A node task migration and scheduling system based on digital twinning

Abstract

The application provides a node task migration and scheduling system based on digital twinning, relates to the technical field of computer system structure and data processing, and quantifies the computing resource interference in the multi-tenant shared environment by sensing the cross-tenant noise coefficient and state synchronization complexity entropy in the node micro-architecture; extracts the spatial discrete variance of the mobile terminal trajectory feature to generate an adaptive migration decision hysteresis factor, which is converted into a decision damping to suppress the invalid high-frequency reciprocating migration; a digital twinning sandbox is constructed before physical cutting, a composite fault operator is injected into the bottom layer by a chaotic scene injection engine, and a fidelity integrator is used to calculate the risk-adjusted service continuity retention degree; finally, the bottom layer controller is executed by the safe baseline comparison linkage to perform physical flow cutting. The application completes network boundary deduction without consuming physical bandwidth, avoids the interruption risk caused by state hard switching, and effectively guarantees the smooth transition of stateful services.

Description

Technical Field

[0001] This invention relates to the field of computer system architecture and data processing technology, specifically to a node task migration and scheduling system based on digital twins. Background Technology

[0002] With the evolution of edge computing and distributed system architecture, a large number of high-computing-power-consuming defensive computing tasks are being deployed to widely distributed edge computing nodes. These heterogeneous computing nodes, carrying complex computing tasks, collectively constitute a defense system, with each physical or virtual computing entity in the system serving as a defense node. The core concept of this system is "computing power fusion and resource scheduling": logically merging these geographically dispersed, heterogeneous underlying hardware computing, storage, and I / O resources into a computing power pool through virtualization and resource scheduling engines. Its goal is to match the optimal underlying computing environment in real time for each dynamic business flow or computing session in the system, achieving task computation and state processing with the lowest latency. However, in mobile computing scenarios, when the main body of the task request (such as a terminal roaming across regions or a dynamically routed data flow) undergoes a high-speed spatial relocation, to ensure the continuity of computing tasks, the "stateful service instance" carrying the business must be seamlessly migrated across different defense nodes. Such stateful services typically maintain a large memory context, a complex process execution tree, and a continuously updated state machine. The computing node currently hosting the service instance is called the source defense node j, and the node about to take over the context state is called the target defense node i.

[0003] When performing stateful system-level service migrations, existing technologies, particularly traditional resource schedulers, rely primarily on macroscopic operating system-level telemetry data such as overall CPU utilization and remaining memory space. This approach lacks awareness of the underlying micro-architecture's operational status and cannot penetrate into the compute nodes to assess the actual performance degradation caused by micro-level resource contention in multi-tenant concurrent environments (such as cache pollution in multi-core CPUs, memory bus contention across non-uniform memory access (NUMA) architectures, and I / O scheduling starvation). Furthermore, it fails to quantify the lack of "computing affinity" between different physical nodes due to differences in instruction set compatibility and memory page replacement strategies. This computational allocation based on macroscopic "blind box" metrics is tantamount to random scheduling in a performance minefield. Existing migrations are mostly passively triggered, lacking pre-simulated dirty page synchronization, which easily leads to severe "context switching cliffs" (instantaneous computational vacuum periods). Meanwhile, due to the lack of predictive logic based on sandbox simulation, when faced with sudden anomalies in underlying resources (such as I / O interruption blocking causing the target node storage to "freeze" or state data block write timeout), the failure rate of cross-node memory state transfer remains high, which can easily damage the reliability of the computing power system. Summary of the Invention

[0004] The purpose of this invention is to provide a node task migration and scheduling system based on digital twins to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A node task migration and scheduling system based on digital twins includes:

[0007] The topology and state awareness module is configured to construct an opportunity migration domain map containing multiple candidate target defense nodes for the source defense node serving the mobile terminal; detect the link characteristics and micro-architecture state from the source defense node to each candidate target defense node to calculate the link resilience coefficient and computing power affinity index, and generate a dynamic candidate migration path set.

[0008] The predictive migration decision module is configured to evaluate and generate a net migration benefit index for each candidate migration path based on a dynamic candidate migration path set, combined with the predicted trajectory of the mobile terminal and the underlying operating characteristics of the candidate target defense nodes; select the one with the highest net migration benefit index as the optimal target defense node, and generate a migration control scheme that includes a state-progressive synchronization strategy and a traffic switching window.

[0009] The simulation and evaluation module is configured to build a digital twin sandbox in the system memory according to the migration control scheme, simulate the gravitational potential energy transfer process from state pre-synchronization to complete business traffic switching in the digital twin sandbox, extract composite fault scenarios from the preset disturbance library and inject them into the digital twin sandbox, calculate the instantaneous user experience fidelity function through definite integral accumulation operation, and combine multi-scenario probability aggregation to accurately calculate the service continuity retention degree.

[0010] The decision execution and network cutover module is configured to compare the service continuity retention with a preset security baseline threshold. When the service continuity retention is greater than or equal to the security baseline threshold, it issues a physical flow table switching command to the underlying network controller to perform a bidirectional physical cutover of network status and traffic. When the service continuity retention is less than the security baseline threshold, it triggers the underlying circuit breaker mechanism and destroys the migration control scheme.

[0011] Compared with the prior art, the beneficial effects of the present invention are:

[0012] This invention introduces cross-tenant noise figure and state synchronization complexity entropy to deeply probe memory bus contention, I / O queue starvation, and branch prediction interference in the microarchitecture of computing nodes, eliminating reliance on easily distorted macroscopic operating system load metrics. This mechanism successfully quantifies and avoids the "performance dark matter effect" in multi-tenant shared environments, effectively preventing millisecond-level instruction pauses or process freezes caused by underlying microarchitecture resource contention after stateful services undergo cross-node hot migration, ensuring accurate and stable supply of computing resources.

[0013] This invention calculates the spatial discrete variance by extracting the time series of mobile terminal speed and heading angle, and uses an inverse activation function to generate an adaptive migration decision hysteresis factor, transforming high-frequency trajectory clutter in physical space into mathematical damping for scheduling decisions. This mechanism imposes penalty resistance on the lingering behavior of terminals at base station boundaries, improves the stability of routing decisions during network topology changes, effectively reduces the migration pendulum effect caused by terminal directional shifts, and reduces frequent back-and-forth migrations without service benefits and wide area network link instability.

[0014] This invention constructs a digital twin sandbox and coordinates it with a chaotic fault injection engine. Before performing a real memory context cutover, it injects complex system-level fault operators such as I / O interrupt blocking, memory synchronization timeout, and state machine drift into the sandbox's underlying layer. It then uses definite integrals to precisely calculate the risk-adjusted "service continuity retention." This mechanism completes extreme stress simulation of the scheduling architecture without consuming actual hardware computing power. It enhances the system's resilience to fault takeover in the face of sudden hardware anomalies, avoids process crashes caused by traditional direct "hard switching," and promotes a smooth transition and continuous operation of massive memory states between nodes. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the overall architecture and operational process of a node task migration and scheduling system based on digital twins.

[0016] Figure 2 This is a schematic diagram of the overall system flow of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0019] Example 1:

[0020] Please see Figures 1 to 2 The present invention provides a technical solution:

[0021] A node task migration and scheduling system based on digital twins is applied in a widely distributed computing network of nodes to perform stateful services that are highly sensitive to computing resources and network latency for high-speed mobile communication terminals, including:

[0022] The topology and state awareness module is configured to construct an opportunity migration domain map containing multiple candidate target defense nodes for the source defense node serving the mobile terminal; detect the link characteristics and micro-architecture state from the source defense node to each candidate target defense node to calculate the link resilience coefficient and computing power affinity index, and generate a dynamic candidate migration path set.

[0023] The topology and state awareness module includes a link quality assessment unit. This unit is configured to analyze in real-time the micro-burst packet loss patterns, long-tail effect of delay distribution, and congestion feedback delay of the protocol stack between the source defense node and candidate target defense nodes. It extracts the variance of the micro-burst packet loss rate, the long-tail ratio of the 99th percentile network delay to the average network delay, and the absolute hysteresis time of the transmission control protocol stack congestion feedback signal within a preset time window. Preset regularization penalty weights are assigned to each weight, and a nonlinear exponential decay model is used to output a link resilience coefficient that characterizes the link's ability to withstand state-synchronized traffic collapse. The increase in the variance of the micro-burst packet loss rate and the long-tail ratio leads to an exponential decay of the link resilience coefficient.

