Cloud edge-end collaborative computing power scheduling system supporting dynamic task migration

Abstract

The invention discloses a cloud edge end collaborative computing power scheduling system supporting task dynamic migration, which relates to the technical field of intelligent computing power scheduling and comprises a path modeling module, a conflict identification module, a time sequence checking module, a rehearsal switching module and an interference suppression module, and constructing a routing convergence time sequence profile and a link playback spectrum, modeling an overlapping window of the old path and the new path and a life cycle of a control signal, and generating a path evolution baseline for subsequent conflict identification. By means of path modeling, fingerprint identification, causal check, shadow rehearsal, interference suppression and the like, time sequence consistency of a control instruction and a data stream, continuous path switching and real-time extinguishing of link interference are realized, and the stability and reliability of cloud edge collaborative scheduling are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent computing power scheduling technology, and more specifically to a cloud-edge-device collaborative computing power scheduling system that supports dynamic task migration. Background Technology

[0002] The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration is an adaptive resource collaboration system for multi-layered computing environments. It achieves real-time optimal scheduling of tasks across the cloud, edge computing, and devices by constructing intelligent task offloading, dynamic computing power allocation, and migration management mechanisms. The system first establishes a multi-dimensional evaluation index system, including edge-side energy consumption, execution time, cloud transmission overhead, and latency, forming a quantitative basis for task execution location decisions. Through a hybrid decision model integrating the Analytic Hierarchy Process (AHP) and Deep Neural Networks (DNN), index weights and real-time perceived data are input into the neural network to automatically learn the optimal execution strategy for different task types (such as computationally intensive and real-time response tasks). When the cloud execution probability output by the model exceeds a preset threshold, the task is intelligently offloaded to the cloud for execution; otherwise, it remains on the edge computing end for processing, thus achieving a dynamic balance between energy consumption, latency, and computational efficiency, forming the system's intelligent decision engine layer.

[0003] During task execution, the system utilizes a dynamic migration mechanism to adaptively flow computing power between the cloud, edge computing, and devices. The system continuously monitors edge CPU / GPU load, memory usage, and cloud task queue length and idle computing power. When edge load consistently exceeds a threshold and task time remains long, task migration to the cloud is triggered. When cloud resources are idle and transmission time is acceptable, some tasks can be migrated back to the edge computing. The migration process employs data snapshot technology to save intermediate task states, ensuring seamless execution after migration, and reduces transmission latency through a bandwidth adaptive adjustment mechanism. Ultimately, the system forms a closed-loop architecture consisting of a decision engine layer, a dynamic migration control layer, and a collaborative optimization layer. This enables real-time resource awareness, self-scheduling of computing power, and self-repair during migration between the cloud and edge computing, comprehensively improving the efficiency, reliability, and adaptability of integrated cloud-edge-device collaborative computing.

[0004] The existing technology has the following shortcomings:

[0005] In existing technologies, when cloud-edge collaborative systems perform large-scale computing power switching during peak task migration periods, the transmission links between the cloud and edge nodes are often in a state of dynamic reconstruction with frequent connection and disconnection. Due to the convergence delay in routing protocols during frequent backtracking and reconnection, the network topology table cannot be updated in a timely manner, resulting in some old paths still maintaining connections while new paths have been activated, forming link overlap areas. Within this overlap area, residual control packets and newly generated control commands may propagate simultaneously, causing overlapping control signals and interference. This leads to the repeated execution or premature triggering of task scheduling commands, resulting in misaligned execution of task migration commands and disordered data packet order. Such latent interference is particularly prominent in high-concurrency migration scenarios, causing minor issues such as single-node communication anomalies, and severe issues such as paralysis of the cloud-edge control channel, global migration command failure, and computing power scheduling link interruption, thereby significantly reducing the system's reliability and continuous operation capability.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a cloud-edge-device collaborative computing power scheduling system that supports dynamic task migration, in order to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration, comprising a path modeling module, a conflict identification module, a timing verification module, a pre-simulation switching module, and an interference suppression module:

[0009] The path modeling module, based on the dynamic topology of cloud-edge transmission links, constructs a route convergence time profile and link playback spectrum, models the overlap window of old and new paths and the life cycle of control signals, and generates a path evolution baseline for subsequent conflict identification.

[0010] The conflict identification module generates a fingerprint feature set of control signals based on the path evolution baseline. By analyzing the timing deviation and transmission trajectory of the control signals, it constructs a control command contradiction relationship diagram to identify control command misalignment sources and false triggering intervals.

[0011] The timing verification module establishes a causal verification chain across network protocol layers based on the control command contradiction relationship diagram. By comparing the consistency and monotonicity of the timestamps of control signals and data packets, it completes the time redirection of scheduling commands and data packets.

