A cluster error correction parameter learning method and system

Abstract

This invention specifically relates to a cluster error correction parameter learning method and system, encompassing the fields of distributed cluster communication and error correction technology. It includes: a cell state space definition module; a dynamic neighborhood topology construction module; a local transition rule engine module; an asynchronous random update engine module; and an entropy monitoring and stability detection module. In this invention, cell neighbors are defined along both physical and logical dimensions to accurately adapt to dynamic cluster changes such as node online / offline status, link failures, and replica migration. Furthermore, direct reading from hardware registers enables microsecond-level acquisition of physical parameters. Discretized encoding and lightweight custom messages significantly reduce communication and computational overhead, meeting the core requirement of lightweight cluster interaction.

Description

Technical Field

[0001] This invention relates to the field of distributed cluster communication and error correction technology, and in particular to a cluster error correction parameter learning method and system. Background Technology

[0002] In distributed cluster communication scenarios, physical layer transmission is susceptible to errors due to factors such as link interference, hardware aging, and network congestion. Accurate error correction parameter configuration is crucial for ensuring reliable data transmission and stable business operation. Current mainstream cluster error correction schemes mostly adopt a centralized scheduling architecture, where a central node uniformly collects the entire cluster status, calculates error correction parameters, and issues configuration commands. This model not only makes the central node a performance bottleneck, with communication and computational overhead increasing exponentially with cluster size, but also poses a single point of failure risk. Furthermore, error correction rules are mostly triggered by static thresholds, failing to consider the local neighborhood status of the cluster for collaborative decision-making. This can easily lead to resource congestion caused by excessive local error correction or high error rates due to insufficient error correction. The rationality of resource allocation and system robustness are insufficient to meet the application requirements of large-scale distributed clusters.

[0003] While existing technologies have attempted to apply the distributed self-organizing characteristics of cellular automata to cluster management, the engineering implementation of this model in error correction scenarios has not yet been achieved. Core engineering problems such as standardized definition of cell states and dynamic adaptation of neighborhood topology have not been solved. In addition, traditional error correction schemes all adopt the method of synchronously updating error correction configuration across the entire network, which is prone to instantaneous impact on power and network, causing the system to fall into periodic oscillations. Frequent configuration switching will also wear out hardware encoders / decoders.

[0004] Furthermore, the industry lacks a quantitative assessment mechanism for the global emergent state of cluster error correction systems, making it impossible to accurately determine the steady state of the system. When the system falls into a disordered and chaotic state due to extreme failures, there are no effective non-intrusive fallback recovery methods, which further limits the real-time performance and stability of cluster error correction. Summary of the Invention

[0005] The purpose of this invention is to provide a cluster error correction parameter learning method and system to solve the above-mentioned problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A cluster error correction parameter learning system includes:

[0008] The cellular state space definition module is configured to define the 3D state vectors of cellular physical perception, error correction configuration, and virtual pheromone, and to discretize and quantize the physical parameters.

[0009] The dynamic neighborhood topology construction module is configured to define cell neighbors in both physical and logical dimensions, dynamically maintain the adjacency matrix and topology structure, and achieve low-overhead interaction between cells using only pheromone states through heartbeat messages.

[0010] The local transition rule engine module is configured with a built-in hardware-level excitation-inhibition-memory decay rule library, which independently completes the state transition decision for the next time step based on the cell's own state and the concentration of pheromones in the neighborhood.

[0011] The asynchronous random update engine module is configured to achieve asynchronous cell updates through a probability activation mechanism, and is equipped with hardware and software latches to prevent state jitter.

[0012] The entropy monitoring and stability detection module is configured to quantitatively evaluate the emergent convergence state of the system, determine the steady state, and broadcast a seed reset to trigger cell reset when a chaotic state is detected.

[0013] Preferably, the cell state space definition module specifically includes:

[0014] Acquire raw physical layer parameters of the node through the hardware register direct read interface;

[0015] Including physical bit error rate, signal-to-noise ratio, I / O queue length, and scalable acquisition link utilization, node CPU utilization, and memory usage;

[0016] Normalize each original continuous parameter to map the value to the [0,1] interval;

[0017] The normalized values ​​are binned at equal intervals and divided into K levels.

[0018] Integrate the discrete levels of all parameters into The vector form;

[0019] Pattern definition: In this context, each mode corresponds to a set of directly callable hardware and software error correction parameters. It is a single-valued discrete quantity.

[0020] Preferably, the method also includes virtual pheromones. calculate:

[0021] Preset physical bit error rate threshold and I / O queue length threshold, and obtain the physical bit error rate and I / O queue length corresponding to n consecutive heartbeat cycles. Subtract the corresponding physical bit error rate threshold and I / O queue length threshold from each physical bit error rate and I / O queue length in turn according to the time series. If the obtained value is greater than 0, it is retained and recorded as the abnormal difference in bit error rate and abnormal difference in queue length.

[0022] Take the three largest abnormal differences in bit error rate and queue length respectively, and calculate the average to obtain the average abnormal difference in bit error rate and the average abnormal difference in queue length;

[0023] After normalizing the mean difference in bit error rate and the mean difference in queue length, these are used as the major and minor semi-axes of an ellipse, respectively, to construct an ellipse model. The area of ​​the ellipse model is then calculated and denoted as the virtual pheromone. ;

[0024] The calculation results are normalized to a scalar range of [0, 100], where 0 represents that the node is not abnormal and 100 represents that the node is on the verge of collapse.

[0025] This is the only publicly disclosed state parameter of a cell, broadcast to its neighbors via heartbeat messages.

[0026] Preferably, the dynamic neighborhood topology construction module specifically includes:

[0027] Configure a sparse adjacency matrix for each cell The matrix dimension is , The total number of cells in the cluster, matrix elements ,in Representative cell non-cellular Neighbors Representative cell For cells Neighbors Neighbor weight coefficient; Assign values ​​based on the neighbor type;

[0028] The adjacency matrix is ​​stored using a linked list sparse storage method, storing only non-zero elements;

[0029] Physical neighbor determination: Based on the cluster hardware topology information database, which pre-stores the rack number and switch port number of all nodes;

[0030] Logical neighbor determination: implemented based on the metadata information base of a distributed file system;

[0031] The physical neighbor list and logical neighbor list are deduplicated and merged to generate the final neighbor list of the current cell. Weight coefficients are then assigned according to the neighbor type, and the sparse adjacency matrix is ​​updated accordingly. .

