Online dynamic adjustment method, system and equipment for control parameters and medium

Abstract

The invention provides an online dynamic adjustment method, system and device for control parameters and a medium, and the method comprises the steps: collecting storage use data from each processor node, processing state fluctuation in a multi-node environment through employing a distributed collection protocol, and obtaining the current storage occupation proportion of each node; determining high-load state nodes by adopting a threshold comparison method according to the current storage occupation proportion of each node; analyzing a storage state of a processor node adjacent to the high-load state node, and determining a target node of the migration task; and sending a part of data packets of the high-load node to a target node through a transmission path based on an encrypted transmission protocol to complete online dynamic adjustment. According to the method, real-time monitoring of the storage state is realized in a multiprocessor environment, and storage resources are dynamically balanced through a collaboration mechanism among nodes.

Description

Technical Field

[0001] This invention relates to the field of information technology, and specifically to an online dynamic adjustment method, system, device, and medium for control parameters. Background Technology

[0002] In today's era of rapid technological advancement, the application of artificial intelligence (AI) technology has permeated all walks of life, becoming a core driving force for social progress and industrial innovation. Especially in resource-constrained processor environments, how to efficiently utilize storage space to optimize the performance of AI models has become a crucial research area. Research in this field not only concerns the improvement of technical performance but also directly impacts the stability and responsiveness of smart devices in practical applications, possessing undeniable value in enhancing user experience and industrial efficiency.

[0003] However, current methods for optimizing processor storage space often struggle to adapt to the complex environments of multiple devices working collaboratively. Many solutions lack the ability to respond promptly to dynamic changes in storage resources when multiple processor nodes are running simultaneously, leading to excessive storage space consumption on some nodes while resources on others are underutilized. This imbalance is particularly pronounced under high-load scenarios, severely impacting the overall system efficiency.

[0004] A deeper technical challenge lies in the real-time monitoring of storage space and the coordination of cross-node resource allocation. First, because the storage status of each processor node fluctuates constantly during operation, without real-time monitoring, it's difficult to accurately determine whether a node is about to reach its storage limit, thus missing optimization opportunities. Second, this monitoring lag further exacerbates the difficulty of cross-node resource allocation. In a multi-node environment, relying solely on the independent adjustment of a single node cannot solve the overall resource shortage problem; a collaborative mechanism between nodes is necessary to balance storage pressure. For example, in an intelligent transportation system supported by multiple processor nodes, if a node is responsible for processing real-time traffic data and its storage space is about to run out, and other nodes cannot share the data storage task in time, data processing will be interrupted, affecting the accuracy of traffic control.

[0005] Therefore, how to achieve real-time monitoring of storage status in a multi-processor environment and dynamically balance storage resources through inter-node collaboration mechanisms has become a key issue that this research urgently needs to address. Solving this problem will directly affect the stable operation and resource utilization efficiency of artificial intelligence systems in complex scenarios. Summary of the Invention

[0006] To address the problem that existing solutions lack timely response to dynamic changes in storage resources when multiple processor nodes operate simultaneously, leading to excessive storage space consumption on some nodes while resources on other nodes are underutilized, this invention provides an online dynamic adjustment method for control parameters, comprising: Data on storage usage is collected from each processor node, and a distributed collection protocol is used to process state fluctuations in a multi-node environment to obtain the current storage occupancy ratio of each node. High-load nodes are determined by using a threshold comparison method based on the current storage usage ratio of each node. Analyze the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task; Based on the encrypted transmission protocol, some data packets from high-load nodes are sent to the target node through the transmission path to complete online dynamic adjustment.

[0007] Preferably, the step of determining nodes in a high-load state using a threshold comparison method based on the current storage occupancy ratio of each node includes: Compare the current storage usage ratio of each node with a preset first threshold; The nodes whose storage usage ratio exceeds the preset first threshold are identified as high-load nodes.

[0008] Preferably, the step of analyzing the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task includes: Based on the storage status of the adjacent processor nodes of the high-load state node, calculate the storage occupancy ratio of each adjacent processor node. Processor nodes with storage occupancy ratios less than a preset second threshold are selected as candidate nodes, and a candidate node set is constructed from these candidate nodes. Based on the candidate node set, the collected fluctuation results are organized to obtain the trajectory of the change in the occupancy ratio of each node; The method of extracting trend data based on the change trajectory of the occupancy ratio is used to determine whether it indicates potential load fluctuations, and to obtain a resource assessment of candidate nodes that integrate fluctuation trends; Target nodes are selected from candidate node resource evaluations that incorporate fluctuation trends using load balancing algorithms, reinforcement learning algorithms, or game theory methods.

[0009] Preferably, a load balancing algorithm is used to select the target node from the candidate node resource evaluation based on the fusion fluctuation trend, including: The current storage availability ratio is calculated based on the current storage occupancy ratio of candidate nodes according to the fusion fluctuation trend. Based on trend data, linear regression slope or exponential smoothing algorithm is used to predict the changing trend of the current storage availability ratio and determine the trend coefficient. Based on the current storage availability ratio, trend coefficient, current storage occupancy ratio of candidate nodes, and transmission overhead between candidate nodes and high-load nodes, combined with the comprehensive evaluation value calculation formula, the comprehensive evaluation value of each candidate node is obtained. Select the candidate node with the highest comprehensive evaluation value as the target node.

[0010] Preferably, the comprehensive evaluation value is calculated using the following formula:

[0011] In the formula, For comprehensive evaluation, The weights are the proportion of current available storage. This represents the current percentage of available storage. This is the trend coefficient. The weights of the trend coefficients, This represents the current storage usage percentage of the candidate nodes. The weights of the currently available proportion of candidate nodes are stored. To reduce transmission overhead between high-load state nodes, Weights for transmission overhead between high-load state nodes.

