to the change of the load assessment coefficient in a monitoring cycle, and predict the load trend of each node in the key management system in the next monitoring cycle;

[0011] S3. According to the predicted load trend results, identify the hotspot nodes in each node in the key management system, and determine whether the hotspot nodes have potential performance bottlenecks;

[0012] S4. Adjust the load balancing strategy based on the results of the hot nodes and the nodes that determine whether there are potential performance bottlenecks.

[0013] As a further technical solution, the process of obtaining the load trend coefficient is:

[0014] Step 1: Divide a monitoring cycle into sub-periods;

[0015] Step 2: Fit and obtain the actual curve and reference curve of the load evaluation coefficient of each node in each sub-period changing with time;

[0016] Step 3: By formula:

[0017]

[0018] Calculate the first The load trend factor of each node ;

[0019] in, is the coefficient of variation, is the conversion coefficient, through the formula:

[0020] Calculated;

[0021] , The first sub-period is The start and end time of each node, , Respectively Sub-period The start and end time of each node, , The first sub-period, The reference value of the change of the sub-period; , The first sub-period, The actual curve of the load evaluation factor changing with time in a sub-cycle, , They are the reference curves of the load evaluation coefficient changing with time in the 1st sub-cycle and the nth sub-cycle respectively.

[0022] As a further technical solution, the process of predicting the load trend of each node in the key management system in the next monitoring cycle is as follows:

[0023] Step 1: The load trend coefficient of the previous monitoring cycle calculated for each node The preset load trend threshold interval Make comparisons;

[0024] Step 2: If , it is predicted that the load of the current node is on an increasing trend, otherwise it goes to step 3;

[0025] Step 3: If , it is predicted that the load of the current node is in a stable trend, otherwise it goes to step 4;

[0026] Step 4: If , it is predicted that the load of the current node is in a downward trend.

[0027] As a further technical solution, the working process of S3 is:

[0028] S31, the load evaluation factor With preset load assessment thresholds For comparison, if and , then the current node is judged to be a hotspot node;

[0029] S32. Obtain various performance parameters of the current node in real time based on AI, where the performance parameters are the number of key transactions per second, the number of key query requests per second, throughput, number of concurrent users, and error rate;

[0030] S33, comparing each performance parameter with a preset performance warning value, if at least one performance parameter item exceeds the performance warning value, it is preliminarily determined that the current hotspot node has a potential performance bottleneck;

[0031] S34, the hotspot nodes preliminarily determined to have potential performance bottlenecks are added to the secondary analysis queue for secondary analysis again, and the list of nodes with potential performance bottlenecks is finally determined based on the analysis results.

[0032] As a further technical solution, the process of performing secondary analysis in S34 is:

[0033] By formula:

[0034]

[0035] Calculate the performance index of the current node ;

[0036] in, is the performance index of the current node, is the performance indicator warning value of the current node, is the total number of samples, For the The corresponding load evaluation coefficient obtained from the sampling points ;

[0037] The performance index of the current node The preset performance threshold of the current node Compare, if , it is judged that the current node has a potential performance bottleneck, otherwise, it is judged that the current node does not have a potential performance bottleneck.

[0038] As a further technical solution, the performance index of the current node By formula:

[0039]

[0040] Calculate and obtain;

[0041] in, , , , , They are the number of key transactions per second, the number of key query requests per second, the throughput, the number of concurrent users, and the error rate. , , , , They are the warning values ​​of key transactions per second, key query requests per second, throughput, concurrent users, and error rate. , , , , is the preset weight coefficient.

[0042] As a further technical solution, the load information is obtained through system monitoring tools, log records or performance monitoring software.

[0043] As a further technical solution, the process of adjusting the load balancing strategy is as follows:

[0044] Hot nodes with potential performance bottlenecks are classified as first-level high-load nodes, and nodes without potential performance bottlenecks are classified as second-level high-load nodes. The nodes under the conditions are divided into three levels of high-load nodes;

[0045] Arrange the high-load nodes at each level in descending order according to the load evaluation coefficient to form a priority queue;

[0046] According to the order of level 1 > level 2 > level 3, the high-load nodes in the priority queue of each level are moved to the nodes that meet the requirements in order. The low-load nodes that meet the conditions perform task migration and the low-load nodes are arranged in ascending order of load evaluation coefficients to receive tasks in sequence.