[0032] Preferably, the method further includes:

[0033] The topology update sub-unit monitors cluster topology change events in real time. When an event is triggered, it automatically updates the adjacency matrix and neighbor list of the current cell.

[0034] The message uses UDP packets, formatted as: Cell ID + Virtual Pheromones +Check bit +Time stamp +Reserved bits;

[0035] The message is broadcast to all nodes in the neighbor list at a preset period, using a no-retransmission mechanism.

[0036] Neighbors are handled using a sliding window averaging method. Value, window size is 5 heartbeat cycles;

[0037] Configure a cache entry for each neighbor to store its latest information. The cached items use an expiration and eviction mechanism, taking into account the values, timestamps, and packet loss counts.

[0038] Preferably, the local transfer rule engine module specifically includes:

[0039] Read its own time from the cell status register. state vector ;

[0040] Read all neighbor times set of states ;

[0041] Input the data into the rule base and calculate according to preset rules and priorities;

[0042] Output the error correction configuration sub-state for the next time step. And update the virtual pheromone state simultaneously. ;

[0043] Will Write to the cell state register to complete one state transition;

[0044] Activating biological rules:

[0045] Calculate the neighborhood-weighted average pheromone concentration and normalize it to the range [0, 100].

[0046] Threshold calibration: Threshold Initialize configurable parameters for the cluster;

[0047] Triggering condition: When And if its own physical error rate does not exceed the normal hardware threshold, the biological rule is triggered; If the limit has been exceeded, it will directly implement its own error correction and improvement without triggering panic transmission.

[0048] Execution logic: Configure the cluster with the highest level of error correction as preset;

[0049] Pheromones Update: After the rule is triggered, itself It sends early warning signals to neighboring areas, creating a chain reaction of panic.

[0050] Preferably, the method further includes:

[0051] Suppressing biological rules:

[0052] Pre-detection: Neighbor error correction status detection, self-status detection;

[0053] Triggering condition: The rule is triggered when both checks are satisfied;

[0054] Execution logic: This reduces its own error correction level by 1 level, freeing up network bandwidth and computing resources;

[0055] Pheromones Update: After the rule is triggered, itself It transmits resource availability signals to neighboring areas;

[0056] Memory decay rule:

[0057] Triggering condition: consecutive The heartbeat cycle did not trigger the activation and inhibition rules, and all of its physical parameters were within the normal hardware threshold.

[0058] Execution logic: Error correction configuration and virtual pheromones decay exponentially;

[0059] Attenuation Termination: When the error correction configuration drops to level 0 or the virtual pheromone drops to 0, attenuation stops and the current state is maintained.

[0060] Preferably, the asynchronous random update engine module specifically includes:

[0061] Define a uniform clock cycle for the cluster It is consistent with the cell state update cycle and heartbeat cycle, serving as the time benchmark for cell updates;

[0062] Each clock cycle Within the cell, generate a random number between 0 and 1; then, combine this random number with the preset activation probability. Comparison, if random numbers Then the cell is awakened and performs state transition calculations; if a random number... Then the cell goes into hibernation, maintaining its current state. It remains unchanged until the next clock cycle;

[0063] When a cell is awakened and it is determined that an error correction configuration switch needs to be performed, the latch sub-unit is updated; if only the pheromone state is updated and the error correction configuration switch is not performed, the latch is not triggered.

[0064] Maintain a state switching counter and time window. Each time a cell performs an error correction configuration switch, the counter is incremented by 1. After the time window ends, the counter is reset. If the counter value is greater than or equal to a preset threshold, it is determined to be a local oscillation.

[0065] Preferably, the entropy monitoring and stability detection module specifically includes:

[0066] Collect the error correction configuration sub-state of all cells in the cluster. All valid fetched data The values ​​are statistically analyzed to generate an error correction configuration level distribution table, recording the number of cells at each level k. And calculate the total number of effective cells. ;

[0067] Transform the error correction configuration level distribution table into a discrete probability distribution. ;

[0068] Calculate the global entropy of the system , each moment global entropy The timestamps are stored in the entropy database to form entropy change curves, providing a data basis for convergence decision and chaotic state detection;

[0069] Based on entropy change curve detection, the amplitude of entropy fluctuation is obtained. With fluctuation frequency ;

[0070] Preset fluctuation range threshold Fluctuation frequency threshold ;when and When this happens, the system is determined to have entered a chaotic state.

[0071] A cluster error correction parameter learning method includes:

[0072] Define a three-dimensional state vector for cellular physical perception, error correction configuration, and virtual pheromone, and discretize and quantize the physical parameters.

[0073] Define cell neighbors, dynamically maintain the topology and adjacency matrix, and achieve low-overhead interaction of pheromone states between cells through heartbeat messages;

[0074] Based on the cell's own state and the concentration of pheromones in its neighborhood, the cell independently completes the state transition decision for the next time step, relying on the built-in hardware-level rule base.

[0075] Asynchronous cell updates are achieved through a probabilistic activation mechanism, coupled with hardware and software latching for anti-jitter.

[0076] The convergence state of the system is evaluated by Shannon entropy quantification, and steady state is determined by two indicators. When a chaotic state is detected, the seed is broadcast to trigger cell reset.

[0077] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0078] 1. This invention defines cell neighbors by both physical and logical dimensions, accurately adapting to dynamic changes in the cluster such as node online / offline, link failure, and replica migration; and achieves microsecond-level acquisition of physical parameters through direct reading of hardware registers, while discretization encoding and lightweight custom messages significantly reduce communication and computing overhead, meeting the core requirements of lightweight cluster interaction.

[0079] 2. This invention designs a hardware-level rule base by integrating a biological excitation-inhibition-memory decay mechanism. All rules are computationally complex and prioritized to achieve independent local decision-making in cells. Excitation rules enable collaborative early warning and error correction for abnormal states, inhibition rules prevent resource congestion caused by excessive error correction, and memory decay rules enable resource release when there are no faults. The three work together to achieve local on-demand allocation of error correction resources, which greatly improves the utilization rate of cluster resources. At the same time, the invention achieves asynchronous cell updates through a probabilistic activation mechanism, and reduces the probability of state jitter to 0 with a dual hardware and software latching mechanism. The oscillation detection subunit can achieve dynamic suppression of local oscillations and neighborhood collaboration, fundamentally eliminating the periodic oscillations of the system. Attached Figure Description

[0080] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0081] Figure 1 This is a system structure diagram of the present invention;

[0082] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0083] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0084] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0085] Example 1

[0086] Its specific implementation method is combined with the appendix Figure 1 and attached Figure 2 Please provide a detailed explanation.