[0012] Preferably, the step of sending a portion of the data packets from the high-load node to the target node via the transmission path based on the encrypted transmission protocol to complete the online dynamic adjustment includes: Obtain the communication link between the target node and the high-load state node, determine the data migration path, and obtain the migration task allocation scheme; According to the migration task allocation scheme, some data packets of the high-load node are sent to the target node through an encrypted transmission protocol, and it is determined whether state fluctuations occur during the transmission process. When state fluctuations occur, adjust the migration rate to obtain confirmation of completed migration data. Confirm the data that has been migrated, update the storage status records of each node, and complete the online dynamic adjustment.

[0013] Preferably, after sending a portion of the data packets from the high-load node to the target node via the transmission path based on the encrypted transmission protocol to complete the online dynamic adjustment, the method further includes: Based on the resource assessment of candidate nodes with fused fluctuation trends, an extended assessment index is generated by integrating the sequence of trend data with the judgment results of potential load fluctuations. The resource allocation results are verified using a consistency check method to obtain the consistency check results. Determine whether the extended evaluation metric matches the consistency verification result. If they do not match, trigger an additional verification loop; otherwise, do not trigger an additional verification loop.

[0014] Preferably, the present invention further includes: Determine whether the overall system has reached dynamic equilibrium. When equilibrium is reached, obtain the optimized storage usage distribution. Based on the optimized storage usage distribution, a resource utilization report is generated. It is then determined whether there are any remaining high-load nodes in the report. If so, the collection and allocation process is repeated until the iterative results of continuous monitoring are obtained. Otherwise, the resource utilization report is output.

[0015] Furthermore, the present invention also provides an online dynamic adjustment system for control parameters, comprising: The data processing module is used to collect storage usage data from each processor node, and uses a distributed collection protocol to process state fluctuations in a multi-node environment to obtain the current storage occupancy ratio of each node. The status judgment module is used to determine the high-load nodes based on the current storage usage ratio of each node using a threshold comparison method; The target determination module is used to analyze the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task. The data migration module is used to send a portion of the data packets from high-load nodes to the target node via the transmission path based on an encrypted transmission protocol, thereby completing online dynamic adjustment.

[0016] Preferably, the status determination module is specifically used for: Compare the current storage usage ratio of each node with a preset first threshold; The nodes whose storage usage ratio exceeds the preset first threshold are identified as high-load nodes.

[0017] Preferably, the target determination module includes: The computing unit is used to calculate the storage occupancy ratio of each adjacent processor node based on the storage status of the adjacent processor nodes of the high-load state node. The comparison unit is used to select the processor nodes with storage occupancy ratios less than a preset second threshold as candidate nodes, and to construct a candidate node set from the candidate nodes. The data processing unit is used to process the collected fluctuation results based on the candidate node set to obtain the trajectory of the change in the occupancy ratio of each node. The fusion unit is used to determine whether potential load fluctuations are indicated by extracting trend data based on the trajectory of the change in the occupancy ratio, and to obtain a resource assessment of candidate nodes with fused fluctuation trends. The node selection unit is used to select target nodes from the evaluation of candidate node resources that integrate fluctuation trends using load balancing algorithms, reinforcement learning algorithms, or game theory methods.

[0018] Preferably, the specific implementation steps of selecting the target node from the candidate node resource evaluation based on the fusion fluctuation trend using the load balancing algorithm in the node selection unit include: The current storage availability ratio is calculated based on the current storage occupancy ratio of candidate nodes according to the fusion fluctuation trend. Based on trend data, linear regression slope or exponential smoothing algorithm is used to predict the changing trend of the current storage availability ratio and determine the trend coefficient. Based on the current storage availability ratio, trend coefficient, current storage occupancy ratio of candidate nodes, and transmission overhead between candidate nodes and high-load nodes, combined with the comprehensive evaluation value calculation formula, the comprehensive evaluation value of each candidate node is obtained. Select the candidate node with the highest comprehensive evaluation value as the target node.

[0019] Preferably, the comprehensive evaluation value is calculated using the following formula:

[0020] In the formula, For comprehensive evaluation, The weights are the proportion of current available storage. This represents the current percentage of available storage. This is the trend coefficient. The weights of the trend coefficients, This represents the current storage usage percentage of the candidate nodes. The weights of the currently available proportion of candidate nodes are stored. To reduce transmission overhead between high-load state nodes, Weights for transmission overhead between high-load state nodes.

[0021] Preferably, the data migration module is specifically used for: Obtain the communication link between the target node and the high-load state node, determine the data migration path, and obtain the migration task allocation scheme; According to the migration task allocation scheme, some data packets of the high-load node are sent to the target node through an encrypted transmission protocol, and it is determined whether state fluctuations occur during the transmission process. When state fluctuations occur, adjust the migration rate to obtain confirmation of completed migration data. Confirm the data that has been migrated, update the storage status records of each node, and complete the online dynamic adjustment.

[0022] Preferably, it further includes: a verification module, used for: Based on the resource assessment of candidate nodes with fused fluctuation trends, an extended assessment index is generated by integrating the sequence of trend data with the judgment results of potential load fluctuations. The resource allocation results are verified using a consistency check method to obtain the consistency check results. Determine whether the extended evaluation metric matches the consistency verification result. If they do not match, trigger an additional verification loop; otherwise, do not trigger an additional verification loop.

[0023] In another aspect, this application also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, an online dynamic adjustment method for control parameters as described above is implemented.