[0024] The topology and state awareness module also includes a computing power compatibility measurement unit and a path index generation unit. The computing power compatibility measurement unit is configured to extract the multi-core consistency protocol of the processor cache, the memory page replacement strategy across nodes with non-uniform memory access architecture, and the application programming interface compatibility of the instruction set of the secure execution environment, and output the computing power affinity index through a Boolean matching measurement function. The path index generation unit is configured to perform a geometric mean operation on the link resilience coefficient calculated by the link quality assessment unit and the computing power affinity index calculated by the computing power compatibility measurement unit to obtain the migration path fidelity index; remove nodes whose index is not higher than the dynamic admission threshold, and merge the remaining nodes into the dynamic candidate migration path set.

[0025] This embodiment is specifically applied to a high-speed mobile communication terminal, which passes through different edge computing base station coverage areas during its movement. The communication terminal maintains a stateful encrypted communication channel with the current source network security protection node (source defense node j). As the communication terminal moves, the scheduling system needs to perceive, in real time, dozens of candidate target defense nodes i within surrounding base stations, assessing whether they possess the physical and network conditions to take over the encrypted channel and its security context state.

[0026] The specific steps executed by the topology and state awareness module are as follows:

[0027] S101. Continuously monitor the widely distributed computing nodes. For the source defense node j that is providing network layer security protection instances (such as stateful encrypted traffic filtering services) for high-speed mobile communication terminals (such as communication terminals carried during high-speed rail travel), delineate the opportunity migration domain map on the geographical and network topology based on the movement vector direction of the communication terminal. The scheduling system sends microarchitecture probes and network link probe messages to all candidate target defense nodes i in the map to obtain the underlying state characteristics.

[0028] S102, Operational Link Quality Assessment Unit, calculates the link resilience coefficient, including:

[0029] Real-time analysis is performed on the micro-burst packet loss patterns, long-tail effects of latency distribution, and congestion feedback delays in the protocol stack between the source defense node j and the target defense node i. Since stateful services require the transmission of large amounts of in-memory state data during migration, even minor network jitter can lead to state synchronization failure. The scheduling system calculates and outputs a link resilience coefficient, characterizing the link's ability to withstand failure, using a nonlinear fitting formula. The calculation formula is as follows:

[0030]

[0031] Here, σloss,i,j represents the variance of the micro-burst packet loss rate of the link from node j to node i within the past preset time window. The packet loss rate itself is a ratio, and its variance is a dimensionless pure number.

[0032] τ 99,i,j With τ avg,i,j Δ represents the 99th percentile network delay and the average network delay of the link between source defense node j and target defense node i, respectively. Both are measured in time (e.g., milliseconds, ms). The ratio of the two is used to quantify the long-tail effect of the delay. Dividing them cancels out the dimensions, resulting in a dimensionless pure number; Δ ack,i,j表征 The absolute delay between the TCP stack congestion feedback signals returned by the source defense node j and the target defense node i, expressed in time (e.g., milliseconds, ms); Δ normThe maximum tolerable congestion feedback delay constant is the preset baseline for the system, with the dimension being time (e.g., milliseconds, ms). The ratio of the two normalizes the feedback delay to a dimensionless pure number; α, β, and γ are the preset regularization penalty weights for the system; the penalty weight for micro-burst packet loss variance is α=2.0, the penalty weight for long-tail delay is β=1.5, and the penalty weight for congestion feedback hysteresis is γ=1.0.

[0033] When the link experiences severe intermittent stuttering (i.e., an increased long-tail effect ratio) or slow congestion feedback, the corresponding penalty term will increase rapidly, leading to a decrease in the link resilience coefficient. (The value ranges between (0,1]) It drops exponentially, thus strictly filtering out inferior network links that "meet the average bandwidth requirements but are extremely unstable at the micro level" in the mathematical model.

[0034] S103. Run the computing power compatibility measurement unit to calculate the computing power affinity index, including:

[0035] The scheduling system quantifies and compares the matching degree between the target defense node i and the source defense node j in terms of underlying hardware microarchitecture. Since security protection instances involve frequent read / write operations and encryption / decryption operations on the underlying memory, differences in microarchitecture can lead to severe performance degradation after migration. Based on the microarchitecture feature evaluation table, the scheduling system extracts processor cache coherence protocols, memory controller scheduling policies, and API compatibility of the Secure Execution Environment (TEE), and calculates the computing power affinity index. .

[0036]

[0037] Where k represents the dimension index of the microarchitectural features, k=1 for cache consistency, k=2 for memory scheduling, k=3 for security environment, and F j,k With F i,k These are the microarchitectural physical feature identifiers that characterize the source defense node j and the target defense node i in the k-th dimension, respectively. This is a Boolean matching metric function that outputs dimensionless discrete pure numbers, such as 1.0 for a perfect match, 0.4~0.8 for partial compatibility, and 0.0 for incompatibility. The weights are the corresponding to the k-th dimension features; the evaluation rules for the computing power affinity index are shown in Table 1.

[0038] Table 1: Reference Table for Preset Standards of Hardware Feature Extraction and Weight Assignment for Computing Power Compatibility

[0039]

[0040] S104, Path Index Generation Unit, generates a dynamic candidate migration path set, including:

[0041] The scheduling system will use the link resilience coefficient calculated above. Computing power affinity index The migration path fidelity index is calculated by performing fusion. :

[0042]

[0043] The use of geometric mean (square root) instead of simple arithmetic mean is to adhere to the "weakest link" principle in computing systems. When the link is extremely unstable, Approaching 0 or completely incompatible with computing power When the index approaches 0, the overall fidelity index will be forcibly flattened to an extremely low value by multiplication, thus preventing the risk of a single index being artificially high and masking the significant defects of another index.

[0044] Set an entry threshold, denoted as P. threshold The preferred value is 0.70. The migration path fidelity index of each candidate path is compared. The conditions will be met. The target defense node i is incorporated into the dynamic candidate migration path set and directly proposed. The specific data extrapolation for the nodes is shown in Table 2.

[0045] Table 2: Candidate Node Path Index Generation and Initial Screening Deduction Table

[0046]

[0047] In this embodiment, a deep, low-level probing and physical state assessment are performed through a topology and state awareness module. At the network channel assessment level, the scheduling system drills down to the micro-network state, extracting in real time the variance of micro-burst packet loss rate, the long-tail effect ratio of delay distribution, and the protocol stack congestion feedback hysteresis time. After relativizing and dimensionlessizing the delay parameters with physical units, a nonlinear exponential decay mathematical model is used to amplify the negative weight of micro-network jitter, thereby calculating the link resilience coefficient to capture potential network collapse risks affecting state synchronization. At the computing power assessment level, the scheduling system delves into the hardware micro-architecture, quantifying and comparing the CPU cache coherence protocol, non-uniform memory access scheduling strategy, and secure execution environment instruction set of the source and target defense nodes. A quantified computing power affinity index is output through a Boolean matching metric function and a weighted algorithm. Based on the principle of the weakest link in the computing system, the scheduling system performs a geometric average operation on the link resilience coefficient and the computing power affinity index to generate a dimensionless migration path fidelity index, and removes nodes below the dynamic admission threshold to generate a dynamic candidate migration path set. The beneficial effects of the above technical solution are as follows: by introducing a nonlinear exponential penalty mechanism, the system can accurately filter out inferior communication links that meet the average bandwidth requirements but have unstable micro-states; through the physical coupling algorithm of micro-architecture-level computing power matching and geometric average, it can prevent the risk of a single macro-indicator being artificially high and masking deep compatibility defects, and provide rigorous, high-fidelity and highly feasible basic decision support for the smooth migration of stateful services in wide-area edge networks.

[0048] Example 2:

[0049] The predictive migration decision module, connected to the topology and state awareness module, is configured to evaluate and generate a net migration benefit index for each candidate migration path based on the opportunity migration domain map, and combine the motion vector prediction of the mobile terminal to select the optimal target defense node i, and generate a migration control scheme that includes a state progressive synchronization strategy and a traffic switching window.

[0050] The predictive migration decision module further includes:

[0051] The path initial screening unit is configured to remove candidate migration paths whose migration path fidelity index is lower than the dynamically adjusted admission threshold.

[0052] The migration benefit calculation unit is configured to calculate, for the remaining paths after the path screening unit has been screened, the difference between the expected latency reduction brought about by migrating to each candidate target defense node i within a future time window T and the instantaneous service interruption penalty caused by the migration, in combination with the predicted trajectory of the mobile terminal.