[0087] Appendix Figure 1 This invention provides a structural block diagram of a cluster error correction parameter learning system, which shows the connection relationship between the cell state space definition module and the entropy monitoring and stability detection module, and marks the main functional interaction flow of each module.

[0088] Appendix Figure 2 The flowchart of a cluster error correction parameter learning method provided in this embodiment of the invention illustrates the complete steps from defining cellular physical perception to evaluating the system convergence state using Shannon entropy quantification.

[0089] In this embodiment, it includes:

[0090] The cellular state space definition module is configured to define the 3D state vectors of cellular physical perception, error correction configuration, and virtual pheromone, and to discretize and quantize the physical parameters to provide low-overhead standardized data support for subsequent interactions and state transitions.

[0091] Specifically, it includes:

[0092] This module is the core foundation for the engineering implementation of cellular automata in distributed clusters. It enables the standardized collection, quantization, structured storage, and real-time updating of the state of a single cell (node). It not only defines the complete set of attributes of a cell at a certain moment, but also provides a unified format of engineering data carrier for subsequent neighborhood information interaction and local state transitions.

[0093] At the same time, the mapping between biological genes and computer parameters is realized through the engineering processing of parameters, so that the phenotype (actual operating state) of cells can be quantified, calculated and transmitted.

[0094] Physical sensing sub-state:

[0095] Acquisition layer: Acquires raw physical layer parameters of nodes through hardware register direct read interfaces (such as PCIe / SMBus), without needing to go through the operating system kernel mode forwarding, and the acquisition latency is controlled in the microsecond level;

[0096] The core acquisition parameters include physical bit error rate, signal-to-noise ratio, and I / O queue length. It can expand the acquisition link utilization, node CPU utilization, and memory usage to adapt to the hardware characteristics of different clusters.

[0097] Quantization layer: A discretization algorithm using normalization and equidistant binning is employed to transform continuous numerical values ​​into K discrete levels. The specific steps are as follows:

[0098] Normalize each original continuous parameter, mapping the value to the [0,1] interval, as shown in the formula. ,in The hardware threshold for this parameter (such as physical bit error rate) (the maximum bit error rate that the hardware can tolerate).

[0099] The normalized values ​​are binned at equal intervals and divided into K levels (Example K=6, 0=healthy, 1=minor interference, 2=moderate interference, 3=severe interference, 4=fault warning, 5=on the verge of collapse).

[0100] Integrate the discrete levels of all parameters into The vector form, such as This indicates a healthy physical bit error rate, slight interference with signal-to-noise ratio, and moderate interference with I / O queue length.

[0101] The discretized level values ​​are stored as 4-bit binary numbers (which can cover the level requirements of K≤16), which greatly reduces storage and transmission overhead and is suitable for subsequent lightweight neighborhood communication.

[0102] Error correction configuration sub-state:

[0103] Pattern definition: In this system, each mode corresponds to a set of hardware and software error correction parameters that can be directly called. The parameter combinations are pre-stored in the error correction configuration library of the node and are loaded directly through hardware instructions when switching modes, without any additional computational overhead.

[0104] Exemplary pattern definition:

[0105] (No verification): All error correction algorithms are turned off, and only data is passed through. This method is suitable for nodes that are in perfect health.

[0106] (Basic verification): Parity check is enabled, error correction overhead is ≤5%, suitable for minor interference;

[0107] (Lightweight coding): With Hamming code verification enabled, error correction overhead is ≤10%, suitable for moderate interference;

[0108] (Medium-level coding): RS code verification is enabled, error correction overhead is ≤20%, suitable for heavy interference;

[0109] (High redundancy coding): With strong LDPC code verification enabled, error correction overhead is ≤30%, suitable for fault early warning;

[0110] (Extreme Redundancy): Enables three-replica data backup and LDPC code verification, with error correction overhead ≤50%, suitable for nodes on the verge of collapse.

[0111] It is a single-valued discrete quantity, stored as a 3-bit binary number. The mode switching is triggered by the subsequent local transfer rule engine. After the triggering, the error correction parameters are loaded through hardware-level instructions.

[0112] Also includes virtual pheromones calculate:

[0113] Preset physical bit error rate thresholds and I / O queue length thresholds, and obtain the physical bit error rate and I / O queue length corresponding to n (4≤n, specifically set by staff) consecutive heartbeat cycles. According to the time series, subtract the corresponding physical bit error rate threshold and I / O queue length threshold from each physical bit error rate and I / O queue length. If the obtained value is greater than 0, it is retained and recorded as the abnormal difference in bit error rate and queue length. If it is less than 0, it is discarded.

[0114] Sort each abnormal difference in bit error rate and abnormal difference in queue length in descending order of value, and take the three largest abnormal differences in bit error rate and abnormal differences in queue length to calculate the mean of abnormal difference in bit error rate and mean of abnormal difference in queue length.

[0115] After normalizing the mean difference in bit error rate and the mean difference in queue length, these are used as the major and minor semi-axes of an ellipse, respectively, to construct an ellipse model. The area of ​​the ellipse model is then calculated and denoted as the virtual pheromone. ;

[0116] The calculation results are normalized to a scalar range of [0, 100], where 0 represents that the node is not abnormal and 100 represents that the node is on the verge of collapse, so that the neighboring cells can directly determine the degree of abnormality by the numerical value without additional parsing.

[0117] It is the only publicly disclosed state parameter of a cell, broadcast to the neighborhood via heartbeat messages, stored as an 8-bit unsigned integer, occupying only 1 byte of transmission space, achieving extremely lightweight communication.

[0118] The dynamic neighborhood topology construction module is configured to define cell neighbors in both physical and logical dimensions, dynamically maintain the adjacency matrix and topology structure, and achieve low-overhead interaction between cells using only pheromone states through lightweight heartbeat messages, adapting to the dynamic characteristics of clusters.

[0119] Specifically, it includes:

[0120] This module forms the engineering foundation for implementing local interactions within cells. Its core function is to dynamically generate, maintain, and adaptively update the cell neighborhood topology, while also enabling lightweight neighborhood state interaction. It breaks through the limitations of the fixed chessboard topology of traditional cellular automata, adapting to the irregular physical topology and dynamically changing business topology of actual distributed clusters. Through the definition of neighbors in both physical and logical dimensions, the neighborhood topology not only conforms to the underlying hardware communication characteristics but also matches the error correction requirements of upper-layer distributed services. At the same time, a lightweight heartbeat mechanism enables low-overhead and highly reliable transmission of neighborhood states.