[0024] In another aspect, this application also provides a computer-readable storage medium having an executable program stored thereon, which, when executed, implements an online dynamic adjustment method for control parameters as described above.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides an online dynamic adjustment method for control parameters, comprising: collecting storage usage data from each processor node; processing state fluctuations in a multi-node environment using a distributed acquisition protocol to obtain the current storage occupancy ratio of each node; determining high-load nodes based on a threshold comparison method according to the current storage occupancy ratio of each node; analyzing the storage status of processor nodes adjacent to the high-load nodes to determine the target node for the migration task; and sending a portion of the data packets from the high-load nodes to the target node via a transmission path based on an encrypted transmission protocol to complete the online dynamic adjustment. This invention achieves real-time monitoring of storage status in a multi-processor environment and dynamically balances storage resources through a collaborative mechanism between nodes. It solves the problem that existing solutions lack timely response capabilities to dynamic changes in storage resources when multiple processor nodes are running simultaneously, leading to excessive storage space occupation on some nodes while other nodes' resources are not fully utilized. Attached Figure Description

[0026] Figure 1 This is a flowchart of an online dynamic adjustment method for control parameters according to the present invention; Figure 2 This is a schematic diagram of an electronic device structure according to the present invention. Detailed Implementation

[0027] To better understand the present invention, the following description, in conjunction with the accompanying drawings and embodiments, will further illustrate the content of the present invention.

[0028] Example 1: A method for online dynamic adjustment of control parameters, such as Figure 1As shown, it includes: Step 1: Collect storage usage data from each processor node, use a distributed collection protocol to process state fluctuations in a multi-node environment, and obtain the current storage occupancy ratio of each node; Step 2: Determine the high-load nodes based on the current storage usage ratio of each node using a threshold comparison method; Step 3: Analyze the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task; Step 4: Based on the encrypted transmission protocol, send some data packets from the high-load node to the target node through the transmission path to complete the online dynamic adjustment.

[0029] The following section provides a further explanation of each step.

[0030] Step 1: Collect storage usage data from each processor node, use a distributed collection protocol to handle state fluctuations in a multi-node environment, and obtain the current storage usage ratio of each node, including: By collecting storage usage data from each processor node and using a distributed collection protocol to process state fluctuations in a multi-node environment, real-time storage status indicators are obtained, and the current storage occupancy ratio of each node is calculated.

[0031] Step 1 includes the following specific implementation steps: S1: Determine the target time for data acquisition: During the acquisition period, determine a target timestamp for aligning the data acquisition actions of multiple nodes.

[0032] S2: Coordinate node acquisition action: Based on the target timestamp, coordinate multiple processor nodes so that each node performs snapshot acquisition of local storage data in time alignment.

[0033] S3: Calculate and report the ratio: Each node calculates its current storage occupancy ratio based on the collected local storage usage data, and reports the ratio and collected metadata.

[0034] S4: Handling State Fluctuations: Receive data reported by nodes, and correct or mark the storage occupancy ratio based on the deviation between the actual collection time of each node and the target timestamp, so as to eliminate or identify the impact of state fluctuations caused by asynchronous collection times, and obtain a set of time-aligned storage occupancy ratios.

[0035] Optionally, the "coordination" described in step S2 can be achieved in any of the following ways: Distributed negotiation is implemented as follows: each node negotiates a consistent logical time or batch number as the batch identifier through a distributed consensus protocol (such as Raft or Paxos), and performs collection when the agreed conditions are met locally.

[0036] Alternatively, it can be implemented based on a global clock. The specific steps are as follows: all nodes access a high-precision global clock source and agree to perform data acquisition simultaneously at a certain absolute clock moment.

[0037] Alternatively, a centralized collaborative approach can be implemented, with the following steps: a central coordinator broadcasts a collection instruction containing the target timestamp to all nodes, and the nodes perform collection when their local time reaches that timestamp.

[0038] Step 2: Determine high-load nodes based on the current storage usage ratio of each node using a threshold comparison method, including: Compare the current storage usage ratio of each node with a preset first threshold; The nodes whose storage usage ratio exceeds the preset first threshold are identified as high-load nodes.

[0039] Step 2 specifically includes: Based on the current storage occupancy ratio of each node, a threshold comparison method is used to analyze resource utilization. If the storage occupancy ratio of a node exceeds a preset threshold, the node is determined to be in a high-load state, and the node collaboration process is initiated. Otherwise, the node is not in a high-load state.

[0040] Step 3: Analyze the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task, including: Based on the storage status of the adjacent processor nodes of the high-load state node, calculate the storage occupancy ratio of each adjacent processor node. Processor nodes with storage occupancy ratios less than a preset second threshold are selected as candidate nodes, and a candidate node set is constructed from these candidate nodes. Based on the candidate node set, the collected fluctuation results are organized to obtain the trajectory of the change in the occupancy ratio of each node; The method of extracting trend data based on the change trajectory of the occupancy ratio is used to determine whether it indicates potential load fluctuations, and to obtain a resource assessment of candidate nodes that integrate fluctuation trends; A load balancing algorithm is used to select target nodes from the candidate node resource evaluation based on the fusion fluctuation trend.

[0041] Step 3 specifically includes: For identified high-load nodes, the available resource list is obtained by querying the storage status of adjacent processor nodes. This allows for the identification of nodes with low occupancy rates, resulting in a set of candidate nodes that can share storage tasks. Based on this candidate node set, the collected fluctuation results are processed using a method that extracts trend data from occupancy rate changes. The method determines whether the trend data indicates potential load fluctuations, leading to a resource assessment of candidate nodes that incorporate fluctuation trends. Finally, a load balancing algorithm, reinforcement learning algorithm, or game theory method is used to select target nodes from this resource assessment.