[0053] The optimal decision unit is configured to combine the migration path fidelity index with the difference to calculate the final net migration benefit index, and select the node with the highest index as the optimal target defense node.

[0054] The predictive migration decision module also includes a node service quality analysis engine, configured to construct a node service quality vector that integrates multi-dimensional features before generating a migration control plan. This vector includes at least: state synchronization complexity entropy, cross-tenant noise coefficient, service recovery time gradient, and migration decision hysteresis factor.

[0055] The calculation of cross-tenant noise figure is based on the monitoring of the memory bus contention intensity of candidate target defense node i, the starvation probability of the I / O scheduling queue, and the interference of defense tasks running on it on the branch prediction of the CPU pipeline, in order to assess the "performance dark matter" effect caused by microarchitectural resource contention.

[0056] The migration decision lag factor is an adaptive parameter used to proactively delay migration decisions when the uncertainty of the mobile terminal trajectory is high, in order to avoid the "migration pendulum" effect caused by prediction errors.

[0057] The predictive migration decision module executes the logic of selecting the optimal target defense node and generating a migration control plan, which is implemented through the following units:

[0058] The risk prediction unit is configured to predict the failure rate risk of the service quality score of the source defense node j falling below the preset SLA in the future, based on the predicted trajectory of the mobile terminal and the node service quality vector.

[0059] The migration triggering unit is configured to select the node with the highest net migration benefit index from the candidate migration path set as the optimal target defense node i when the failure rate risk exceeds the dynamic threshold, and to start the state progressive synchronization mechanism.

[0060] The takeover capability judgment unit is configured to continuously monitor the convergence deviation of the state replicas between the source defense node j and the target defense node i during synchronization. When the convergence rate of this deviation exceeds the slope of the preset learning curve, the target defense node i is determined to have hot takeover capability.

[0061] Based on the "dynamic candidate migration path set" and its corresponding "migration path fidelity index" in the embodiments, this embodiment will elaborate in detail how the scheduling system executes extremely complex predictive migration decisions. The core of this embodiment aims to solve the three major technical challenges commonly faced by existing technologies when migrating stateful network security protection services (such as distributed web application firewalls with complex context session memory, anti-DDoS connection tracing and cleaning instances, etc.) for high-speed mobile terminals (such as intelligent logistics heavy truck platoons in automatic cruising state) in wide-area distributed computing nodes: "microarchitecture resource trampling (performance dark matter effect)," "invalid repeated migration caused by trajectory fluctuations (migration pendulum effect)," and "business disruption caused by state synchronization."

[0062] The scheduling system, acting as a single, strictly executing entity, performs the following deeply coupled algorithmic steps through its internally integrated predictive migration decision module:

[0063] S201, the running node service quality analysis engine quantifies "performance dark matter" and constructs a node service quality vector. The specific principle is as follows: In traditional edge cloud scheduling strategies, the system often relies solely on macroscopic, shallow indicators such as "CPU utilization" or "remaining memory" reported by the target defense node's operating system kernel. However, in multi-tenant shared edge computing nodes, even a node with only 30% CPU utilization may have its underlying memory bus bandwidth or L3 cache completely exhausted by other high I / O-intensive tenant tasks. This causes newly migrated stateful services to suffer severe "performance dark matter," resulting in millisecond-level or even second-level instruction pauses.

[0064] To completely eliminate this hidden danger, the scheduling system constructs the node service quality vector V of the target defense node i before generating the migration control plan. QoS (i). The construction of this vector involves the following high-dimensional computational sub-steps:

[0065] S2011, Calculate the cross-tenant noise figure N cross (i), obtained in the following way:

[0066] The scheduling system directly reads the microarchitecture running status of the target defense node i by issuing underlying hardware performance-monitor-counters (PMC) probes, collecting three core underlying parameters: memory bus contention intensity, I / O scheduling queue starvation probability, and CPU pipeline branch prediction interference rate. To ensure consistency of computational dimensions, the scheduling system first performs dimensionless processing on each physical sample value, and then calculates the cross-tenant noise figure using the following nonlinear weighted aggregation formula. :

[0067]

[0068] Among them, B mem,used (i) and B mem,max (i) represents the current memory bus bandwidth occupied by target defense node i and the theoretical maximum bandwidth, respectively, both in bytes per second (Byte / s). Their ratio constitutes the dimensionless base rate of memory contention; c is the system-preset bus congestion penalty index, with a preferred value of 1.5; utilizing the exponential amplification effect, when memory bandwidth usage exceeds a specific threshold, the penalty term increases non-linearly; P starve (i) represents the proportion of starved requests (dimensionless pure number, value range [0,1]) in the underlying I / O scheduling queue of target defense node i whose request waiting time exceeds the preset tolerance limit (e.g., 10ms); R miss (i): Represents the branch prediction failure rate (Branch-Misprediction-Rate) caused to the CPU pipeline by other concurrent tasks running on the target defense node i. This value is a dimensionless pure number; κ is a scalar of branch prediction interference sensitivity, with an optimal value of 5.0; μ1, μ2, and μ3 are the normalized fusion weights of the three microarchitectural interference features, respectively, and satisfy μ1+μ2+μ3=1. The memory bus bandwidth contention weight μ1=0.45, the I / O starvation probability weight μ2=0.35, and the branch prediction interference weight μ3=0.20 (totaling 1.0); cross-tenant noise coefficient The final output is a dimensionless pure number in the interval [0,1]. The closer the value is to 1, the more severe the "performance dark matter" effect is at the bottom layer of the target defense node i. That is, the node appears to be idle on the surface, but in fact, its internal microarchitecture resources have been severely overdrawn. If defense services that perform high-frequency regular expression matching or encryption / decryption calculations are migrated to this node, they will face catastrophic performance degradation.

[0069] S2012, Calculate the entropy of state synchronization complexity E sync (j) and the service recovery time gradient G recover (i) Specifically, this includes: Since stateful services contain a large number of TCP session connection tables and hacker attack tracing state machines, the scheduling system needs to evaluate the real-time mutation rate of these state memory pages on the source defense node j. Based on the generation rate of dirty pages, the scheduling system calculates the state synchronization complexity entropy E using information entropy theory. sync (j):

[0070]

[0071] Where H represents the total number of memory blocks allocated to this instance by the source defense node j. The entropy value represents the probability that the h-th memory block is modified within a unit monitoring period; the higher the entropy value, the more active the business state, and the greater the risk that the data consistency generated during migration will be difficult to converge.

[0072] Simultaneously, the scheduling system calculates the service recovery time gradient G of the target defense node i. recover (i) refers to the rate at which the target node reloads the received state data into the CPU cache and restores the execution context.

[0073] Finally, the scheduling system fuses the aforementioned dimensionless features to generate the node service quality vector V for the target defense node i. QoS (i): ;

[0074] S202. For high-speed mobile terminals, their movement trajectories are often constrained by the complexity of the physical road network. If a communication terminal lingers or hovers at low speed at the edge of densely populated base station coverage overlap areas in cities (such as large intersections or roundabouts), the traditional greedy scheduling algorithm based on "shortest distance or strongest signal" will cause stateful services to migrate frequently and ineffectively between multiple adjacent edge defense nodes. This is a migration pendulum effect. To completely eliminate the "migration pendulum effect," the scheduling system incorporates a high-order kinematic evaluation engine to calculate the migration decision hysteresis factor H. hyst (v,θ) actively applies mathematical damping to unstable topological changes. The specific calculation steps are as follows:

[0075] The scheduling system extracts the instantaneous velocity scalar time series V={v1,v2,...,vn} and the heading angle time series Θ={θ1,θ2,...,θn} of the communication terminal within the current time window ΔW.

[0076] Calculate the spatial discrete variance of the terminal motion vector, or simply the trajectory variance. :

[0077] Where Var() represents calculating the statistical variance of the sequence; and System-calibrated speed and heading weighting coefficients. Configuration speed variance weighting. Heading deviation variance weight In edge computing topologies, a terminal's "acceleration and deceleration (velocity variance)" typically only changes the time it takes to reach the next base station along the current route, while "frequent turning (heading angle variance)" directly determines that the terminal will repeatedly hop between the coverage sectors of multiple different base stations. Therefore, the scheduling system assigns a higher convergence weight to the heading angle to precisely curb the "migrating pendulum effect" caused by directional shifts.