[0121] This module generates an initial adjacency matrix during cluster initialization and maintains the adjacency matrix in real time through a topology update subunit during cluster operation. It also configures a heartbeat sending and receiving subunit and a neighbor state caching subunit for each cell, which are responsible for sending and receiving heartbeat messages and caching and updating neighbor states, respectively. The overall module is a distributed architecture, with each cell independently maintaining its own adjacency matrix and neighbor states, without the participation of a central node.

[0122] Adjacency Matrix Engineering implementation:

[0123] Matrix definition: Assign a sparse adjacency matrix to each cell. The matrix dimension is , The total number of cells in the cluster, matrix elements ,in Representative cell non-cellular Neighbors Representative cell For cells Neighbors Neighbor weight coefficient;

[0124] Sparse storage: Since the number of neighbors of each cell is much smaller than the total number of cells in the cluster (for example, the number of neighbors of each cell is ≤20), a linked list sparse storage method is used to store the adjacency matrix, storing only non-zero elements, which greatly reduces storage overhead;

[0125] The weight is assigned based on the neighbor type. The weight of physical neighbors is 0.7, and the weight of logical neighbors is 0.3 (configurable). If a cell is both a physical neighbor and a logical neighbor, the weights are added together to 1.0, ensuring that neighbors with higher correlation have a greater impact during state interactions.

[0126] Definition and determination algorithm of two-dimensional neighbors:

[0127] Physical neighbor determination: This is implemented based on a cluster hardware topology information database, which pre-stores the rack number and switch port number of all nodes. The determination algorithm is as follows:

[0128] Read the rack number of the current cell Access switch number ;

[0129] Traverse the cluster nodes and number the racks as follows Or the switch number is The node is determined to be a physical neighbor;

[0130] Exclude its own node and nodes with link failures, and generate a list of physical neighbors.

[0131] Logical neighbor determination: Based on the metadata information repository of the distributed file system, taking HDFS as an example, the determination algorithm is as follows:

[0132] Read the numbers of all HDFS Blocks carried by the current cell. ;

[0133] Retrieve each from the metadata repository The node where all replicas reside;

[0134] These replica nodes are the logical neighbors of the current cell;

[0135] Exclude the node itself and its replicas from failing nodes, and generate a list of logical neighbors.

[0136] The physical neighbor list and logical neighbor list are deduplicated and merged to generate the final neighbor list of the current cell. Weight coefficients are then assigned according to the neighbor type, and the sparse adjacency matrix is ​​updated accordingly. ;

[0137] Topology dynamic update mechanism:

[0138] The topology update sub-unit monitors cluster topology change events in real time. When an event is triggered, it automatically updates the adjacency matrix and neighbor list of the current cell. Triggering events include:

[0139] Node going online or offline (physical topology change);

[0140] Link failure / recovery (physical topology change);

[0141] Distributed file system replica migration / reconstruction (logical topology change);

[0142] Node rack / switch migration (physical topology change); update latency ≤1s, ensuring that the neighborhood topology is consistent with the actual cluster state.

[0143] Engineering design of the heartbeat detection unit:

[0144] Message design: A custom lightweight UDP message is used, with a total message length of only 12 bytes. The format is: Cell ID (4 bytes) + Virtual pheromone. (1 byte) + check bit (1 byte) + timestamp (4 bytes) + reserved bits (2 bytes), with no redundant fields, achieving the ultimate small message transmission;

[0145] Transmission strategy: Broadcast messages to all nodes in the neighbor list at a preset period (example 100ms), using a no-retransmission mechanism (because...). (Updated in real time, with the latest value rebroadcast in the next cycle), further reducing communication overhead;

[0146] Packet loss tolerance: Due to the possibility of slight packet loss in UDP packets, the neighbor state buffer subunit uses a sliding window averaging method to handle neighbors. The value is set at a window size of 5 heartbeat cycles, which allows for accurate perception of the true status of neighbors even if individual packets are lost.

[0147] Communication overhead quantification: Each cell sends a 12-byte UDP packet every 100ms. If the number of neighbors of each cell is ≤20, the heartbeat communication bandwidth of a single cell is ≤19.2Kbps, and the overall heartbeat communication overhead of the cluster accounts for ≤0.1%, achieving the technical effect of extremely low communication overhead.

[0148] State Cache: The neighbor state cache sub-unit is configured with a cache entry for each neighbor to store its latest state. Values, timestamps, and packet loss counts are used in the cache. The cached items adopt an expiration and eviction mechanism. If a neighbor fails to send a message for 10 consecutive heartbeat cycles, it is determined to be a neighbor failure and is temporarily removed from the neighbor list until communication is restored.

[0149] The local transition rule engine module is configured with a built-in hardware-level excitation-inhibition-memory decay rule library, which independently completes the state transition decision for the next time step based on the cell's own state and the concentration of pheromones in the neighborhood.

[0150] Specifically, it includes:

[0151] This module is the core decision-making unit of the entire distributed emergent control system, corresponding to the state transition function of the cellular automaton. The core implementation enables local independent decision-making and real-time updates of cell states. Each cell is configured with an independent rule engine instance. The engine has a built-in engineered rule library that integrates biological excitation-inhibition mechanisms. Without the need for central node scheduling, it can independently determine the state vector for the next time step using only local information from its own state and the states of its neighbors. The calculation is performed, and all rules are hardware-level calculations with O(1) complexity, ensuring the real-time nature of the decision and its feasibility in engineering implementation; at the same time, the rule base is configurable to adapt to the error correction needs of different clusters.

[0152] The engineering execution steps of the state transition equation:

[0153] State transition equation The execution is completed entirely within the cell. The specific steps are as follows:

[0154] The state-aware subunit reads its own time from the cell state register. state vector (The core is the physical bit error rate) );

[0155] Read all neighbor times from the neighbor state cache subunit. set of states (The core is each neighbor) With weight );

[0156] The rule matching and execution subunit inputs the above data into the rule base and performs calculations according to preset rules and priorities;

[0157] Output the error correction configuration sub-state for the next time step. And update the virtual pheromone state simultaneously. ;

[0158] The status output subunit will Write to the cell state register to complete one state transition;

[0159] The rule base adopts a priority-based hierarchical architecture, condition matching, and numerical calculation. All rules are implemented using hardware-level logic circuits with O(1) complexity. They can be executed independently or in combination. The rule priorities from high to low are: suppression rules > activation rules > memory decay rules, ensuring that the system can make the optimal decision when multiple rules are triggered simultaneously. The specific engineering implementation of each rule is as follows:

[0160] Activating biological rules:

[0161] Pre-calculation: The rule matching execution subunit first calculates the neighborhood weighted average pheromone concentration. To traverse the current cell All neighboring cells Normalize to the range [0,100] to eliminate the influence of the number of neighbors on the calculation results;

[0162] Threshold calibration: Threshold Initialize configurable parameters for the cluster and calibrate through cluster load testing (example) (This is triggered when the average anomaly level in the neighborhood exceeds 60), and can be modified according to the cluster hardware tolerance.