[0042] Furthermore, the selection of target nodes from candidate node resource assessments based on fusion fluctuation trends using a load balancing algorithm includes: The current storage availability ratio is calculated based on the current storage occupancy ratio of candidate nodes according to the fusion fluctuation trend. Based on trend data, linear regression slope or exponential smoothing algorithm is used to predict the changing trend of the current storage availability ratio and determine the trend coefficient. Based on the current storage availability ratio, trend coefficient, current storage occupancy ratio of candidate nodes, and transmission overhead between candidate nodes and high-load nodes, combined with the comprehensive evaluation value calculation formula, the comprehensive evaluation value of each candidate node is obtained. Select the candidate node with the highest comprehensive evaluation value as the target node.

[0043] Furthermore, a negative reinforcement learning algorithm is used to select target nodes from the candidate node resource evaluation that incorporates fluctuation trends, including: The state of a distributed storage system is defined as the environmental information observed by the agent. The state space S contains two types of information: node-level state and system-level state.

[0044] Node-level state involves collecting a multi-dimensional state vector for each processor node in the cluster, including but not limited to: Storage dimensions: current storage occupancy rate and storage capacity change rate (obtained through linear fitting of historical data). Performance metrics: current CPU utilization, memory utilization, network I / O throughput, and disk I / O latency. Health dimensions: node availability status, hardware error count. Trend dimension: Future load forecasting indicators (such as the expected change in storage usage within the next 5 minutes) obtained from the analysis of historical fluctuation results.

[0045] System-level status: Aggregated information describing the overall status of the cluster, including: average cluster storage utilization, Storage utilization standard deviation (used to measure balance), number of data migration tasks currently in progress, and overall network congestion level.

[0046] Action space definition: "Select a target node" or "Do not migrate for now" are defined as actions that the agent can execute. Specifically, action space A includes: selecting any node from the candidate node set as the target node for this migration task. The no_op (no operation) action indicates that no migration is performed in the current state, and the agent waits for the next decision cycle.

[0047] Reward function design: The reward function R is used to guide the agent to learn the optimal policy, and its design takes into account both immediate benefits and long-term effects.

[0048] Instant Reward: Calculated immediately after the next migration action is completed. This instant reward includes the following: Storage balancing reward: After migration, calculate the standard deviation of the storage occupancy ratio of all nodes. The reward value is proportional to the reduction in the standard deviation.

[0049] Performance Bonus: If the I / O latency or CPU utilization of the source hotspot node decreases significantly after migration, a positive bonus will be awarded.

[0050] Cost penalty: The migration operation itself consumes network bandwidth and computing resources, and the reward function includes a negative penalty term that is proportional to the amount of data migrated.

[0051] Constraint violation penalty: If the migration causes the target node to exceed the security threshold or violates any service level agreement (SLA), a large negative reward will be given.

[0052] Long-term rewards: A discount-based cumulative reward mechanism is adopted. The goal of the agent is to maximize the expected value of the cumulative reward over multiple decision cycles in the future, rather than the single instant reward.

[0053] Deep Q-Network (DQN) or its variants (such as Double DQN and DuelingDQN) are preferred as the core algorithm for reinforcement learning.

[0054] The network structure includes an input layer, a feature extraction layer, and a value output layer.

[0055] Input layer: Receives the state space vector St. Feature extraction layer: Composed of several fully connected layers or convolutional layers, used to automatically learn high-order representations of state features.

[0056] Value output layer: For each possible action a∈A, output a Q value Q(St,a), which represents the long-term value expectation of performing action a in state St.

[0057] (2) Training process 1. Experience collection: The agent explores in a real cluster or high-fidelity simulation environment and stores the tuple (St,At,Rt,St+1) of each decision in the experience replay buffer.

[0058] 2. Offline learning: Periodically sample a batch of empirical data from the buffer to update the weight parameters of the Q-network. Backpropagation is performed by minimizing the temporal difference error between the predicted Q-value and the target Q-value.

[0059] 3. Exploration and Utilization: In the early stages of training, a high exploration rate (such as an ε-greedy strategy) is adopted to allow the agent to try more random actions to explore the environment. As training progresses, the exploration rate is gradually reduced, making the agent rely more on the learned strategy (i.e., choosing the action with the highest Q value).

[0060] 4. Model Deployment and Updates: The trained policy network can be deployed in the production environment's decision engine. The system can be set to continuous learning mode, periodically fine-tuning the model with new production data to adapt to changes in workload patterns.

[0061] Since training directly on a production cluster carries risks, this embodiment recommends building a digital twin simulation environment.

[0062] System modeling: Based on historical monitoring data, statistical modeling is performed on the storage growth pattern, I / O performance, network interaction, etc. of each node in the cluster.

[0063] Load generation: Use real workload tracking playback or statistical models to generate dynamic loads in the simulator that closely resemble the production environment.

[0064] Accelerated training: In the simulator, a large number of training instances can be run in parallel, quickly completing the iteration and convergence of the policy, and then the validated policy can be applied to the production system.

[0065] Step 3 involves selecting the target node by quantifying the evaluation indicators and choosing the maximum value from them. The specific quantification process is as follows: (1) Quantification of evaluation indicators For each candidate node i, calculate a comprehensive score, which combines: • Current storage availability percentage S i =1 - Occupancy rate (the higher the value, the better).

[0066] • Trend coefficient T i : If the occupancy rate is decreasing or stabilizing, T i >1 (bonus point); If the occupancy rate is increasing, Ti <1 (discount factor), the faster the increase, the larger the discount. T i The proportion of occupancy can be quantified by predicting the occupancy ratio after a certain period of time using linear regression slope and exponential smoothing.

[0067] • Current compute load C of the node i (CPU / memory usage, etc.) Avoid selecting a node with large storage space but full CPU load.

[0068] • Transmission overhead D between the source node H and the source node H i (Network bandwidth, latency, distance, etc.)

[0069] Assuming we need to receive M amounts of data, predict the change in the storage usage ratio of node i after the migration is completed, ensuring it remains below the threshold.