[0078] By using the inverse sigmoid activation function, the trajectory variance is mapped to a dynamic transfer decision hysteresis factor. :

[0079]

[0080] Among them, H max The maximum damping coefficient preset for the system is ζ (preferred value is 5.0); ζ is the damping response kurtosis coefficient (pure number); the value range of the damping response kurtosis coefficient ζ is [0.5, 2.0], with a specific preferred value of ζ=1.2; ζ determines the "sensitive filter window" of the scheduling system for changes in the trajectory of communication terminals. If the value of ζ is too small (e.g., 0.2), the hysteresis factor grows too slowly, and the system cannot apply sufficient damping in time when the terminal enters the intersection, which may still trigger misjudgment migration; if the value of ζ is too large (e.g., 5.0), the function will degenerate into a step function, causing the system's scheduling decision to have abrupt mathematical oscillations (jumps) between "allow" and "lock". When the preferred value is 1.2, it can effectively mitigate the hysteresis of the trajectory variance. It provides a rapid penalty boost when breaking through the critical point, while ensuring the continuous smoothness of the calculated gradient, perfectly filtering out the occasional glitches and jitters of the GPS signal. This is the critical trajectory disorder variance threshold defined by the system. When the communication terminal is traveling at a constant speed in a straight line on a highway or railway track, the trajectory variance is small. After substituting into the formula, the exponential term tends to positive infinity, and the denominator tends to infinity. At this time, the hysteresis factor (undamped state) encourages the scheduling system to migrate nodes quickly and decisively to ensure low network latency. Conversely, when the terminal is turning at low speed under a complex overpass, or when GPS signal drift causes a large trajectory variance, the exponential term tends to 0 after substituting into the formula. This factor will be used as a penalty term in subsequent revenue calculations, forcibly lowering the expected revenue of migration, thereby actively "freezing" migration commands. The system is required to tolerate slightly higher transmission latency of the source node before the terminal trajectory is completely stable, rather than triggering high-risk invalid state migrations.

[0081] S203. Run the migration benefit calculation unit to calculate the net migration benefit index. Specifically, for the remaining safe paths after being screened by the "path initial screening unit" in Example 1 (i.e., removing candidate migration paths whose migration path fidelity index is lower than the dynamically adjusted entry threshold), the scheduling system begins to calculate the difference between the expected benefit and the instantaneous penalty that can be brought by migrating to each candidate target defense node i within the future time window T. The scheduling system calculates the difference between the expected latency reduction and the service interruption penalty ΔBenefit(i) based on the following mathematical integral and penalty deduction combination formula:

[0082]

[0083] in, This represents the predicted physical network interaction delay (in milliseconds, ms) between the communication terminal and the source defense node j within a future time t, assuming no migration occurs. Ψ(t) represents the predicted physical network interaction delay (in milliseconds) between the communication terminal and the target defense node i within a future time t, assuming the migration is completed; Ψ(t) is the service sensitivity time distribution function (a dimensionless pure number). Among them, 1.0 is the basic sensitivity constant, which ensures that even during periods of low-frequency requests, the reduction in physical latency can still generate basic benefits. Let $t$ be the predicted number of concurrent network requests (QPS, times / second) from the terminal at a future time $t$. This is the upper limit of the peak concurrent users recorded in the history of this terminal; The sensitivity amplification extremum coefficient is preferably set to 2.5;

[0084] If, within the predicted time period (for example, an autonomous heavy truck is about to enter a complex road condition, it needs to conduct high-frequency road network collaborative perception with the cloud). Approaching the peak value, Ψ(t) will be amplified to a maximum of 3.5 (i.e., 1.0 + 2.5 × 1.0). In mathematical integration, this is equivalent to a "gravitational accelerator," forcing the scheduling system to recognize that even a reduction of only 5ms in a single network latency would result in enormous cumulative business benefits, thereby actively promoting high-quality node migration.

[0085] The integral term represents the total amount of communication time (in milliseconds, ms) saved for the application layer over the entire time window T if the service is migrated to the target defense node i, which is closer to the terminal's physical topology. A larger value indicates a greater benefit from the network proximity advantage brought about by the migration. C base The base time constant (in milliseconds) represents the inherent computational overhead of a single complete state memory page synchronization. The overall physical meaning of the penalty term is that the product of three dimensionless pure numbers and the base time constant results in an output representing the "equivalent penalty time for service lag caused by the migration operation itself" (in milliseconds). If the underlying "performance dark matter" of the target defense node i is rampant, and the state modification of the source node is extremely drastic, coupled with the communication terminal being in a state of trajectory chaos, this penalty term will expand dramatically.

[0086] The above calculation results are nonlinearly mapped to normalized pure numbers to obtain the difference evaluation score SΔ(i), whose range is mapped to [ Within the interval [1,1]. If the score is less than 0, it indicates that the cost of implementing the migration plan far exceeds the network reduction benefits it brings, making it a losing proposition.

[0087] S204. Run the optimal decision-making unit, integrate and output the net migration benefit index, specifically, the migration path fidelity index P obtained in Example 1 is used as the net migration benefit index. fid (i,j) (a pure number representing the matching degree between the micro-bursts and the micro-architecture of the underlying network), and the difference evaluation score S calculated in this embodiment. Δ (i) Orthogonally fused (pure numbers representing macroeconomic business revenue expectations).

[0088] The fusion formula uses the weighted harmonic mean principle to calculate the final net migration gain index Inet(i):

[0089]

[0090] Among them, w p The bias weight is used to adjust the decision skew ratio between underlying fidelity and surface returns. A solution that cuts off all negative returns; To prevent small positive numbers with a denominator of zero, a bias weight w is applied. p The value range is [0.5, 1.5], and the preferred value is w. p =0.8 The harmonic mean is inherently biased towards the smaller value (the weakest link effect). p This is used to disrupt the absolute balance. Because stateful network security protection services (such as DDoS mitigation nodes) have a "zero-tolerance" attitude towards the failure of the underlying operation, the system would rather sacrifice some of the benefits of network proximity access (i.e., surface benefits SΔ(i)) than compromise the stability of the underlying micro-architecture (i.e., migration path fidelity index P). fid (i,j)). Set w p =0.8 (less than 1), which mathematically corresponds to giving the migration path fidelity index P a value of 0.8. fid (i,j) are given a higher priority of "veto power" to ensure the safety and reliability of the incoming node;

[0091] Traverse all nodes in the dynamic candidate migration path set and select the net migration benefit index. The highest value is confirmed as the optimal target defense node i* for this scheduling. The specific multi-dimensional data deduction logic is shown in Table 3.

[0092] Table 3: Full-Scale Simulation and Derivation Table for High-Dimensional Predictive Transfer Decision Making

[0093]

[0094] S205. Even if the optimal target defense node i* is selected, the scheduling system does not blindly and immediately execute a hard switch. To ensure that the network security service of the communication terminal is in an "absolutely continuous" state, the system executes a refined risk prediction and state synchronization takeover process.

[0095] S2051. The SLA failure rate risk prediction scheduling system predicts the signal attenuation of the physical channel between the mobile terminal and the source defense node j due to increased distance, based on the high-speed departure trajectory of the mobile terminal. Combining the Weibull distribution reliability model, it calculates the failure rate risk function R of the source defense node j, which indicates that its service quality score will fall below the preset SLA (Service-Level Agreement, set to a maximum tolerable delay of 50ms) within a future time t. fail (t):

[0096]

[0097] in, This is the timescale constant for SLA failure characteristics calculated based on the current terminal's distance from the target location. For environmental decay shape parameters;

[0098] S2052. Set a risk tolerance threshold R that is dynamically adjusted based on the type of terminal service. th (Preferred range [0.65, 0.85]). The failure rate risk R calculated through real-time monitoring... fail When (t)>Rth, the scheduling system formally triggers the migration instruction to the optimal target defense node i* and forcibly starts the "state progressive synchronization mechanism".

[0099] Explanation of the physical principle: The traditional "Stop-and-Copy" mechanism would cause a complete interruption of business operations during the copying process. The scheduling system adopts a progressive synchronization mechanism, that is, while the source defense node j continues to process business operations, an encrypted tunnel is silently established in the background to transfer a full snapshot of the baseline memory of the source node j to the target node i*. During the baseline transfer, any new business state changes (dirty pages) generated by the source node j are stored in a separate iterative tracing matrix.