[0163] Triggering condition: When And if its own physical error rate does not exceed the normal hardware threshold, the biological rule is triggered; If the limit has been exceeded, it will directly implement its own error correction and improvement without triggering panic transmission.

[0164] Execution logic: Configure the cluster with the highest level of error correction (example) ), ensuring that the error correction level does not exceed the limit;

[0165] Pheromones Update: After the rule is triggered, itself (Not exceeding 100), transmit early warning signals to neighboring areas to achieve a chain reaction of panic;

[0166] By mimicking the lateral inhibition mechanism of organisms, we can achieve local on-demand allocation of cluster resources, prioritizing the supply of resources to nodes with higher anomaly levels.

[0167] Suppressing biological rules:

[0168] Pre-detection: First, complete two detections:

[0169] Neighbor error correction status detection: Iterate through all neighbors and determine whether 100% of the neighbors are in the correct state. Error correction level;

[0170] Self-state detection: Self Whether it is within the normal hardware threshold and whether the link utilization exceeds 80% (congestion judgment threshold, configurable);

[0171] Triggering conditions: When both of the above checks are met, the rule is triggered to determine that the current subnet is experiencing resource congestion due to high redundancy error correction;

[0172] Execution logic: It reduces its own error correction level by 1 level, down to a minimum of level 0 (no verification), thereby freeing up network bandwidth and computing resources;

[0173] Pheromones Update: After the rule is triggered, itself (Not exceeding 0), transmit resource availability signals to neighboring areas;

[0174] By mimicking the lateral inhibition mechanism of organisms, we can achieve local on-demand allocation of cluster resources, prioritizing the supply of resources to nodes with higher anomaly levels.

[0175] Memory decay rule:

[0176] Triggering condition: consecutive One heartbeat cycle (example) If the excitation and inhibition rules are not triggered within 1 second, and all physical parameters are within the normal hardware threshold, it is determined that there is no new fault stimulus.

[0177] Execution logic: The error correction configuration sub-state and the virtual pheromone sub-state decay exponentially, as shown in the following formulas: , ,in Attenuation coefficient (example) (Configurable) To round down, ensure the state is an integer;

[0178] Attenuation Termination: When the error correction configuration sub-state Descended to level 0 or virtual pheromone state When the value drops to 0, the decay stops and the current state is maintained.

[0179] The rule loading subunit supports modifying core rule parameters through cluster initialization configuration files without modifying hardware logic, improving system adaptability. Configurable parameters include:

[0180] Threshold for triggering rules ;

[0181] Congestion determination threshold (link utilization) for suppression rules;

[0182] The decay coefficient of the memory decay rule With trigger cycle ;

[0183] The priority of each rule;

[0184] The step size for pheromone updates (e.g., +10 after the trigger rule is activated).

[0185] The asynchronous random update engine module is configured to achieve asynchronous cell updates through a probability activation mechanism, and is equipped with hardware and software latches to prevent state jitter, suppressing the power and network impacts caused by system periodic oscillations and network-wide synchronous updates;

[0186] Specifically, it includes:

[0187] This module is the key module for achieving global emergent stability of the system. It simulates the asynchronous and non-uniform physiological activity characteristics of organisms. Through an engineered asynchronous random update mechanism, it avoids the problem of simultaneous state switching of the entire network caused by traditional synchronous updates, fundamentally eliminating the possibility of the system falling into periodic oscillations. At the same time, it prevents instantaneous power / network shocks and protects the cluster hardware devices.

[0188] In addition, the state transition stabilization mechanism avoids cell state jitter caused by small fluctuations in neighborhood information, thereby improving the stability and reliability of state transition.

[0189] Global clock synchronization subunit:

[0190] It provides a unified lightweight clock reference for the entire cluster, adopts a simplified version of the NTP protocol, uses multiple idle nodes in the cluster as clock sources to synchronize the clock to all cells, and the clock synchronization error is ≤1ms.

[0191] Define a uniform clock cycle for the cluster (Example) =100ms), consistent with the cell state update cycle and heartbeat cycle, serving as the time benchmark for cell updates, ensuring that the update timing of all cells is on the same time scale, and avoiding time chaos;

[0192] Engineering implementation of probabilistic activation mechanism:

[0193] Each clock cycle Inside, the probability activation subunit generates a random number between 0 and 1 for the current cell (implemented through a hardware random number generator, with high randomness and no pseudo-random bias).

[0194] The random number is compared with the preset activation probability. (Example) =0.7 (configurable) comparison, if random number Then the cell is awakened and performs state transition calculations; if a random number... Then the cell goes into hibernation, maintaining its current state. It remains unchanged until the next clock cycle;

[0195] The oscillation detection subunit monitors the cell's state switching frequency in real time. If frequent state switching is detected (e.g., ≥5 switching times within 1 second), it is determined to be local oscillation and automatically... Reduce the value to 0.3 to decrease the number of activations and suppress oscillations; if the oscillations disappear, restore it to the initial value. ;

[0196] Engineering design for updating latches:

[0197] When a cell is awakened and the rule matches the execution subunit's determination that an error correction configuration switch needs to be performed (i.e. If only the pheromone state is updated and the error correction configuration switch is not performed, then latching will not be triggered.

[0198] Latching implementation: A dual latching mechanism using hardware register latching and software semaphores is employed.

[0199] At the hardware level: Lock the error correction configuration register and prohibit any write operations until the latch duration ends;

[0200] At the software level: Set a latch semaphore to prevent the rule engine from triggering new error correction configuration switching until the semaphore is released.

[0201] The latch duration L is a configurable parameter (example L=500ms). It is calibrated through cluster stress testing to ensure that it is greater than the hardware execution time of error correction configuration switching (≤1ms) to avoid interference from new status information during the switching process.