[0070] (2) Formal example of load balancing algorithm Define the overall evaluation value of node i (the larger the value, the better):

[0071] In the formula, Let i be the comprehensive evaluation value of node i. The weights are the proportion of current available storage. This represents the current storage availability percentage for node i. Let be the trend coefficient of node i. The weights of the trend coefficients, This represents the current storage usage percentage of node i. The weights of the currently available proportion of candidate nodes are stored. The transmission overhead between node i and the node in the high-load state. Weights for transmission overhead between high-load state nodes.

[0072] The chosen strategy can be: • Maximum value method: directly select E i The largest node.

[0073] • Random weighted method: according to E i The probability distribution is randomly selected to avoid multiple migrations competing for the same optimal node at the same time. • Multi-objective optimization: If multiple migration tasks are performed in parallel, the objective can be to minimize the variance of storage usage of each node after migration, and the target node for each migration can be selected using greedy or linear programming. (3) Specific algorithm for fluctuation trend fusion It's possible to model a future storage usage prediction curve for each candidate node (e.g., for the next 5 minutes or 1 hour). Load balancing algorithms can: • Compare which node still has the lowest and most stable storage usage at the expected migration completion time.

[0074] • Use reinforcement learning or control theory to select nodes that can maximize the long-term storage load balance of the system.

[0075] Step 4: Based on the encrypted transmission protocol, send a portion of the data packets from the high-load node to the target node through the transmission path to complete the online dynamic adjustment, including: Obtain the communication link between the target node and the high-load state node, determine the data migration path, and obtain the migration task allocation scheme; According to the migration task allocation scheme, some data packets of the high-load node are sent to the target node through an encrypted transmission protocol, and it is determined whether state fluctuations occur during the transmission process. When state fluctuations occur, adjust the migration rate to obtain confirmation of completed migration data. Confirm the data that has been migrated, update the storage status records of each node, and complete the online dynamic adjustment.

[0076] Step 4 specifically includes: The system obtains the communication links between selected nodes, determines the data migration path, and obtains a migration task allocation scheme. Based on the obtained migration task allocation scheme, it sends some data packets from high-load nodes to the target nodes through an encrypted transmission protocol, determines whether state fluctuations occur during transmission, adjusts the migration rate if state fluctuations occur, and obtains confirmation of completed migration data. By summarizing the confirmation of completed migration data, it updates the storage status records of each node and completes online dynamic adjustment.

[0077] Step 4 is followed by: Based on the resource assessment of candidate nodes obtained from the fusion fluctuation trend, extended assessment indicators are generated by integrating the trend data sequence with the potential load fluctuation judgment results, and the correlation basis with subsequent migration confirmation is determined. The resource allocation results are verified using a consistency check method to determine whether the overall system has reached dynamic balance. When balance is achieved, the optimized storage occupancy distribution is obtained. Based on the optimized storage occupancy distribution, the extended evaluation indicators are compared with the consistency check results. If there is a mismatch between the trend data sequence and the transmission feedback, an additional verification loop is triggered to obtain the balance confirmation after correlation and fusion.

[0078] The additional verification loop specifically includes: The cycle of "collecting and storing usage data → analyzing trends → generating extended evaluation indicators → load balancing decisions → verification" is repeated. This time, the transmission feedback information from before is incorporated (to correct the trend model) to obtain a more accurate understanding and ultimately make a "balance confirmation after correlation and fusion"—that is, a more reliable balance state judgment based on actual feedback correction.

[0079] The invention also includes: generating a resource utilization report based on the optimized storage occupancy distribution; determining whether there are any remaining high-load nodes in the report; if so, repeatedly executing the aforementioned collection and allocation process to obtain iterative results of continuous monitoring; otherwise, directly outputting the optimized storage occupancy distribution; and adjusting the parameters of the first threshold and the second threshold to adapt to changes in the multi-node environment by analyzing the optimization efficiency indicators in the iterative results, determining whether the adjustment improves system stability, and obtaining the final dynamic resource balance configuration, wherein the first threshold is greater than the second threshold. These two data can be defined manually or specified by technical standards.

[0080] Example 2 This embodiment of a method for online dynamic adjustment of control parameters specifically includes the following steps: S101. By collecting storage usage data from each processor node, using a distributed collection protocol to process state fluctuations in a multi-node environment, real-time storage status indicators are obtained, and the current storage occupancy ratio of each node is obtained.

[0081] S102. Based on the current storage occupancy ratio of each node, the resource utilization is analyzed using a threshold comparison method. If the storage occupancy ratio of a certain node exceeds the preset threshold, it is determined that the node is in a high-load state and the node collaboration process needs to be started.

[0082] S103. For a given high-load node, obtain a list of available resources by querying the storage status of adjacent processor nodes, determine which nodes have a low occupancy rate, and obtain a set of candidate nodes that can share the storage task.

[0083] S104. Based on the obtained candidate node set, the collected fluctuation results are processed using the method of extracting trend data from the trajectory of occupancy ratio changes. It is determined whether the trend data indicates potential load fluctuations, and a candidate node resource assessment with integrated fluctuation trends is obtained.

[0084] S105. Based on the candidate node resource assessment obtained from the fusion fluctuation trend, by integrating the trend data sequence and the potential load fluctuation judgment results, an extended assessment index is generated to determine the correlation basis with subsequent migration confirmation.

[0085] S106. Using a load balancing algorithm, target nodes are selected from the candidate node resource evaluation of the fusion fluctuation trend, the communication links between the selected nodes are obtained, the data migration path is determined, and the migration task allocation scheme is obtained.

[0086] Step S106 is implemented in detail as follows: (1) Quantification of evaluation indicators For each candidate node i, calculate a comprehensive score, which combines: • Current storage availability percentage S i =1 - Occupancy rate (the higher the value, the better).