[0012] The pre-switching module constructs a reversible queue fence and shadow channel switching sequence based on the verification results of the causal check chain. Before the main path switching, it performs a consistency pre-switching through the shadow path and performs the main path switching operation after verifying the stability of the control signal.

[0013] The interference suppression module, based on the pre-simulation results of the shadow channel switching sequence and the buffer isolation mechanism provided by the reversible queue fence, performs phase conjugate timing back-turn suppression operation and polarization channel isolation linkage operation. It completes path drift and control signal deduplication processing within the peak period of network congestion, constructs a dynamic control closed loop, and realizes real-time suppression of link interference and stability assurance of task migration.

[0014] Preferably, the path evolution baseline generation steps are as follows:

[0015] Based on the physical communication structure and connection node relationship between the cloud and the edge, the connection status changes of edge nodes, scheduling strategy execution information, task migration request response time and cloud resource feedback information are collected in continuous time periods. The connection status and key transmission parameters of each path at each moment are recorded to construct a routing convergence time sequence profile.

[0016] The connection state is associated with the task scheduling behavior. Based on the number of hops, transmission delay and access node information of each hop path, the complete transmission path of the control command from the sending node to the receiving node is reconstructed, the link playback spectrum is constructed and the overlapping areas of the path are marked.

[0017] Based on the propagation trajectory of control commands in the link playback spectrum, the start and end times of the propagation of control signals between the old and new paths, the path intersection time window, and the command execution feedback time are marked, and a life cycle description model of control signals is established.

[0018] Combining the aforementioned time-series profile, link playback spectrum, and lifecycle model, a unified mapping is applied to the time-labeled sequence to label the active path set, control signal set, and path overlap relationship at each moment, thus constructing a path evolution baseline.

[0019] Preferably, when constructing the path evolution baseline, the active path set, control signal set, and path overlap relationship corresponding to each time label are jointly labeled with network congestion level, bandwidth fluctuation frequency, and link switching response delay to enhance the adaptability of the path evolution baseline to dynamic changes in links.

[0020] Preferably, the steps for constructing the control instruction contradiction relationship diagram are as follows:

[0021] Based on the propagation trajectory of control signals in the path evolution baseline, the generation time, forwarding time and arrival time of control commands in each path segment are collected, and a tag set containing timestamps, path numbers, node numbers, signal numbers and target behaviors is constructed.

[0022] Based on the control signal tag set, its propagation timing, path overlap and feedback behavior are extracted to generate a unique fingerprint structure. The signal feature vector is then normalized by combining historical task behavior to construct a fingerprint feature set.

[0023] Based on fingerprint feature sets, control signal pairs with consistent targets and propagation conflicts are selected, a control command conflict relationship graph is constructed, and the behavioral conflict paths and propagation time windows are marked.

[0024] Based on the information of conflict edges in the contradiction relationship graph, the propagation path of the control signal is traced back, and the switching time and link behavior in the path evolution baseline are combined to identify the source of instruction misalignment and the false triggering interval.

[0025] Preferably, each conflict edge in the control command conflict relationship graph is accompanied by a propagation time window, path intersection segment identifier, and conflict level index to help identify high-incidence segments of propagation timing disorder and command misalignment source locations.

[0026] Preferably, a causal verification chain across network protocol layers is established based on the control command contradiction relationship diagram. Consistency comparison and monotonicity checks are performed on the timestamps of control signals and data packets. The time redirection steps are as follows:

[0027] Based on the control command conflict relationship diagram, conflicting control command pairs are extracted and associated with the transmission time behavior of their corresponding data packets to establish a time mapping pair between control signals and data packets.

[0028] Based on the time mapping pair, the consistency comparison between the control signal trigger time and the data packet reception time is performed to check whether there is a deviation in the order of the two at the logic layer and the transport layer, and to construct a preliminary causal conflict chain.

[0029] Based on consistency deviation information, a causal verification chain covering the network transmission layer, scheduling triggering layer and task execution layer is generated, and the task state switching points and control branch relationships are marked.

[0030] By referencing the causal verification chain and path evolution baseline, time redirection is performed on control signals and data messages with sequence deviations to adjust their execution timing and achieve consistent restoration of task scheduling behavior.

[0031] Preferably, time redirection includes advancing the activation time of the control signal to the synchronization time point before the arrival of the task data packet, and delaying the processing operation of the task data packet that is triggered earlier than the control signal until it is confirmed by the control signal.