[0047] Beneficial effects of the present invention:

[0048] The present invention can predict the load trend of each node, identify hotspot nodes according to the prediction results, and pre-judge which nodes have potential performance bottlenecks to make real-time dynamic adjustments to the pre-set load balancing strategy in a timely manner, thereby preventing improper configuration of hotspot nodes from causing performance bottlenecks when the next monitoring cycle is running, resulting in performance limitations of the key management system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention will be further described below in conjunction with the accompanying drawings.

[0050] Figure 1 It is a diagram of the method steps of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] See also Figure 1 As shown, the present invention is a key management system load balancing method based on AI, comprising:

[0053] S1. Collect the load information of each node in the key management system and store it as historical data, wherein the load information includes CPU usage, memory occupancy and network bandwidth;

[0054] S2. Process the historical data based on the pre-trained neural network model to obtain the load assessment coefficient of each node in the key management system, and obtain the load trend coefficient according to the change of the load assessment coefficient in a monitoring cycle, and predict the load trend of each node in the key management system in the next monitoring cycle;

[0055] S3. According to the predicted load trend results, identify the hotspot nodes in each node in the key management system, and determine whether the hotspot nodes have potential performance bottlenecks;

[0056] S4. Adjust the load balancing strategy based on the results of the hot nodes and the nodes that determine whether there are potential performance bottlenecks.

[0057] The process of obtaining the load trend coefficient is:

[0058] Step 1: Divide a monitoring cycle into sub-periods;

[0059] Step 2: Fit and obtain the actual curve and reference curve of the load evaluation coefficient of each node in each sub-period changing with time;

[0060] Step 3: By formula:

[0061]

[0062] Calculate the first The load trend factor of each node ;

[0063] in, is the coefficient of variation, is the conversion coefficient, which is determined based on historical experimental data, using the formula:

[0064] Calculated;

[0065] , The first sub-period is The start and end time of each node, , Respectively Sub-period The start and end time of each node, , The first sub-period, The reference value of the sub-period change is determined based on a comprehensive analysis of historical data and experimental data; , The first sub-period, The actual curve of the load evaluation factor changing with time in a sub-cycle, , They are the reference curves of the load evaluation coefficient changing with time in the 1st sub-cycle and the nth sub-cycle respectively.

[0066] In this embodiment, a key management system load balancing method is provided that can avoid improper configuration causing certain nodes to become performance bottlenecks, and a specific method for obtaining a load trend coefficient is given. Specifically, the load information of each node in the key management system is first collected and stored as historical data, and the load information includes CPU usage, memory occupancy and network bandwidth. Then, the historical data is processed based on a pre-trained neural network model to obtain a load evaluation coefficient of each node in the key management system, and a load trend coefficient is obtained according to the change of the load evaluation coefficient in a monitoring cycle, and the load trend of each node in the key management system in the next monitoring cycle is predicted. Then, according to the predicted load trend result, the hotspot nodes in each node in the key management system are identified, and it is determined whether the hotspot nodes have potential performance bottlenecks. Finally, the load balancing strategy is adjusted according to the results of the hotspot nodes and the nodes for determining whether there are potential performance bottlenecks. Through the above technical solution, the load trend of each node can be predicted, and the hotspot nodes can be identified according to the predicted results. By pre-determining which nodes have potential performance bottlenecks, the pre-set load balancing strategy can be adjusted in real time and dynamically, so as to prevent the improper configuration of the hotspot nodes from causing performance bottlenecks when the next monitoring cycle is running, resulting in the problem of limited performance of the key management system.

[0067] At the same time, through the formula: Calculate the first The load trend factor of each node By accumulating the deviation between the actual curve of the load assessment coefficient changing with time in each sub-period and the reference curve, and then obtaining the accurate load assessment of the current node by averaging, it provides effective data support for the subsequent judgment of potential performance bottlenecks and reduces the probability of misjudgment.

[0068] The process of predicting the load trend of each node in the key management system in the next monitoring cycle is as follows:

[0069] The load trend coefficient of the previous monitoring cycle calculated for each node The preset load trend threshold interval For comparison; if , it is predicted that the load of the current node is on the rise; if , then it is predicted that the load of the current node is in a stable trend; if , it is predicted that the load of the current node is in a downward trend; through the above technical solution, an accurate and rapid prediction of the load trend of the current node is achieved.