[0087] • Trend coefficient T of node i i : If the occupancy rate is decreasing or stabilizing, T i >1 (bonus point); If the occupancy rate is increasing, T i <1 (discount factor), the faster the increase, the larger the discount. T i The proportion of occupancy can be quantified by predicting the occupancy ratio after a certain period of time using linear regression slope and exponential smoothing.

[0088] • Current compute load C of the node i (CPU / memory usage, etc.) Avoid selecting a node with large storage space but full CPU load.

[0089] • Transmission overhead D between the source node H and the source node H i (Network bandwidth, latency, distance, etc.)

[0090] Assuming we need to receive M amounts of data, predict the change in the storage usage ratio of node i after the migration is completed, ensuring it remains below the threshold.

[0091] (2) Formal example of load balancing algorithm Define the overall evaluation value of node i (the larger the value, the better):

[0092] In the formula, Let i be the comprehensive evaluation value of node i. The weights are the proportion of current available storage. This represents the current storage availability percentage for node i. Let be the trend coefficient of node i. The weights of the trend coefficients, This represents the current storage usage percentage of node i. The weights of the currently available proportion of candidate nodes are stored. The transmission overhead between node i and the node in the high-load state. Weights for transmission overhead between high-load state nodes.

[0093] The chosen strategy can be: • Maximum value method: directly select E i The largest node.

[0094] • Random weighted method: according to E i The probability distribution is randomly selected to avoid multiple migrations competing for the same optimal node at the same time. • Multi-objective optimization: If multiple migration tasks are performed in parallel, the objective can be to minimize the variance of storage usage of each node after migration, and the target node for each migration can be selected using greedy or linear programming. (3) Specific algorithm for fluctuation trend fusion It's possible to model a future storage usage prediction curve for each candidate node (e.g., for the next 5 minutes or 1 hour). Load balancing algorithms can: • Compare which node still has the lowest and most stable storage usage at the expected migration completion time.

[0095] • Use reinforcement learning or control theory to select nodes that can maximize the long-term storage load balance of the system. The following example illustrates this: Assume there are 3 candidate nodes: node 1, node 2, and node 3.

[0096] The rate of increase corresponds to the trend in the table. The trend coefficient of node i can be defined as follows, and in this embodiment, i takes the value of 1, 2, or 3: T i =1-α×rate of ascent Let α = 0.1, then: The trend coefficient T1 for node 1: T1 = 1 - 0.1 × 0.5 = 0.95 • Trend coefficient T2 for node 2: T2=1 • Trend coefficient T3 for node 3: T3 = 1 - 0.1 × (-0.2) = 1.02 Normalized storage availability ratio: S1=1-0.4=0.6, S2=0.7, S3=0.65 (if it is just a comparison, 0.6, 0.7, 0.65 can be directly used in the formula), where S1, S2, and S3 are the normalized storage availability ratios of the trend coefficients of nodes 1, 2, and 3, respectively. CPU load factor: 1-C i We get 0.7, 0.4, 0.5, where C i This represents the current CPU load of node i. Delay factor normalization: Assuming a delay of 1-5ms is mapped to 0-1, the larger the delay, the lower the score. For example, using 1-(D i -1) / 4 gives 0.75, 0, 1. Set weights =0.4, =0.3, =0.2, =0.1, then calculate (simple weighted sum): E1=0.4×0.6+0.3×0.95+0.2×0.7-0.1×0.75=0.59 E2=0.4×0.7+0.3×1+0.2×0.4-0.1×0=0.66 E3=0.4×0.65+0.3×1.02+0.2×0.5-0.1×1=0.566 Among them, E1, E2, and E3 are the comprehensive evaluation values ​​of nodes 1, 2, and 3, respectively. At this time, node 2 has the highest score. Although the latency is slightly higher, the storage trend is stable and the current occupancy is the lowest. The comprehensive evaluation is the best, so node 2 is selected as the target node.

[0097] 5. Optional advanced methods Based on reinforcement learning: Each choice is treated as a decision, and the reward is defined as the improvement in the overall load balance of the system over a period of time after the migration. The optimal strategy is learned through Q-learning and other methods. Game theory approach: When multiple migration tasks are carried out in parallel, candidate nodes are "resources" and multiple source nodes are "players", which are allocated through stable matching or bidding mechanisms. Dynamic weight adjustment: Automatically adjusts the weights in the formula based on the current overall load status of the cluster (e.g., increasing the latency weight when the network is congested). ). The specific implementation steps of the multi-task collaborative transfer method based on game theory are as follows: The relationship between multiple source nodes (migration demanders) and multiple candidate target nodes (resource providers) is modeled as a two-sided matching problem.

[0098] Participants: All "source nodes" that need to migrate data constitute set P, and all "candidate nodes" that can be used as targets constitute set Q. The preference list includes: each source node p i For all candidate nodes q j There is a preference ranking, based on the resource evaluation score of candidate nodes that incorporate fluctuation trends. Each candidate node q... j For all source nodes p iThere is also a preference sorting, which can be based on the size of the migrated data, the priority of the source node, or the estimated load status of the node itself after migration.

[0099] Matching objective: Find a stable matching scheme such that there is no such pair (p i ,q j They tend to prefer each other rather than the currently matched object.

[0100] The Gale-Shapley "delayed acceptance" algorithm is preferred for finding stable matches. The specific process is as follows: 1. Proposal Phase: All source nodes send a "migration proposal" to the candidate node ranked first according to their own preference list.

[0101] 2. Decision-making phase: After receiving a proposal, each candidate node temporarily accepts the proposal of the source node it favors most, based on its own preference list, and rejects all other proposals.

[0102] 3. Iteration Phase: The rejected source node sends a proposal to the next candidate node in its preference list. The candidate node receiving the new proposal replaces the current temporary acceptor if the new proposer is superior; otherwise, it rejects the proposal.