[0032] Preferably, based on the causal verification chain results, a reversible queue fence and shadow channel switching sequence is constructed. After performing consistency pre-rehearsal through the shadow path, the main path switching operation steps are as follows:

[0033] Based on the time redirection results of the causal verification chain, a buffer structure with state preservation capability is constructed to temporarily encapsulate scheduling instructions and data packets and configure rollback flags;

[0034] Based on the control instructions in the buffer structure, a physically isolated path consistent with the main path structure is selected as the shadow path, and the instruction stream is injected in sequence, and the propagation trajectory and response behavior of each node are recorded.

[0035] By evaluating the timing consistency of the control signal propagation behavior in the shadow path, it is determined whether it meets the stable propagation conditions, and intermediate state consistency is compared.

[0036] Release the control commands in the buffer structure to the main path, perform a path switching operation and load the response template in the shadow path until the control signal completes the confirmation feedback and updates the task scheduling table.

[0037] Preferably, before the control signal is released to the main path, the relay behavior and command response results in the shadow path are templated, and the corresponding template is loaded in the main path node to match the execution state, so as to ensure the consistency and continuity of the control signal response during path switching.

[0038] Preferably, based on the pre-simulation results of the shadow channel switching sequence and the buffer isolation mechanism of the reversible queue fence, the following steps are performed to conduct path drift and control signal deduplication during the peak period of network congestion:

[0039] Based on the shadow channel switching sequence, a pre-simulation trajectory and buffer isolation mechanism are used to perform phase conjugate timing back-turn suppression operation, reverse mapping of dense propagation instructions and reset release time;

[0040] Based on the location of vulnerable nodes in the path topology, polarization channel isolation linkage operation is implemented to group and isolate the propagation channels and set signal labels to avoid crosstalk.

[0041] During network congestion periods, micro-periodic path drift operations are performed using the stability assessment results of shadow paths, and hop-by-hop task transfers are completed by calling the control signal states recorded in the cache structure.

[0042] After the path drift is completed, the control signal deduplication process is performed based on the fingerprint characteristics of the control signal, and the valid instruction status is written back through the reversible queue fence to build a dynamic control closed loop.

[0043] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0044] This invention constructs a refined path evolution baseline and control signal feature set through path modeling and fingerprinting analysis, significantly improving the ability to detect potential link conflicts and abnormal control signal behavior. It restores logical consistency between control commands and data flows through causal verification mechanisms and time redirection strategies. Combined with shadow path pre-simulation and rollback mechanisms, it ensures that command execution during main path switching has verification support and fault tolerance capabilities. Finally, it establishes a dynamic real-time extinguishing mechanism for link interference using propagation timing compression and channel physical isolation technologies. In summary, this invention can achieve accuracy in control signal scheduling, continuity in path switching, and stability in computing power migration in dynamic network environments, thereby comprehensively improving the reliability and robustness of the cloud-edge-device integrated scheduling system. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0046] Figure 1 This is a schematic diagram of the cloud-edge-device collaborative computing power scheduling system that supports dynamic task migration according to the present invention. Detailed Implementation

[0047] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0048] This invention provides, for example Figure 1 The cloud-edge-device collaborative computing power scheduling system shown includes a path modeling module, a conflict identification module, a timing verification module, a pre-simulation switching module, and an interference suppression module.

[0049] The path modeling module, based on the dynamic topology of cloud-edge transmission links, constructs a route convergence time profile and link playback spectrum, models the overlap window of old and new paths and the life cycle of control signals, and generates a path evolution baseline for subsequent conflict identification.

[0050] In cloud-edge collaborative computing environments with high concurrency during task migration, to effectively address control signal conflicts caused by frequent link reconfiguration, a path evolution modeling process needs to be constructed based on the dynamic topology characteristics of the transmission path for conflict source identification. This process aims to obtain the temporal characteristics of path changes and control signal behavior during task scheduling. The specific steps are as follows:

[0051] Based on the actual physical communication structure and connection node relationships deployed between the cloud and the edge, basic data such as the connection status changes of edge nodes, scheduling policy issuance records, task migration request response times, and cloud resource scheduling feedback are collected within a specific time period. During the data collection process, it is essential to ensure the entire connection trajectory, including the edge-side network access link, the forwarding path from the edge to the intermediate layer gateway, and the final destination to the cloud master node, is captured. The status of each path at each sampling time point must be clearly defined as one of four states: connected, disconnected, undergoing reconstruction, or switching. Key performance indicators such as connection time, hop count, and bandwidth stability between nodes are also recorded. Based on this, all link states are sorted according to the timeline, and a two-dimensional time series graph is constructed with time as the horizontal axis and path as the vertical axis to establish the connection relationship between edge nodes and cloud nodes, thus constructing a preliminary routing convergence time series profile. This profile describes the continuity, reconstruction frequency, and stability changes of the end-to-end path from the start to the completion of task scheduling, providing a data foundation for subsequent refined modeling.