[0070] The working process of S3 is as follows:

[0071] S31, the load evaluation factor With preset load assessment thresholds For comparison, if and , then the current node is judged to be a hotspot node;

[0072] S32. Obtain various performance parameters of the current node in real time based on AI, where the performance parameters are the number of key transactions per second, the number of key query requests per second, throughput, number of concurrent users, and error rate;

[0073] S33, comparing each performance parameter with a preset performance warning value, if at least one performance parameter item exceeds the performance warning value, it is preliminarily determined that the current hotspot node has a potential performance bottleneck;

[0074] S34, the hotspot nodes preliminarily determined to have potential performance bottlenecks are added to the secondary analysis queue for secondary analysis again, and the list of nodes with potential performance bottlenecks is finally determined based on the analysis results.

[0075] The process of secondary analysis in S34 is as follows:

[0076] By formula:

[0077]

[0078] Calculate the performance index of the current node ;

[0079] in, is the performance index of the current node, is the performance indicator warning value of the current node, is the total number of samples, For the The corresponding load evaluation coefficient obtained from the sampling points ; The performance index of the current node By formula:

[0080]

[0081] Calculate and obtain;

[0082] in, , , , , They are the number of key transactions per second, the number of key query requests per second, the throughput, the number of concurrent users, and the error rate. , , , , They are the warning values ​​of key transactions per second, key query requests per second, throughput, concurrent users, and error rate. , , , , It is a preset weight coefficient, which is selected and formulated based on historical data;

[0083] The performance index of the current node The preset performance threshold of the current node Compare, if , it is judged that the current node has a potential performance bottleneck, otherwise, it is judged that the current node does not have a potential performance bottleneck.

[0084] In this embodiment, a method for determining whether a node has a potential performance bottleneck is provided. Specifically, the load evaluation coefficient With preset load assessment thresholds For comparison, if and , then the current node is judged to be a hotspot node, and then the performance parameters of the current node are obtained in real time through AI, and the performance parameters are the number of key transactions per second, the number of key query requests per second, throughput, number of concurrent users, and error rate; then, by comparing the various performance parameters with the preset performance warning values, a preliminary judgment is made on whether the current hotspot node has a potential performance bottleneck. Obviously, if it does, it means that the hotspot node will have performance problems in the future, which will eventually affect the performance of the entire system. However, in order to prevent prejudgment, a re-judgment is performed through a secondary analysis method, combining the initial judgment and the re-judgment process to achieve the purpose of accurately judging the hotspot nodes with potential performance bottlenecks; the specific secondary analysis process is through the formula Calculate the performance index of the current node , and then substitute into Calculate the performance index of the current node Obviously, if the number of key transactions per second, the number of key query requests per second, the throughput, and the number of concurrent users are larger, the performance index The larger the error rate, the smaller the performance index. The larger the performance index Performance indicator warning value The larger the ratio, the more likely the hotspot node is to be fully loaded and lacks redundancy. Once more nodes are added, performance problems will inevitably occur. Therefore, the performance index of the hotspot node directly leads to a performance bottleneck in the entire key management system, affecting performance. It indicates the volatility of the load assessment coefficient of the node in the last monitoring cycle. The greater the volatility, the worse the stability of the node, and therefore the more likely it is to be suddenly fully loaded.

[0085] It should be noted that the real-time acquisition of various performance parameters of the current node through AI in the present invention is deep learning or machine learning in the prior art, which will not be elaborated here; the number of key transactions per second is the number of key management related transactions that can be processed per second; the number of key query requests per second is the number of key query requests processed per second; the throughput is the amount of data or the number of request tasks that can be processed per unit time; the number of concurrent users is the maximum number of users that can be processed simultaneously; the error rate is the ratio of the number of errors occurring when processing requests per unit time to the total number of requests.

[0086] The load information is obtained through system monitoring tools, log records or performance monitoring software. The performance monitoring software can be used , or , it is sufficient to monitor the performance parameters, and no further details will be given here.

[0087] The process of adjusting the load balancing policy is as follows:

[0088] Hot nodes with potential performance bottlenecks are classified as first-level high-load nodes, and nodes without potential performance bottlenecks are classified as second-level high-load nodes. The nodes under the conditions are divided into three levels of high-load nodes;

[0089] Arrange the high-load nodes at each level in descending order according to the load evaluation coefficient to form a priority queue;

[0090] According to the order of level 1 > level 2 > level 3, the high-load nodes in the priority queue of each level are moved to the nodes that meet the requirements in order. The low-load nodes that meet the conditions perform task migration and the low-load nodes are arranged in ascending order of load evaluation coefficients to receive tasks in sequence.