[0202] Anti-jitter effect implemented: The latching mechanism ensures that within L milliseconds after a cell completes its error correction configuration switch, it will not perform a new switch due to small fluctuations in neighborhood information. Through stress testing, the probability of cell state jitter can be reduced to 0, avoiding the wear and tear on the hardware encoder / decoder caused by frequent switching and improving the lifespan of the hardware.

[0203] The oscillation detection subunit maintains a state transition counter and a time window (exemplary time window = 1s). Each time the cell performs an error correction configuration switch, the counter is incremented by 1. After the time window ends, the counter is reset. If the counter value is ≥ a preset threshold (exemplary 5 times), it is determined to be a local oscillation.

[0204] Suppression measures: In addition to dynamically reducing the activation probability In addition, the decay coefficient λ of the memory decay rule will be temporarily increased to 0.2 to accelerate state decay and allow cells to quickly recover to a stable state. If local oscillations spread to the neighborhood (multiple neighbors experience oscillations), an oscillation suppression signal will be broadcast to the neighborhood, and neighbor cells will decrease synchronously. This enables neighborhood-based cooperative inhibition.

[0205] The entropy monitoring and stability detection module is configured to evaluate the emergent convergence state of the system using Shannon entropy quantification, determine the steady state through dual indicators, and broadcast a reset seed to trigger cell reset when a chaotic state is detected, thereby achieving non-invasive global observation and fallback recovery.

[0206] Specifically, it includes:

[0207] This module is a lightweight global observer of the entire decentralized system. Its core implementation includes quantitative evaluation of the global emergence effect of the system, real-time determination of the convergence state, and detection and fallback handling of chaotic states. This module has no control authority and does not participate in the local state transition decision of any cell. It only provides a global perspective to the system through distributed entropy calculation and state detection.

[0208] Simultaneously, a fallback recovery mechanism for chaotic states is designed. When the system falls into a chaotic state due to extreme failures and cannot achieve emergent convergence through local rules, a lightweight global intervention is used to allow the system to start evolving again, ensuring the final stability of the system. This module is designed non-intrusively, does not occupy cluster business resources, and does not affect the normal operation of the system.

[0209] Engineering implementation of the Shannon entropy calculation unit:

[0210] Collect the error correction configuration sub-state of all cells in the cluster. No other state parameters are collected, because the distribution of Scfg directly reflects the global error correction state of the cluster and is the core indicator for evaluating the emergence effect.

[0211] Data Acquisition Method: A lightweight pull mode is adopted, and the entropy acquisition sub-unit pulls data from all cells at a preset period (example period = 1 second, much longer than the heartbeat period). The value is that the retrieved message is a custom UDP message, only 8 bytes long, with extremely low communication overhead.

[0212] If a cell fails to be retrieved (e.g., due to a fault or offline status), then it will be... Values ​​marked as invalid and not included in entropy calculation, and the accuracy of entropy calculation is not affected when the percentage of failed nodes is ≤10%;

[0213] All valid data retrieved The values ​​are statistically analyzed to generate an error correction configuration level distribution table, recording the number of cells at each level k (0≤k≤m). And calculate the total number of effective cells. ;

[0214] Transform the error correction configuration level distribution table into a discrete probability distribution. , , representing the proportion of cells in the cluster that are configured at level k to the total number of effective cells;

[0215] Calculate the system's global entropy using the Shannon entropy formula. , It uses a base-2 logarithm, and the entropy value is measured in bits.

[0216] The physical meaning of entropy:

[0217] The smaller the value, the more concentrated the distribution of error correction configuration levels in the cluster, the closer the system is to a global steady state, and the better the emergent convergence effect.

[0218] The larger the value, the more discrete the error correction configuration level distribution of the cluster, the more disordered the system is, and the worse the emergent convergence effect.

[0219] This means that the error correction configuration level of all cells is completely consistent, and the system reaches an absolute steady state.

[0220] Each moment global entropy The timestamps are stored in the entropy database to form entropy change curves, providing a data basis for convergence decision and chaotic state detection;

[0221] Engineering implementation of convergence decision logic:

[0222] Decision Indicators: Two core indicators are used to determine whether the system has converged to a global steady state. Both indicators are configurable parameters.

[0223] The rate of change of entropy characterizes the trend of entropy change; the rate of change of entropy is calculated as follows: , The period representing the entropy value calculation period;

[0224] The total bit error rate (TBER) of a cluster represents its actual operating status; the total bit error rate of a cluster... ,in The business weight of the node;

[0225] Decision thresholds: Two thresholds are preset, including the entropy change rate threshold. Total Bit Error Rate Safety Threshold ,when When the entropy value tends to stabilize, it is determined that the entropy value is stable; when When the cluster is in normal operating condition, it is determined that the cluster is running normally.

[0226] When both indicators are satisfied simultaneously and the entropy value remains unchanged for M consecutive calculation cycles (for example, M=10, i.e., 10s), it is determined that the entire system has reached the global optimal equilibrium state through local interaction and self-organization.

[0227] After convergence is determined, a convergence completion signal is broadcast to the entire network. Upon receiving the signal, the cells will adjust their activation probabilities. Reduce it to 0.3 to decrease the number of state updates and maintain system steady state.

[0228] Detection and fallback handling of chaotic states:

[0229] Based on entropy change curve detection, the amplitude of entropy fluctuation is obtained. With fluctuation frequency ;

[0230] Fluctuation amplitude: Calculate the difference between the maximum and minimum entropy values ​​over 10 consecutive periods;

[0231] Fluctuation frequency: Calculates the number of extreme points of entropy within 10 seconds;

[0232] Preset fluctuation range threshold Fluctuation frequency threshold ;

[0233] when and When the system enters a chaotic state, it is determined that the system has entered a chaotic state, that is, the local rules of the system have failed due to extreme failures (such as large-scale link failures or sudden network-wide bit errors), and the system cannot achieve emergent convergence through local interactions.

[0234] Only when the state is determined to be chaotic after three consecutive entropy value calculation cycles will a fallback process be performed to avoid misjudgment caused by accidental entropy value fluctuations.

[0235] Chaotic state fallback mechanism:

[0236] The chaotic state processing subunit generates a global reset seed, which is a 64-bit random number and is unique, ensuring that the initial state is different each time it is reset;

[0237] Reset Seed Broadcast: Broadcast the chaotic state and reset seed signal to the entire network via the cluster multicast protocol. The multicast message is a custom UDP message with a length of only 16 bytes, ensuring that all normal cells in the network can receive it.