[0052] After obtaining the route convergence time profile, the aforementioned connection state information needs to be correlated and extended with task scheduling behavior to further trace the propagation trajectory of control signals during path changes and construct a link replay spectrum. This step relies on the path hop count, transmission delay, and access node identification results calibrated in the previous stage to reconstruct the complete physical link from the issuing node to the receiving node for each scheduling control command via each hop path. During replay, an actual transmission time window is allocated for the control command distribution for each hop, and the dwell time, repeated forwarding behavior, and arrival order of the control signal on different links are recorded and compared in conjunction with the actual existence time of the path. By comparing the behavioral differences of different control commands near path switching points during the same task scheduling process, path overlap areas where signals may overlap are identified, and a transmission event chain is established to describe all possible path behavior branches and intersections during the control packet generation and confirmation execution process.

[0053] After constructing the link replay spectrum, the propagation trajectory of control signals during the replay process should be further utilized to refine the modeling of the overlap window between the old and new paths, clearly marking the lifecycle of each control signal. At this stage, the generation time, path in point, path out point, arrival time, and execution feedback time of the control signals need to be jointly mapped onto the constructed time series graph. Specifically, for each control signal, its propagation start and end times in the old path need to be determined, and it needs to be marked whether it was sent before the new path was activated. If not, its possible dwell time should be calculated, and the possible new path forwarding segment it might enter should be identified. Similarly, the propagation start point of control commands generated in the initial stage of new path activation needs to be compared with the time overlap of commands still in effect in the old path, extracting the overlap time window and marking the interval where signal interference may occur. Based on identifying the overlap intervals of control signals, a unique identifier should be established for the propagation behavior of each control packet, marking its effective duration, expected execution node, and execution command number, ensuring that the scope of each control behavior during the path evolution process is clearly identifiable. Through this stage of modeling, we can accurately characterize the mutual interference potential and timing crossover behavior of control signals during the path reconstruction process, forming a complete description of the correlation between the life cycle of control signals and link reconstruction.

[0054] Based on the time-series profiles, path playback spectra, and control signal propagation lifecycle data obtained in the three stages described above, a path evolution baseline is synthesized. This path evolution baseline serves as a continuous time-series reference to support subsequent conflict identification, control command misalignment verification, and causal relationship tracing. Specifically, in its construction, path switching events, control signal generation and propagation behavior, and the state values ​​of each link at the same moment in each task scheduling cycle are uniformly mapped to a set of multi-dimensional time-labeled sequences. At each time-label point, the active path set, effective control signal set, and potential overlapping relationship group are labeled. Simultaneously, to enhance the dynamic adaptability of this baseline, transmission environment factor variables need to be incorporated, such as recording network congestion levels, bandwidth fluctuation frequency, and link switching response delays at each time point, to assist in the subsequent dynamic prediction of control signal failure or misorder risks. Based on this, the path evolution baseline, as a unified bridge between scheduling behavior and dynamic changes in links, provides quantitative support and behavioral boundary constraints for subsequent control command conflict analysis, scheduling command redirection, and the establishment of migration stability assurance strategies.

[0055] The conflict identification module generates a fingerprint feature set of control signals based on the path evolution baseline. By analyzing the timing deviation and transmission trajectory of the control signals, it constructs a control command contradiction relationship diagram to identify control command misalignment sources and false triggering intervals.

[0056] Building upon the established path evolution baseline, to accurately identify misaligned execution and false triggering behaviors during control command propagation, key propagation features of control signals are extracted. A time-series behavior map is constructed to reveal their propagation conflict relationships and mutual influence paths during link evolution. The specific steps are as follows:

[0057] Based on the control signal propagation lifecycle provided by the path evolution baseline, refined identification labels for control signals in different propagation paths should be established. Specifically, for each identified control signal path segment, the control command generation time at the link ingress point, the relay forwarding time at intermediate hops, and the command landing time at the target execution node should be collected. The node identifier, transmission direction, and duration of each hop in its propagation trajectory should also be fully recorded. Based on this, the behavior information of the control signal in the path evolution baseline and its path propagation status should be jointly encoded to generate a raw label with a five-tuple containing a timestamp, path number, node number, unique signal number, and target behavior identifier. Subsequently, feature analysis should be performed on all raw control signal labels to extract information such as propagation mode, start and end status, hop count distribution, time distribution, and response feedback status, constructing a parameter space for signal behavior classification. Combining the propagation patterns of each type of task scheduling command in historical migration events, the propagation behavior of control signals in the current task should be aligned and compared to identify signal samples that deviate significantly from normal behavior trajectories and mark them as candidates for abnormal behavior.