[0103] 4. Termination condition: Repeat step 3 until no source node is rejected or all source nodes have completed matching. At this point, the matching result is stable.

[0104] For more complex scenarios, the storage and I / O resources of candidate nodes can be abstracted into tradable commodities, introducing a market mechanism.

[0105] Virtual currency: The system allocates a virtual budget to each node.

[0106] Bidding Mechanism: When a candidate node is contested by multiple source nodes, the source nodes must bid (commit to paying virtual currency) to compete for the resource. Bidding can be based on the urgency of their migration needs (such as the degree to which the current load exceeds a threshold).

[0107] Equilibrium price: Through an iterative bidding process, a market clearing state is reached, at which point resource allocation can reflect the relative value of different migration tasks, thereby maximizing overall benefits.

[0108] The reason for considering "integrating fluctuation trends" is that trends have already been extracted using historical fluctuation data in previous steps. The purpose is to avoid selecting a node that is currently idle but about to be filled by other tasks, which could lead to migration failure or immediately trigger a secondary migration. Therefore, when selecting a node, the load balancing algorithm uses trend data to correct its assessment of the node's current capacity, which is equivalent to performing dynamic capacity prediction, thereby improving the effectiveness of migration.

[0109] S107. Based on the obtained migration task allocation scheme, send some data packets of the high-load node to the target node through an encrypted transmission protocol, determine whether state fluctuations occur during the transmission process, and adjust the migration rate if state fluctuations occur to obtain data confirmation that the migration has been completed.

[0110] S108. By summarizing the data that has been migrated, update the storage status records of each node, verify the resource allocation results using the consistency check method, determine whether the overall system has reached dynamic balance, and obtain the optimized storage occupancy distribution.

[0111] S109. Based on the obtained optimized storage occupancy distribution, by comparing the extended evaluation index with the consistency verification results, if there is a mismatch between the trend data sequence and the transmission feedback, an additional verification loop is triggered to obtain the balance confirmation after correlation and fusion.

[0112] S1010. Based on the optimized storage occupancy distribution, generate a resource utilization report, determine whether there are any remaining high-load nodes in the report, and if so, repeatedly execute the aforementioned collection and allocation process to obtain iterative results of continuous monitoring.

[0113] S1011. By analyzing the optimization efficiency index in the iteration results, adjust the threshold parameters to adapt to the changes in the multi-node environment, determine whether the adjustment improves system stability, and obtain the final dynamic resource balance configuration.

[0114] Example 3 Based on the same inventive concept, the present invention also provides an online dynamic adjustment system for control parameters, comprising: The data processing module is used to collect storage usage data from each processor node, and uses a distributed collection protocol to process state fluctuations in a multi-node environment to obtain the current storage occupancy ratio of each node. The status judgment module is used to determine the high-load nodes based on the current storage usage ratio of each node using a threshold comparison method; The target determination module is used to analyze the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task. The data migration module is used to send a portion of the data packets from high-load nodes to the target node via the transmission path based on an encrypted transmission protocol, thereby completing online dynamic adjustment.

[0115] Preferably, the status determination module is specifically used for: Compare the current storage usage ratio of each node with a preset first threshold; The nodes whose storage usage ratio exceeds the preset first threshold are identified as high-load nodes.

[0116] Preferably, the target determination module includes: The computing unit is used to calculate the storage occupancy ratio of each adjacent processor node based on the storage status of the adjacent processor nodes of the high-load state node. The comparison unit is used to select the processor nodes with storage occupancy ratios less than a preset second threshold as candidate nodes, and to construct a candidate node set from the candidate nodes. The data processing unit is used to process the collected fluctuation results based on the candidate node set to obtain the trajectory of the change in the occupancy ratio of each node. The fusion unit is used to determine whether potential load fluctuations are indicated by extracting trend data based on the trajectory of the change in the occupancy ratio, and to obtain a resource assessment of candidate nodes with fused fluctuation trends. The node selection unit is used to select target nodes from the evaluation of candidate node resources that integrate fluctuation trends using load balancing algorithms, reinforcement learning algorithms, or game theory methods.

[0117] Preferably, the specific implementation steps of selecting the target node from the candidate node resource evaluation based on the fusion fluctuation trend using the load balancing algorithm in the node selection unit include: The current storage availability ratio is calculated based on the current storage occupancy ratio of candidate nodes according to the fusion fluctuation trend. Based on trend data, linear regression slope or exponential smoothing algorithm is used to predict the changing trend of the current storage availability ratio and determine the trend coefficient. Based on the current storage availability ratio, trend coefficient, current storage occupancy ratio of candidate nodes, and transmission overhead between candidate nodes and high-load nodes, combined with the comprehensive evaluation value calculation formula, the comprehensive evaluation value of each candidate node is obtained. Select the candidate node with the highest comprehensive evaluation value as the target node.

[0118] Preferably, the comprehensive evaluation value is calculated using the following formula:

[0119] In the formula, For comprehensive evaluation, The weights are the proportion of current available storage. This represents the current percentage of available storage. This is the trend coefficient. The weights of the trend coefficients, This represents the current storage usage percentage of the candidate nodes. The weights of the currently available proportion of candidate nodes are stored. To reduce transmission overhead between high-load state nodes, Weights for transmission overhead between high-load state nodes.

[0120] Preferably, the data migration module is specifically used for: Obtain the communication link between the target node and the high-load state node, determine the data migration path, and obtain the migration task allocation scheme; According to the migration task allocation scheme, some data packets of the high-load node are sent to the target node through an encrypted transmission protocol, and it is determined whether state fluctuations occur during the transmission process. When state fluctuations occur, adjust the migration rate to obtain confirmation of completed migration data. Confirm the data that has been migrated, update the storage status records of each node, and complete the online dynamic adjustment.