[0100] S206, the takeover capability judgment unit, performs millisecond-level smooth traffic switching; after entering the state progressive synchronization stage, the core task of the scheduling system is to find a brief "golden traffic switching window" to achieve seamless business cutover.

[0101] S2061. With multiple iterative incremental synchronizations in the background, the state memory copy on the target node i* becomes increasingly closer to the real-time state of the source node j. The scheduling system continuously calculates the state copy convergence deviation D at the end of the f-th synchronization round. conv (f). Since the source node j is continuously processing new business and generating dirty pages while synchronizing, the convergence deviation is physically represented as "the volume of residual dirty data that has not yet been synchronized (unit: megabytes MB)". The Markov chain state equation for the convergence deviation is expressed as:

[0102]

[0103] in, The dynamic variation rate (MB / s) of the source node's service status. The synchronous transmission bandwidth (MB / s) reserved for the backend. This refers to the time consumed per round.

[0104] S2062. Calculate the slope of the learning curve and determine the hot-swap capability. Calculate the discrete derivative of the convergence deviation between two consecutive iterations, i.e., the slope of the learning curve. conv (f):

[0105]

[0106] When Slope conv (f) The value remains negative (indicating that the residual dirty data is rapidly shrinking), and the absolute value convergence rate exceeds the preset extreme slope threshold, while also satisfying the remaining absolute deviation. When a single atomic transmission can be completed within a very small interval (e.g., less than 2ms, i.e., reaching the microsecond-level memory copy limit) when a network request initiated by the terminal is initiated, the takeover capability judgment unit of the scheduling system determines that the hot takeover capability of the target defense node i* is qualified.

[0107] The scheduling system sends OpenFlow route update flow tables to the SDN (Software-Defined Networking) controller of the edge network. The instruction physically switches the uplink network traffic of the communication terminal to the target defense node i* within milliseconds, completing the smooth takeover of the entire stateful service. The takeover state simulation is shown in Table 4.

[0108] Table 4: Real-time Tracking Deduction Table for State Progressive Synchronization and Hot-Connection Judgment

[0109]

[0110] In this embodiment, the system collects memory bus contention intensity, input / output scheduling queue starvation probability, and branch prediction interference rate from the underlying hardware performance counters. After dimensionless processing, cross-tenant noise coefficients are generated, and a node service quality vector is constructed by combining it with state synchronization complexity entropy to quantify the contention state of microarchitectural resources. The system extracts the speed and heading time series of the mobile terminal to calculate the spatial discrete variance, maps it to a migration decision hysteresis factor through an inverse sigmoid activation function, and applies dynamic mathematical damping to the fluctuating physical trajectory. Subsequently, the system performs time integration on the expected latency reduction bonus, deducts the instantaneous service penalty time derived from the aforementioned noise coefficients and hysteresis factors, and substitutes the difference and migration path fidelity index into the harmonic mean formula with bias weights to output the net migration benefit index to lock the optimal target defense node. The system combines a reliability model to predict service degradation risk to trigger progressive state synchronization, calculates the convergence deviation of residual dirty data in real time based on the Markov chain state equation, and performs physical flow switching when the slope of the convergence learning curve meets the single network request interval transmission condition. The beneficial effects of the above technical solution are as follows: By introducing cross-tenant noise coefficients, the system reduces its dependence on macro-level resource indicators, establishes a quantitative evaluation mechanism for contention of underlying memory and input / output buses, and prevents instruction pauses caused by resource overdraft after service migration; by utilizing dynamic migration decision lag factors, physical trajectory fluctuations are transformed into scheduling penalty resistance, avoiding frequent back-and-forth migrations and network routing jitter caused by terminal direction shifts; during the state cutover phase, progressive memory iterative tracing is performed based on convergence models and slope judgment criteria, controlling the handover gaps of stateful services to a small number of residual dirty pages, ensuring a smooth transition and continuous operation of business traffic between distributed nodes.

[0111] Example 3:

[0112] The simulation and evaluation module, connected to the predictive migration decision module, is configured to perform digital twin inference based on the opportunity migration domain map, simulate the service gravity field transfer process from state pre-synchronization startup to complete business traffic switching, and calculate the service continuity retention rate that is imperceptible to users within a preset evaluation period.

[0113] The simulation and evaluation module further includes a service fidelity integrator configured to integrate the instantaneous user experience fidelity function over time over the entire time interval from triggering pre-synchronization to completing traffic switching.

[0114] The instantaneous user experience fidelity function takes the effective interaction frame rate and the proportion of requests that do not time out during the migration process as positive inputs, and the overhead of "network shadow traffic" generated during state synchronization and the opportunity cost of the target defense node i due to resource reservation as negative penalty terms.

[0115] The simulation and evaluation module also includes a chaotic scene injection engine, configured to perform a disturbance-resistant expectation evaluation, the specific functions of which are:

[0116] Complex fault scenarios are injected into digital twin simulations from a disturbance library containing edge network "gray rhino" events such as network link partitions and node "death".

[0117] Complex fault scenarios include:

[0118] The first Phantom Bandwidth Scenario refers to a complex network instability scenario where the link from the source defense node j to the target defense node i has a second-level average bandwidth that meets the migration requirements, but the millisecond-level instantaneous throughput jitter variance exceeds the preset high-frequency oscillation threshold (such as 30% of the average), resulting in the actual time taken for state synchronization being far longer than expected.

[0119] The second scenario of a seemingly dead node refers to a situation where the target defense node i has normal system heartbeat and CPU load, but its I / O service latency for probing state write requests of standard size (e.g., 4KB) continuously exceeds a preset intolerable threshold (e.g., 500 milliseconds), resulting in a "stuck" internal service degradation during the migration process.

[0120] The third pendulum trap scenario refers to a situation where the network topology optimal source defense node changes more than twice within two consecutive decision cycles (e.g., within 10 seconds) due to the unstable trajectory of the mobile terminal. If the system frequently initiates migration based on this, it will lead to the continuous consumption of resources in the ineffective migration process, resulting in "decision jitter".

[0121] This embodiment is implemented through deep coupling of simulation and evaluation modules and chaotic injection deduction. Specifically, it is based on the "migration control scheme (including optimal target defense node i*, state progressive synchronization strategy and traffic switching window)" generated by the predictive migration decision module in Embodiment 2. In order to prevent catastrophic business collapse caused by directly executing physical flow switching in extreme, rare and sudden physical environments, this embodiment will explain in detail how the scheduling system constructs a high-fidelity parallel digital space before issuing this decision to the underlying physical hardware, and accurately calculates the final service continuity retention in the form of mathematical integral by implementing an extremely destructive "gray rhino" event exercise.

[0122] As a single, strictly executing entity, the scheduling system performs the following deeply coupled digital twin and chaotic inference steps through its internally integrated simulation and evaluation modules:

[0123] S301. Construct a digital twin sandbox and mapping service gravity field transfer model:

[0124] Upon receiving the migration control plan, the scheduling system does not immediately execute the OpenFlow flow table distribution on the physical side. Instead, it uses lightweight container isolation technology to instantly generate a digital twin sandbox in the system memory for the source defense node j, the target defense node i*, and the current network topology of the communication terminal.

[0125] The scheduling system constructs a mathematically defined "Service-Gravity-Field" model within this twin sandbox. This model treats service state data as a fluid with mass, and the combined computing power of the source and target nodes, along with the network traction force, as gravitational potential energy. The scheduling system calculates the traction potential function Φ(t) of the target defense node i* on the state data at time t:

[0126]

[0127] in, As an indicator of computing power affinity; The synchronous transmission bandwidth reserved for the backend (MB / s); Dtopo(t) is the logical topology hop count between two nodes in the twin network; This represents the divergence operation on the spatial traction vector; by tracking the gradient change of the service's gravitational field potential function, the scheduling system can accurately simulate the dynamic transfer process across the entire time axis from state pre-synchronization start-up (gravitational field begins to tilt) to complete service traffic switching (gravitational field completes potential energy reversal) without consuming physical network bandwidth.