[0238] Upon receiving the reset seed signal, the cell immediately performs a state reset operation:

[0239] Error correction configuration sub-state Reset to Mode0 (no parity);

[0240] virtual pheromone state Reset to 0;

[0241] Activation probability Restore to initial value;

[0242] Clear the state transition counter and oscillation detection flag;

[0243] The state evolution is restarted by resetting the seed as the initial value of the hardware random number generator.

[0244] After the cells are reset, the system re-interacts and transitions through local rules, emerging again from the disordered state. After stress testing, the system can converge back to the global steady state within 30 seconds after the reset.

[0245] Example 2

[0246] Please see Figure 2 A cluster error correction parameter learning method includes the following parts:

[0247] Define cellular physical perception, error correction configuration, and virtual pheromone three-dimensional state vectors, and discretize and quantize physical parameters to provide standardized data support;

[0248] Cell neighbors are defined by both physical and logical dimensions, the topology and adjacency matrix are dynamically maintained, and low-overhead interaction of pheromone states between cells is achieved through lightweight heartbeat messages.

[0249] Based on the cell's own state and the concentration of pheromones in its neighborhood, the cell independently completes the state transition decision for the next time step, relying on the built-in hardware-level rule base.

[0250] Asynchronous cell updates are achieved through a probabilistic activation mechanism, coupled with hardware and software latching to prevent jitter and suppress system oscillations and resource impacts caused by synchronous updates.

[0251] The convergence state of the system is evaluated by Shannon entropy quantification, the steady state is determined by two indicators, and when a chaotic state is detected, the seed is broadcast to trigger cell reset, so as to achieve global observation and fallback recovery.

[0252] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0253] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0254] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0255] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0256] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0257] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0258] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0259] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0260] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0261] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A cluster error correction parameter learning system, characterized in that, include: The cellular state space definition module is configured to define the 3D state vectors of cellular physical perception, error correction configuration, and virtual pheromones, and to discretize and quantize the physical parameters, specifically including: The raw physical layer parameters of the node are collected through the hardware register direct read interface, including physical bit error rate, signal-to-noise ratio, I / O queue length, link utilization, node CPU utilization, and memory usage. Normalize each original continuous parameter to map the value to the [0,1] interval; The normalized values ​​are binned at equal intervals and divided into K levels. Integrate the discrete levels of all parameters into The vector form, It is in vector form of the physical sensing substate; Pattern definition: In this context, each mode corresponds to a set of directly callable hardware and software error correction parameters. It is a single-valued discrete quantity. This represents the set of all error correction modes preset by the system. Also includes virtual pheromones calculate: Preset physical bit error rate threshold and I / O queue length threshold, and obtain the physical bit error rate and I / O queue length corresponding to n consecutive heartbeat cycles. Subtract the corresponding physical bit error rate threshold and I / O queue length threshold from each physical bit error rate and I / O queue length in turn according to the time series. If the obtained value is greater than 0, it is retained and recorded as the abnormal difference in bit error rate and abnormal difference in queue length. Take the three largest abnormal differences in bit error rate and queue length respectively, and calculate their average values ​​to obtain the average abnormal difference in bit error rate and queue length. When the number of the largest abnormal differences in bit error rate or queue length is less than three, take all of them. After normalizing the mean difference in bit error rate and the mean difference in queue length, these are used as the major and minor semi-axes of an ellipse, respectively, to construct an ellipse model. The area of ​​the ellipse model is then calculated and denoted as the virtual pheromone. ; The calculation results are normalized to a scalar range of [0, 100], where 0 represents that the node is not abnormal and 100 represents that the node is on the verge of collapse. This is the only publicly disclosed state parameter of a cell, which is broadcast to its neighboring domain via a heartbeat message. The dynamic neighborhood topology construction module is configured to define cell neighbors in both physical and logical dimensions, dynamically maintain the adjacency matrix and topology structure, and realize the interaction between cells using pheromones as the payload through heartbeat messages. Specifically, it includes: Physical neighbor determination: Based on the cluster hardware topology information database, which pre-stores the rack number and switch port number of all nodes; Logical neighbor determination: implemented based on the metadata information base of a distributed file system; The physical neighbor list and the logical neighbor list are deduplicated and merged to generate the final neighbor list of the current cell; The local transition rule engine module is configured with a built-in hardware-level excitation-inhibition-memory decay rule library. All rules are implemented by hardware-level logic circuits and independently complete the state transition decision for the next time step based on the cell's own state and the concentration of pheromones in the neighborhood. The asynchronous random update engine module is configured to achieve asynchronous cell update through a probability activation mechanism that randomly wakes up cells with a preset activation probability, and is equipped with hardware and software latches to prevent state jitter. The entropy monitoring and stability detection module is configured to quantitatively evaluate the emergent convergence state of the system, determine the steady state, and broadcast a seed reset to trigger cell reset when a chaotic state is detected.

2. The cluster error correction parameter learning system according to claim 1, characterized in that, The dynamic neighborhood topology construction module specifically includes: Configure a sparse adjacency matrix for each cell The matrix dimension is , The total number of cells in the cluster, matrix elements ,in Representative cell non-cellular Neighbors Representative cell For cells Neighbors Neighbor weight coefficient; Assign values ​​based on the neighbor type; The adjacency matrix is ​​stored using a linked list sparse storage method, storing only non-zero elements; Assign weight coefficients based on neighbor type and update the sparse adjacency matrix. .

3. The cluster error correction parameter learning system according to claim 2, characterized in that, Also includes: The topology update sub-unit monitors cluster topology change events in real time. When an event is triggered, it automatically updates the adjacency matrix and neighbor list of the current cell. The message uses UDP packets, formatted as: Cell ID + Virtual Pheromones +Check bit +Time stamp +Reserved bits; The message is broadcast to all nodes in the neighbor list at a preset period, using a no-retransmission mechanism. Neighbors are handled using a sliding window averaging method. Value, window size is 5 heartbeat cycles; Configure a cache entry for each neighbor to store its latest information. The cached items use an expiration and eviction mechanism, taking into account the values, timestamps, and packet loss counts.

4. The cluster error correction parameter learning system according to claim 1, characterized in that, The local transfer rule engine module specifically includes: Read its own time from the cell status register. state vector ; Read all neighbor times set of states ; Input the data into the rule base and calculate according to preset rules and priorities; Output the error correction configuration sub-state for the next time step. And update the virtual pheromone state simultaneously. ; Will Write to the cell state register to complete one state transition; Activating biological rules: Calculate the neighborhood-weighted average pheromone concentration and normalize it to the range [0, 100]. Threshold calibration: Threshold Configure parameters for cluster initialization; Triggering condition: When And if its own physical error rate does not exceed the normal hardware threshold, the biological rule is triggered; If the limit has been exceeded, it will directly implement its own error correction and improvement without triggering panic transmission. Execution logic: , Configure the cluster with the highest level of error correction as preset; Pheromones Update: After the rule is triggered, itself It sends early warning signals to neighboring areas, creating a chain reaction of panic.