[0121] Preferably, it further includes: a verification module, used for: Based on the resource assessment of candidate nodes with fused fluctuation trends, an extended assessment index is generated by integrating the sequence of trend data with the judgment results of potential load fluctuations. The resource allocation results are verified using a consistency check method to obtain the consistency check results. Determine whether the extended evaluation metric matches the consistency verification result. If they do not match, trigger an additional verification loop; otherwise, do not trigger an additional verification loop.

[0122] Example 4 like Figure 2 As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0123] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to realize the steps of the online dynamic adjustment method of control parameters in the above embodiments.

[0124] Example 5 Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the storage medium here can include both built-in storage media within the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of the online dynamic adjustment method for control parameters in the above embodiments.

[0125] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0129] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for online dynamic adjustment of control parameters, characterized in that, include: Data on storage usage is collected from each processor node, and a distributed collection protocol is used to process state fluctuations in a multi-node environment to obtain the current storage occupancy ratio of each node. High-load nodes are determined by using a threshold comparison method based on the current storage usage ratio of each node. Analyze the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task; Based on the encrypted transmission protocol, some data packets from high-load nodes are sent to the target node through the transmission path to complete online dynamic adjustment.

2. The method as described in claim 1, characterized in that, The step of determining nodes in a high-load state using a threshold comparison method based on the current storage usage ratio of each node includes: Compare the current storage usage ratio of each node with a preset first threshold; The nodes whose storage usage ratio exceeds the preset first threshold are identified as high-load nodes.

3. The method as described in claim 1, characterized in that, The process of analyzing the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task includes: Based on the storage status of the adjacent processor nodes of the high-load state node, calculate the storage occupancy ratio of each adjacent processor node; Processor nodes with storage occupancy ratios less than a preset second threshold are selected as candidate nodes, and a candidate node set is constructed from these candidate nodes. Based on the candidate node set, the collected fluctuation results are organized to obtain the trajectory of the change in the occupancy ratio of each node; The method of extracting trend data based on the change trajectory of the occupancy ratio is used to determine whether it indicates potential load fluctuations, and to obtain a resource assessment of candidate nodes that integrate fluctuation trends; Target nodes are selected from candidate node resource evaluations that incorporate fluctuation trends using load balancing algorithms, reinforcement learning algorithms, or game theory methods.

4. The method as described in claim 3, characterized in that, The target node is selected from the candidate node resource evaluation based on the fusion fluctuation trend using a load balancing algorithm, including: The current storage availability ratio is calculated based on the current storage occupancy ratio of candidate nodes according to the fusion fluctuation trend. Based on trend data, linear regression slope or exponential smoothing algorithm is used to predict the changing trend of the current storage availability ratio and determine the trend coefficient. Based on the current storage availability ratio, trend coefficient, current storage occupancy ratio of candidate nodes, and transmission overhead between candidate nodes and high-load nodes, combined with the comprehensive evaluation value calculation formula, the comprehensive evaluation value of each candidate node is obtained. Select the candidate node with the highest comprehensive evaluation value as the target node.

5. The method as described in claim 4, characterized in that, The formula for calculating the comprehensive evaluation value is as follows: In the formula, For comprehensive evaluation, The weights are the proportion of current available storage. This represents the current percentage of available storage. This is the trend coefficient. The weights of the trend coefficients, This represents the current storage usage percentage of the candidate nodes. The weights of the currently available proportion of candidate nodes are stored. To reduce transmission overhead between high-load state nodes, Weights for transmission overhead between high-load state nodes.

6. The method as described in claim 1, characterized in that, The process of sending a portion of the data packets from the high-load node to the target node via the transmission path based on the encrypted transmission protocol to complete online dynamic adjustment includes: Obtain the communication link between the target node and the high-load state node, determine the data migration path, and obtain the migration task allocation scheme; According to the migration task allocation scheme, some data packets of the high-load node are sent to the target node through an encrypted transmission protocol, and it is determined whether state fluctuations occur during the transmission process. When state fluctuations occur, adjust the migration rate to obtain confirmation of completed migration data. Confirm the data that has been migrated, update the storage status records of each node, and complete the online dynamic adjustment.

7. The method as described in claim 3, characterized in that, After transmitting a portion of the data packets from the high-load node to the target node via the transmission path based on an encrypted transmission protocol to complete online dynamic adjustment, the process also includes: Based on the resource assessment of candidate nodes with fused fluctuation trends, an extended assessment index is generated by integrating the sequence of trend data with the judgment results of potential load fluctuations. The resource allocation results are verified using a consistency check method to obtain the consistency check results. Determine whether the extended evaluation metric matches the consistency verification result. If they do not match, trigger an additional verification loop; otherwise, do not trigger an additional verification loop.

8. An online dynamic adjustment system for control parameters, characterized in that, include: The data processing module is used to collect storage usage data from each processor node, and uses a distributed collection protocol to process state fluctuations in a multi-node environment to obtain the current storage occupancy ratio of each node. The status judgment module is used to determine the high-load nodes based on the current storage usage ratio of each node using a threshold comparison method; The target determination module is used to analyze the storage status of processor nodes adjacent to the high-load node to determine the target node for the migration task. The data migration module is used to send a portion of the data packets from high-load nodes to the target node via the transmission path based on an encrypted transmission protocol, thereby completing online dynamic adjustment.

9. An electronic device, characterized in that, include: At least one processor and memory; The memory and processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, an online dynamic adjustment method for a control parameter as described in any one of claims 1 to 7 is implemented.

10. A readable storage medium, characterized in that, It contains an execution program, which, when executed, implements an online dynamic adjustment method for control parameters as described in any one of claims 1 to 7.

Images