[0128] S302. During the simulation of service gravity field transfer, the scheduling system runs a service fidelity integrator to measure the micro-experience impact of each state copy and bandwidth encroachment on the virtual communication terminal in real time with extremely stringent standards; the scheduling system samples on the time axis and calculates the instantaneous user experience fidelity function F. uext (t). The following nonlinear combination formula with strict hedging of positive gain and negative penalty is adopted:

[0129]

[0130] Among them, FPS act (t) and FPS req τ represents the effective interactive frame rate actually acquired by the simulated terminal at the current moment and the service benchmark required frame rate (unit: frames / second, fps), respectively. The ratio of the two is the dimensionless frame rate compliance index; req (t) and τ SLA These represent the actual response time of the simulated terminal's current network request and the maximum tolerance time specified in the SLA protocol (in milliseconds, ms), respectively. The ratio is dimensionless and incorporating an exponential decay constant. This ensures that if a response times out, the experience gains drop exponentially. (BW)shadow (t) and BW total (t) represents the bandwidth of "shadow traffic" generated covertly in the background to maintain gradual synchronization, and the total link bandwidth (in Mbps), respectively. The ratio constitutes the dimensionless bandwidth occupancy rate; χ² opp (i,τ) represents the rate loss factor (in units of s) of the target defense node i* due to its inability to handle other concurrent tenant tasks because it pre-locked CPU / memory resources to prepare for service migration, resulting in an opportunity cost per unit time. 1 The time integral yields the dimensionless cumulative resource waste pure number; d1, d2, d3, and d4 are preset weights.

[0131] The preferred value for the interaction frame rate gain weight d1 is 0.4. The preferred value for the response latency gain weight d2 is 0.6. Under ideal conditions (i.e., when the actual frame rate is fully compliant and the network response latency approaches 0), the limit sum of the positive gain terms is d1×1 + d2×1 = 1.0. During allocation, the latency weight (0.6) is higher than the frame rate weight (0.4) because, for stateful services in distributed security defense, the continuity of the underlying session's real-time response has a higher priority than the smoothness of the application layer's frame rate. (Exponential decay constant) The preferred value is 3.0, when the response delay is just close to the maximum tolerance limit of SLA, i.e., τ. req (t) / τ SLA When =1, the time delay gain term decays to d2×exp( 3.0), approximately equal to 0.6 × 0.049 ≈ 0.03. This setting causes the positive benefit to drop precipitously to near zero once the latency approaches the protocol's red line.

[0132] The preferred value for the bandwidth occupancy penalty weight d3 is 2.5. The preferred value for the cumulative resource waste penalty weight d4 is 1.8. To ensure that "any slight delay will cause the function value to fall below zero," the penalty weights are set to an amplification factor greater than the positive base value of 1.0. For example, when shadow traffic occupies 40% of the total link bandwidth (dimensionless ratio of 0.4), the negative penalty generated by this item alone reaches 2.5 × 0.4 = 1.0, which will directly offset all the perfect positive gain, forcing the function value to fall into the negative range, thereby triggering the system's "zero tolerance" circuit breaker for the experience judgment of that period in the digital twin simulation.

[0133] Only when the service frame rate is full, the response is extremely fast, and the synchronous shadow traffic is minimal, and the target node has very few idle resources (i.e., "perfect hot takeover"), can the instantaneous fidelity approach 1. Any slight bandwidth preemption or synchronization delay will be rapidly amplified by the negative penalty term, causing the function value to drop below zero.

[0134] S303: Real-world edge networks are far from ideally smooth states; they are riddled with fatal flaws arising from a mix of determinism and randomness. The scheduling system's chaotic scenario injection engine extracts three highly destructive "gray rhino" scenarios from the system's perturbation library and directly injects mathematical perturbation operators into the protocol stack and I / O controller of the digital twin sandbox; including:

[0135] S3031. The jitter threshold scheduling system for the first phantom bandwidth scenario obtains the current average network bandwidth value of the link from the source defense node to the target defense node in the digital twin sandbox; the scheduling system starts the underlying packet sending control script, and within every preset millisecond time window (e.g., 10 milliseconds), it forcibly suppresses the instantaneous packet sending rate of the underlying network port downward, making it lower than the preset jitter threshold (which is set to 30% of the current average network bandwidth value); the scheduling system controls this suppression action to continue for a preset oscillation period within the time window, and then restores the normal rate. By tampering with the packet sending rate based on the threshold, a communication environment with macroscopic connectivity but microscopic high-frequency disconnection is directly generated in the twin sandbox.

[0136] S3032, The I / O latency threshold scheduling system for the second fake dead node scenario sends a keep-alive heartbeat packet to the target defense node in the twin sandbox and confirms that the heartbeat response time returned by its operating system and the CPU load rate are both lower than the preset security baseline; the scheduling system attaches an I / O interception hook in the underlying operating system block device driver of the target defense node; when the scheduling system detects that the sandbox attempts to write a state synchronization data block larger than the preset size (e.g., 4KB) to the target defense node, it immediately triggers the I / O interception hook, forcibly suspending the physical completion time of the data block write operation until the time exceeds the preset fake dead judgment threshold (e.g., a fixed delay of 500 milliseconds); by forcibly modifying the underlying write timestamp in this way, a fault environment in which the node's external indicators are normal but the internal storage bus is blocked is directly generated.

[0137] S3033, The trajectory oscillation threshold scheduling system of the third pendulum trap scenario cuts off the digital twin sandbox's interface for reading the GPS coordinates of the real communication terminal, and instead continuously feeds a set of pre-generated fake coordinate sequences into the sandbox's position input port; the scheduling system controls the generation logic of the fake coordinate sequence: forcing the position coordinates of the communication terminal to continuously cross the physical coverage boundary of two adjacent target defense nodes within a preset continuous observation time window (e.g., 10 seconds); the scheduling system further monitors the boundary crossing behavior and forces the frequency of this crossing behavior to exceed a preset high-frequency oscillation threshold (e.g., set to two or more consecutive boundary crossings in completely opposite directions within 10 seconds), thereby directly causing the optimal target defense node calculated in the sandbox to undergo high-frequency replacement in a short period of time, generating a decision jitter fault environment caused by severe trajectory drift.

[0138] By utilizing the value network of the reinforcement learning and optimization module, and combining the occurrence probability of each fault scenario, the risk-adjusted expectation of the integral value of the instantaneous user experience fidelity function in multiple simulated migrations is calculated. This expected value is the final service continuity retention.

[0139] S304. Run the reinforcement learning value network and the expectation aggregation unit to calculate and output the risk-adjusted service continuity retention expectation value, and perform a rigid bottom line judgment.

[0140] To achieve an absolutely precise mathematical balance between "avoiding the risk of extreme physical network outages" and "maximizing network topology benefits," the scheduling system, after injecting and extrapolating the chaotic scenario in S303, executes the following progressive calculation and execution steps:

[0141] S3041, the runtime environment context awareness and probability prediction unit extracts the context multidimensional feature vector of the current real physical edge network. The feature vector strictly includes: the historical boundary gateway protocol (BGP) routing jitter rate from the source defense node to the target defense node, the nonlinear variance of the current motion vector of the communication terminal, and the real-time computing power load fluctuation frequency of the physical rack where the target defense node is located. The scheduling system generates a normalized scenario occurrence probability distribution set P={P(m0),P(m1),P(m2),P(m3)} through the Softmax output layer of this value network. Among them, the scenario index is strictly defined as n∈{0,1,2,3}, m0 is the ideal smooth transition scenario without injected disturbance, and m1, m2, and m3 are the first phantom bandwidth, the second fake dead node, and the third pendulum trap fault scenario injected in S303, respectively. The scheduling system enforces the sum of the probabilities of the above scenarios to be strictly equal to 1 through a regularization function.

[0142] S3042, Perform time-definite integral calculation of service fidelity for each of the above independent derivation scenarios m n(n∈{0,1,2,3}), the scheduling system operates within the set digital twin evaluation period Teval=t s t d (t) s The time to trigger pre-synchronization, t d Within the planned traffic switching completion time, this continuous time window is uniformly discretized into Z millisecond-level sampling steps Δt; the scheduling system at each discrete sampling step node t z At the location where the step size index z∈{1,2,…,Z}, extract m from S302 for this specific scenario. n Real-time calculated instantaneous user experience fidelity function F uext (t z |m n The service fidelity integrator is run, and definite integral accumulation is performed over the time span according to the Riemann and discrete numerical approximation rules. The accumulated result is divided by the total evaluation period Teval to obtain the absolute time integral experience value E for each specific scenario. score (m n ):

[0143]

[0144] Among them, step size Through the above discrete numerical definite integral and mean-averaging operations, the single millisecond-level experience mutation data caused by high-frequency jitter of the underlying network is smoothed, and the instantaneous experience sequence that fluctuates drastically in the time dimension is tightly compressed and mapped into a dimensionless pure number that represents the comprehensive health level of the communication terminal in this specific exercise scenario throughout the entire migration cycle.