5. A cluster error correction parameter learning system according to claim 4, characterized in that, Also includes: Suppressing biological rules: Pre-detection: Neighbor error correction status detection, self-status detection, as detailed below: Neighbor error correction status detection: Iterate through all neighbors and determine whether 100% of the neighbors are in the correct state. Error correction level; Self-state detection: Self Is it within the normal hardware threshold, and does the link utilization exceed 80%? Triggering condition: When both detections are met, the rule is triggered to determine that the current subnet is experiencing resource congestion due to high redundancy error correction; Execution logic: This reduces its own error correction level by 1 level, freeing up network bandwidth and computing resources; Pheromones Update: After the rule is triggered, itself It transmits resource availability signals to neighboring areas; Memory decay rule: Triggering condition: consecutive The heartbeat cycle did not trigger the activation and inhibition rules, and all of its physical parameters were within the normal hardware threshold. Execution logic: Error correction configuration and virtual pheromones decay exponentially; Attenuation Termination: When the error correction configuration drops to level 0 or the virtual pheromone drops to 0, attenuation stops and the current state is maintained.

6. The cluster error correction parameter learning system according to claim 1, characterized in that, The asynchronous random update engine module specifically includes: Define a uniform clock cycle for the cluster It is consistent with the cell state update cycle and heartbeat cycle, serving as the time benchmark for cell updates; Each clock cycle Within the cell, generate a random number between 0 and 1; then, combine this random number with the preset activation probability. Comparison, if random numbers Then the cell is awakened and performs state transition calculations; if a random number... Then the cell goes into hibernation, maintaining its current state. It remains unchanged until the next clock cycle; When a cell is awakened and it is determined that an error correction configuration switch needs to be performed, the latch sub-unit is updated; if only the pheromone state is updated and the error correction configuration switch is not performed, the latch is not triggered. Maintain a state switching counter and time window. Each time a cell performs an error correction configuration switch, the counter is incremented by 1. After the time window ends, the counter is reset. If the counter value is greater than or equal to a preset threshold, it is determined to be a local oscillation.

7. The cluster error correction parameter learning system according to claim 1, characterized in that, The entropy monitoring and stability detection module specifically includes: Collect the error correction configuration sub-state of all cells in the cluster. All valid fetched data The values ​​are statistically analyzed to generate an error correction configuration level distribution table, recording the number of cells at each level k. And calculate the total number of effective cells. ; Transform the error correction configuration level distribution table into a discrete probability distribution. , , representing the proportion of cells in the cluster that are configured at level k to the total number of effective cells; Calculate the system's global entropy using the Shannon entropy formula. , ; Each moment global entropy The timestamps are stored in the entropy database to form entropy change curves, providing a data basis for convergence decision and chaotic state detection; Based on entropy change curve detection, the amplitude of entropy fluctuation is obtained. With fluctuation frequency ; Preset fluctuation range threshold Fluctuation frequency threshold ;when and When this happens, the system is determined to have entered a chaotic state.

8. A method for learning clustered bit error correction parameters, characterized in that, include: The cellular state space definition steps, configured to define the cellular physical perception, error correction configuration, and virtual pheromone three-dimensional state vector, and to discretize and quantize the physical parameters, specifically include: The raw physical layer parameters of the node are collected through the hardware register direct read interface, including physical bit error rate, signal-to-noise ratio, I / O queue length, link utilization, node CPU utilization, and memory usage. Normalize each original continuous parameter to map the value to the [0,1] interval; The normalized values ​​are binned at equal intervals and divided into K levels. Integrate the discrete levels of all parameters into The vector form, It is in vector form of the physical sensing substate; Pattern definition: In this context, each mode corresponds to a set of directly callable hardware and software error correction parameters. It is a single-valued discrete quantity. This represents the set of all error correction modes preset by the system. Also includes virtual pheromones calculate: Preset physical bit error rate threshold and I / O queue length threshold, and obtain the physical bit error rate and I / O queue length corresponding to n consecutive heartbeat cycles. Subtract the corresponding physical bit error rate threshold and I / O queue length threshold from each physical bit error rate and I / O queue length in turn according to the time series. If the obtained value is greater than 0, it is retained and recorded as the abnormal difference in bit error rate and abnormal difference in queue length. Take the three largest abnormal differences in bit error rate and queue length respectively, and calculate their average values ​​to obtain the average abnormal difference in bit error rate and queue length. When the number of the largest abnormal differences in bit error rate or queue length is less than three, take all of them. After normalizing the mean difference in bit error rate and the mean difference in queue length, these are used as the major and minor semi-axes of an ellipse, respectively, to construct an ellipse model. The area of ​​the ellipse model is then calculated and denoted as the virtual pheromone. ; The calculation results are normalized to a scalar range of [0, 100], where 0 represents that the node is not abnormal and 100 represents that the node is on the verge of collapse. This is the only publicly disclosed state parameter of a cell, which is broadcast to its neighboring domain via a heartbeat message. The dynamic neighborhood topology construction steps are configured to define cell neighbors according to both physical and logical dimensions, dynamically maintain the adjacency matrix and topology structure, and realize the interaction between cells with pheromones as the payload through heartbeat messages. Specifically, these steps include: Physical neighbor determination: Based on the cluster hardware topology information database, which pre-stores the rack number and switch port number of all nodes; Logical neighbor determination: implemented based on the metadata information base of a distributed file system; The physical neighbor list and the logical neighbor list are deduplicated and merged to generate the final neighbor list of the current cell; The local transition rule engine steps are configured with a built-in hardware-level excitation-inhibition-memory decay rule library. All rules are implemented by hardware-level logic circuits and independently complete the state transition decision for the next time step based on the cell's own state and the concentration of pheromones in the neighborhood. The asynchronous random update engine steps are configured to achieve asynchronous cell updates through a probability activation mechanism that randomly wakes up cells with a preset activation probability, combined with hardware and software latches to prevent state jitter. The entropy monitoring and stability detection steps are configured to quantitatively evaluate the emergent convergence state of the system, determine the steady state, and broadcast a seed reset to trigger cell reset when a chaotic state is detected.

Images