[0091] In this embodiment, by migrating tasks of high-load nodes to low-load nodes, load balancing of each node in the entire key management system is achieved, thereby improving overall performance and resource utilization.

[0092] It should be noted that the calculation formulas and various parameters involved in the calculation in the present invention have been dimensionally processed in advance, and the process of dimensionless processing is well known in the industry and will not be described here.

[0093] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A key management system load balancing method based on AI, characterized in that: include: S1. Collect the load information of each node in the key management system and store it as historical data, wherein the load information includes CPU usage, memory occupancy and network bandwidth; S2. Process the historical data based on the pre-trained neural network model to obtain the load assessment coefficient of each node in the key management system, and obtain the load trend coefficient according to the change of the load assessment coefficient in a monitoring cycle, and predict the load trend of each node in the key management system in the next monitoring cycle; S3. According to the predicted load trend results, identify the hotspot nodes in each node in the key management system, and determine whether the hotspot nodes have potential performance bottlenecks; S4. Adjust the load balancing strategy based on the results of hot nodes and nodes that determine whether there are potential performance bottlenecks; The process of obtaining the load trend coefficient is: Step 1: Divide a monitoring cycle into sub-periods; Step 2: Fit and obtain the actual curve and reference curve of the load evaluation coefficient of each node in each sub-period changing with time; Step 3: By formula: ； Calculate the first The load trend factor of each node ; in, is the coefficient of variation, is the conversion coefficient, through the formula: ； Calculated; , The first sub-period is The start and end time of each node, , Respectively Sub-period The start and end time of each node, , The first sub-period, The reference value of the change of each sub-period; , The first sub-period, The actual curve of the load evaluation factor changing with time in a sub-cycle, , They are the reference curves of the load evaluation coefficient changing with time in the 1st sub-period and the nth sub-period respectively; The process of predicting the load trend of each node in the key management system in the next monitoring cycle is as follows: Step 1: The load trend coefficient of the previous monitoring cycle calculated for each node The preset load trend threshold interval Make comparisons; Step 2: If , it is predicted that the load of the current node is on an increasing trend, otherwise it goes to step 3; Step 3: If , it is predicted that the load of the current node is in a stable trend, otherwise it goes to step 4; Step 4: If , it is predicted that the load of the current node is in a downward trend; The working process of S3 is as follows: S31, the load evaluation factor With preset load assessment thresholds For comparison, if and , then the current node is judged to be a hotspot node; S32. Obtain various performance parameters of the current node in real time based on AI, where the performance parameters are the number of key transactions per second, the number of key query requests per second, throughput, number of concurrent users, and error rate; S33, comparing each performance parameter with a preset performance warning value, if at least one performance parameter item exceeds the performance warning value, it is preliminarily determined that the current hotspot node has a potential performance bottleneck; S34, adding the hotspot nodes that are initially determined to have potential performance bottlenecks to the secondary analysis queue for further secondary analysis, and finally determining the list of nodes that have potential performance bottlenecks based on the analysis results; The process of secondary analysis in S34 is as follows: By formula: ； Calculate the performance index of the current node ; in, is the performance index of the current node, is the performance indicator warning value of the current node, is the total number of samples, For the The corresponding load evaluation coefficient obtained from the sampling points ; The performance index of the current node The preset performance threshold of the current node Compare, if , it is judged that the current node has a potential performance bottleneck, otherwise, it is judged that the current node does not have a potential performance bottleneck; The performance index of the current node By formula: ； Calculate and obtain; in, They are the number of key transactions per second, the number of key query requests per second, the throughput, the number of concurrent users, and the error rate. They are the warning values ​​of key transactions per second, key query requests per second, throughput, concurrent users, and error rate. is the preset weight coefficient; The process of adjusting the load balancing policy is as follows: Hot nodes with potential performance bottlenecks are classified as first-level high-load nodes, and nodes without potential performance bottlenecks are classified as second-level high-load nodes. The nodes under the conditions are divided into three levels of high-load nodes; Arrange the high-load nodes at each level in descending order according to the load evaluation coefficient to form a priority queue; According to the order of level 1 > level 2 > level 3, the high-load nodes in the priority queue of each level are moved to the nodes that meet the requirements in order. The low-load nodes that meet the conditions perform task migration and the low-load nodes are arranged in ascending order of load evaluation coefficients to receive tasks in sequence.

2. The AI-based key management system load balancing method according to claim 1 is characterized in that: The load information is obtained through system monitoring tools, log records or performance monitoring software.

Images