[0145] S3043, Run the expectation aggregation unit to calculate the occurrence probability P(m) of each scenario output by S3041. n The corresponding absolute time integral experience value obtained from S3042 The system performs orthogonal product summation; based on this, the scheduling system precisely calculates the risk-adjusted mathematical expectation value, and strictly defines this expectation value as the final service continuity retention S. cont Its underlying aggregation formula is:

[0146]

[0147] Through this step, the scheduling system forcibly reduces the complexity of multidimensional nonlinear fault deduction results and aggregates them into a unique, objective, and comprehensive decision value that is not excessively affected by a single extreme scenario.

[0148] S3044, Decision Execution and Network Cutover Module, configured to compare service continuity retention with a preset security baseline threshold. When the service continuity retention is greater than or equal to the security baseline threshold, a physical flow table switching command is issued to the underlying network controller to perform a bidirectional physical cutover of network status and traffic. When the service continuity retention is less than the security baseline threshold, the underlying circuit breaker mechanism is triggered and the migration control scheme is destroyed.

[0149] The rigid threshold comparison and physical flow cut / circuit break command scheduling system extracts the internally preset and inviolable Service Level Agreement (SLA) security baseline threshold S. baseline The scheduling system will aggregate and calculate the service continuity retention S. cont With the safety baseline threshold S baseline Perform real-time digital comparison:

[0150] When the comparison result is S cont ≥S baseline When the scheduling system determines that the current migration control scheme has sufficient resilience against disturbances, it immediately releases the digital twin isolation lock and issues the real first handover command to the underlying software-defined network (SDN) controller to perform a two-way physical cutover of status and traffic.

[0151] The first switching command includes sending address rewrite and port redirection messages, sending in-flight tunnel redirection messages, and sending old route aging and erasure messages.

[0152] The address rewriting and port redirection message includes: forcing the edge switch to intercept the uplink service data packets sent by the communication terminal, rewriting the destination MAC address and destination IP address in the protocol header from the network address of the original source defense node to the network address of the target defense node in real time, and specifying the physical forwarding port to be changed to the port of the shortest topology path pointing to the target defense node;

[0153] The message sent to redirect in-flight data packets through a tunnel includes: within the same clock cycle as the aforementioned redirection message, a second flow table modification message is sent. The scheduling system uses this message to activate a pre-established underlying virtual extended LAN tunnel between the source defense node and the target defense node, forcibly redirecting "in-flight" TCP sequence data packets that have arrived at the source defense node but have not yet been processed by the service process to the target defense node's receive buffer intact via this tunnel.

[0154] The issuance of old route aging and erasure messages includes: immediately sending a third-party flow table deletion message to the switching devices on the original topology path after detecting that the target defense node has successfully returned the first TCP acknowledgment packet (ACK) to the communication terminal. This deletion message forcibly erases all old state forwarding rules pointing to the source defense node, thereby completing a hard switchover of the physical plane of network traffic within a millisecond time window.

[0155] When the comparison result is S cont baseline If the scheduling system determines that the current solution has too large a risk exposure under the edge network base, it will directly trigger the underlying circuit breaker mechanism, forcibly destroy the current migration control solution and clear the digital twin sandbox memory. At the same time, it will order the target stateful service to remain on the source defense node for silent cooling until the scheduling system detects that the communication terminal has entered the next physical area with highly stable topological characteristics before the evaluation can be restarted.

[0156] ​In this embodiment, before performing physical network cutover, a digital twin sandbox is constructed in memory through a simulation and evaluation module, and a service gravity field transfer model is mapped to simulate the dynamic transfer process from state pre-synchronization to complete traffic switching. During the simulation, the service fidelity integrator uses the effective interaction frame rate and the proportion of responses without timeout as positive gains, and the network shadow traffic bandwidth squeezing and the opportunity cost of target node resource reservation as negative penalties, to calculate the instantaneous user experience fidelity function in real time. At the same time, the chaos scenario injection engine injects a composite fault operator into the protocol stack and input / output controller at the bottom layer of the sandbox, which includes low-level packet sending rate suppression, forced suspension of data block writing timestamps, and high-frequency injection of forged location coordinates, thereby constructing a low-level test environment for micro-disruption of communication links, internal bus blocking of nodes, and trajectory logic drift. The environmental context awareness unit extracts real-time features of the physical network to generate a set of probability distributions for various complex fault scenarios. Combined with the absolute time integral experience value obtained by performing definite integral accumulation based on the Riemann summation rule, the unit performs mathematical orthogonal product summation through the expectation aggregation unit to precisely calculate the dimensionless service continuity retention. The scheduling system compares this service continuity retention with a preset security baseline threshold. If the value meets the threshold, it issues a linked physical flow table switching command to the underlying network controller, including address rewriting, in-flight packet tunneling, and old route state erasure, to perform bidirectional network traffic cutover. If the value does not meet the threshold, it directly triggers the underlying circuit breaker mechanism and destroys the current migration plan. The beneficial effects of the above technical solutions are as follows: by constructing a digital twin space and a composite fault injection mechanism, the system can complete the boundary pressure simulation of multi-dimensional operational faults without consuming real physical network bandwidth; by using definite integral operations and a probability-weighted mathematical expectation model, the instantaneous experience loss of multi-source nonlinearity is reduced to an objective and single quantitative evaluation index, avoiding the risk of connection interruption caused by blindly performing hard state switching in complex physical topologies; and by combining the underlying fine-grained flow table modification instructions and the flight data diversion tunnel mechanism, the system can ensure data consistency and smooth transition of network sessions during the handover process of stateful services between wide-area edge nodes.

[0157] Figure 1This diagram visually illustrates a widely distributed computing node scenario, showing the physical topology of a communication terminal traversing between source defense nodes and multiple candidate target defense nodes, and providing insight into the underlying computing resource status of edge nodes. The technical roadmap below the diagram precisely corresponds to the four core workflow mechanisms of the system in this application: the first step corresponds to the "Topology and State Awareness Module," illustrating that the system first detects node link characteristics and microarchitectural status, quantifying underlying computing resource interference to generate a dynamic candidate migration path set; the second step corresponds to the "Predictive Migration Decision Module," indicating that the system combines the terminal's predicted trajectory to calculate the spatial discrete variance and generate a hysteresis factor, thereby evaluating the net migration benefit index and locking in the optimal target defense node; the third step corresponds to the "Simulation and Evaluation Module," indicating that before physical cutover, the system performs simulation and evaluation in memory... A digital twin sandbox is constructed and a service gravity field transfer model is mapped. By injecting a composite fault operator into the underlying layer and performing definite integral accumulation, the service continuity retention degree after risk adjustment is precisely calculated. The fourth step corresponds to the "Decision Execution and Network Cutover Module", which indicates that the system will finally compare the calculated retention degree with the security baseline threshold. When the threshold is met, the underlying controller will be linked to issue a physical flow table switching command to perform a bidirectional physical cutover of network status and traffic. In this way, the boundary pressure simulation can be completed without consuming physical bandwidth, and the smooth transition and continuous operation of stateful services between nodes can be effectively guaranteed.

[0158] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, min-max-normalization and Z-score standardization.

[0159] The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (e.g., using the widely used pandas library in Python) configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (e.g., a DataFrame data structure). Subsequent algorithm steps will directly query and retrieve the required configuration parameters from this in-memory data structure.

[0160] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A node task migration and scheduling system based on digital twins, applied to widely distributed computing nodes, for performing stateful services on computing resources and network latency for high-speed mobile communication terminals, characterized in that, include: The topology and state awareness module configures source defense nodes serving mobile terminals and constructs an opportunity migration domain map containing multiple candidate target defense nodes; it detects the link characteristics and micro-architecture state from the source defense nodes to each candidate target defense node to calculate the link resilience coefficient and computing power affinity index, and generates a dynamic candidate migration path set. The predictive migration decision module is configured based on a dynamic candidate migration path set, combining the predicted trajectory of the mobile terminal with the underlying operational characteristics of the candidate target defense nodes, to evaluate and generate a net migration benefit index for each candidate migration path. The node with the highest net migration benefit index is selected as the optimal target defense node, and a migration control scheme including a state-progressive synchronization strategy and a traffic switching window is generated. The simulation and evaluation module is configured to build a digital twin sandbox in the system memory according to the migration control scheme, and simulate the gravitational potential energy transfer process from state pre-synchronization to complete business traffic switching in the digital twin sandbox; Complex fault scenarios are extracted from a pre-defined disturbance library and injected into a digital twin sandbox. The instantaneous user experience fidelity function is calculated through definite integral accumulation, and the service continuity retention is precisely calculated by combining the probability aggregation of multiple scenarios. The simulation and evaluation module includes a chaotic scenario injection engine; the chaotic scenario injection engine is configured to inject complex fault scenarios containing the following types into the underlying protocol stack and block device drivers of the digital twin sandbox: First Phantom Bandwidth Scenario: Forcefully suppress the packet sending rate of the underlying network port to be lower than the preset percentage of the average network bandwidth value, and recover after a set oscillation period to generate a micro-disruption environment in the communication link. Second dead node scenario: Attach input / output interception hooks to forcibly suspend the timestamp of writing the data block status response until the timeout threshold is exceeded, generating an internal storage bus blocking environment; The third pendulum trap scenario: virtual coordinate sequences are injected into the position input port, forcing the digital twin terminal to cross adjacent physical boundaries at high frequency within a continuous observation time window, generating a trajectory logic drift environment; The decision execution and network cutover module is configured to compare the service continuity retention with a preset security baseline threshold. When the service continuity retention is greater than or equal to the security baseline threshold, it sends a physical flow table switching command to the underlying network controller to perform a bidirectional physical cutover of network status and traffic. When the service continuity is less than the safety baseline threshold, the underlying circuit breaker mechanism is triggered and the migration control scheme is destroyed. The decision execution and network cutover module issues physical flow table switching instructions to the underlying network controller and performs bidirectional physical cutover of network status and traffic. Specifically, this includes issuing the following linked messages: Issuing address rewrite and port redirection messages: generating a first flow table modification message, forcing the edge switching device to intercept uplink service data packets, and real-time overwriting the destination medium access control address and destination network protocol address in their headers to the network address of the optimal target defense node, and changing the physical forwarding port; Issuing fly-through packet diversion messages: generating a second flow table modification message within the same issuance cycle, activating the underlying virtual extended LAN tunnel between the source defense node and the optimal target defense node, and diverting the transmission control protocol sequence data packets that have reached the source node but have not been processed by the service process to the receive buffer of the optimal target defense node; Issuing old route erasure messages: after detecting that the optimal target defense node returns the first acknowledgment control message, issuing a flow table deletion message to the original topology path switching device, forcibly erasing all state forwarding rules pointing to the original source defense node from physical memory.

2. The node task migration and scheduling system based on digital twins according to claim 1, characterized in that: The topology and state awareness module includes a link quality assessment unit; the link quality assessment unit is configured to analyze in real time the micro-burst packet loss pattern, the long-tail effect of latency distribution, and the congestion feedback delay of the protocol stack in the link between the source defense node and the candidate target defense node. The variance of micro-burst packet loss rate, the ratio of the long tail of the 99th percentile network delay to the average network delay, and the absolute hysteresis time of the transmission control protocol stack congestion feedback signal are extracted from the links within a preset time window. These are then assigned preset regularization penalty weights, and a nonlinear exponential decay model is used to output the link resilience coefficient, which characterizes the link's ability to withstand synchronous traffic collapse under the link's carrying state. The increase in the variance of micro-burst packet loss rate and the ratio of the long tail leads to an exponential decay of the link resilience coefficient.

3. The node task migration and scheduling system based on digital twins according to claim 2, characterized in that: The topology and state awareness module also includes a computing power compatibility measurement unit and a path index generation unit. The computing power compatibility measurement unit is configured to extract the multi-core consistency protocol of the processor cache, the memory page replacement strategy across nodes with non-uniform memory access architecture, and the application programming interface compatibility of the instruction set of the secure execution environment, and output the computing power affinity index through a Boolean matching measurement function. The path index generation unit is configured to perform a geometric mean operation on the link resilience coefficient calculated by the link quality assessment unit and the computing power affinity index calculated by the computing power compatibility measurement unit to obtain the migration path fidelity index; remove nodes whose index is not higher than the dynamic admission threshold, and merge the remaining nodes into the dynamic candidate migration path set.

4. A node task migration and scheduling system based on digital twins according to claim 1, characterized in that: The predictive migration decision module includes a path screening unit and a node service quality analysis engine. The path screening unit is configured to remove candidate paths from the opportunity migration domain map that do not meet the threshold for migration path fidelity. The node service quality analysis engine is configured to read the memory bus contention intensity, input / output scheduling queue starvation probability, and CPU pipeline branch prediction interference rate of the target defense node through underlying hardware probes, and obtain the cross-tenant noise coefficient through dimensionless weighted calculation. It also calculates the dirty page mutation rate of the source defense node's memory to generate the state synchronization complexity entropy, and integrates the service recovery time gradient of the target node to generate a node service quality vector containing multi-dimensional features.

5. A node task migration and scheduling system based on digital twins according to claim 4, characterized in that: Before constructing the node service quality vector and executing the decision, the predictive migration decision module is also configured to calculate the adaptive migration decision hysteresis factor: extract the statistical variance of the instantaneous speed scalar time series and heading angle time series of the mobile terminal within a continuous time window, and generate spatial discrete variance by weighting the set speed and heading weight coefficients. The difference between the spatial discrete variance and the preset critical trajectory disorder variance threshold is substituted into the inverse S-shaped activation function configured with the maximum damping coefficient and the steepness response coefficient to generate a migration decision hysteresis factor, which is used to apply adaptive decision mathematical damping to the high-frequency topology changes caused by mobile terminal trajectory fluctuations.

6. A node task migration and scheduling system based on digital twins according to claim 5, characterized in that: The predictive migration decision module also includes a migration benefit calculation unit and an optimal decision unit; The migration benefit calculation unit is configured to perform mathematical time integration on the expected communication latency reduction of migration to the target defense node within the future time window and the dynamic business request sensitivity distribution function to obtain the expected bonus integral value; and to perform multiplication on the single benchmark calculation loss time constant, state synchronization complexity entropy, cross-tenant noise coefficient and migration decision hysteresis factor to obtain the instantaneous service interruption penalty time. The optimal decision unit configuration is used to calculate the difference score between the expected bonus score and the instantaneous service interruption penalty time, and then substitute this difference score and the migration path fidelity index into the harmonic mean formula with bias weights for fusion, and output the net migration benefit index.

7. A node task migration and scheduling system based on digital twins according to claim 1, characterized in that: The predictive migration decision module also includes a risk prediction unit, a migration triggering unit, and a takeover capability judgment unit. The risk prediction unit is configured to predict the failure rate risk of the source defense node falling below the service level agreement tolerance latency by combining the physical departure trajectory of the mobile terminal. The migration triggering unit is configured to transmit a baseline memory snapshot to the optimal target defense node and establish an iterative tracking matrix to perform incremental state synchronization when the failure rate risk exceeds the dynamic risk tolerance threshold. The takeover capability judgment unit is configured to calculate the convergence deviation of the state replica corresponding to the residual dirty data volume in real time, calculate the slope of the convergence learning curve for continuous iteration, and when the slope of the convergence learning curve is negative and the absolute value exceeds the preset rate threshold, and it is determined that the remaining residual dirty data can be atomically transmitted within the critical minimum value of the network request interval, the hot takeover capability is confirmed to meet the standard and the traffic switching window is locked.

8. A node task migration and scheduling system based on digital twins according to claim 1, characterized in that: The simulation and evaluation module also includes an environmental context awareness and probability prediction unit, a service fidelity integrator, and an expectation aggregation unit. The environmental context awareness and probability prediction unit is configured to extract the historical routing jitter rate and real-time computing load fluctuation frequency of the physical network, and generate a normalized set of probability distributions of composite fault scenarios through the output layer of the value network. The service fidelity integrator is configured to use the effective interaction frame rate achievement index as a positive gain and the network timeout response attenuation, state synchronization shadow traffic bandwidth occupancy rate, and target node reserved resource opportunity cost as negative penalties. It performs time integration calculations using the Riemann summation rule to obtain the absolute time integral experience value for each fault scenario. The expectation aggregation unit is configured to perform mathematical orthogonal product summation on the probability distribution set and the absolute time integral experience value to output the service continuity retention in a reduced dimension.

Images