[0058] After obtaining the candidate set of abnormal behaviors for control signals, a unique identifier for the control signals needs to be constructed based on the aforementioned propagation quintuple and behavioral deviation information, forming a feature vector set for conflict identification. In specific implementation, based on the signal's generation time within the same scheduling cycle, the degree of overlap in the evolutionary baseline before and after path switching, the dwell time on each path segment, and the actual task response delay, the time offset, path difference, and instruction target repetition of the signal with its neighboring signals are calculated, and this set of data is mapped to fingerprint features. When generating feature vectors, the features of confirmed normal signals in historical task migration trajectories should be introduced as a reference baseline to normalize the feature vectors of the current signals, enhancing their stability in anomaly clustering discrimination. Finally, a unique fingerprint structure is formed for each control signal in the current link evolution context. The fingerprint includes signal temporal behavior encoding, path overlap encoding, node propagation trajectory encoding, and feedback behavior expectation difference encoding, used to support subsequent attribution analysis of misalignment sources and conflict paths.

[0059] After establishing the fingerprint feature set, a conflict mapping structure diagram should be constructed based on the behavioral contradictions between control signals during the control command propagation process. In the specific construction process, firstly, within the fingerprint feature space, control signal pairs that overlap in time and have the same target task in the path evolution baseline are selected, and their propagation direction, execution node, and arrival time sequence are assessed for any deviations. For signal pairs that meet the criteria, their control content is further determined. If the command content is different but the target is the same, a behavioral contradiction relationship connection line is established and marked as a potential false triggering relationship. If the command content is the same but the propagation time sequence is inconsistent and occurs near a link switching node, a misalignment relationship connection line is established and marked as a potential repeated execution relationship. This process continues, constructing a control signal conflict map within the fingerprint set, where nodes represent specific control signals and edges represent the propagation behavioral conflict paths between them. To improve the temporal readability and behavioral expressiveness of the conflict identification map, each edge should be appended with a propagation time window, path intersection segment identifier, and conflict level evaluation index to assist in identifying the dominant conflict path in multi-signal intersection segments.

[0060] Based on the constructed control command conflict relationship graph, conflict paths are analyzed to identify control command misalignment sources and false triggering intervals. During implementation, starting from the edge with the highest conflict level index in the graph, the propagation origin and destination of the control signals between its two connected nodes are traced back. Combined with the path switching time and bandwidth changes in the path evolution baseline, the propagation stages that may trigger abnormal control signal behavior are located. All conflict edges connecting the same target nodes are clustered, and time periods with concentrated propagation timing disorder or overlapping feedback responses are extracted and defined as high-incidence intervals for false triggering. Further analysis of the signal propagation path starting node and path change rate within these intervals identifies the node where the earliest control signal mutation or path behavior reconstruction occurs as a candidate for command misalignment source. Finally, the identification of misalignment triggering origins, interference behavior links, and abnormal propagation segments in abnormal control signal propagation paths is completed, providing a structured input basis for subsequent causal verification and path switching safety strategies.

[0061] The timing verification module establishes a causal verification chain across network protocol layers based on the control command contradiction relationship diagram. By comparing the consistency and monotonicity of the timestamps of control signals and data packets, it completes the time redirection of scheduling commands and data packets.

[0062] After identifying potential misalignment sources and false triggering intervals during control signal propagation, to ensure the consistency of scheduling instructions' execution and the timing correctness of task processing results across different path evolution stages, a strict causal chain is established based on contradictory signal behaviors. Time redirection processing is then applied to control instructions and data packets to resolve time misalignment issues between the logic and transport layers caused by path jitter and abnormal propagation order. The specific steps are as follows:

[0063] Based on the constructed control command conflict graph, all conflicting control command pairs are extracted, and their temporal behavior is aligned with that of their corresponding data packets throughout the entire propagation path. Each control command has a unique fingerprint generated during its construction, including its generation time, path entry time, transmission time at each hop relay, target node arrival time, and command confirmation feedback time. Similarly, data packets also possess independent transmission trajectories and multiple delay records. At this stage, the lifecycle of the control command is used as a baseline. Starting from its initial generation time, the corresponding task data packets are associated, and their generation time, packet completion time, sending node path entry time, intermediate node transmission time, and destination node reception time are extracted. During the alignment process, it is necessary to determine whether control signals and data packets share path nodes in the physical transport layer. Within the shared path interval, their entry and exit behaviors within the same time period are compared to identify any time misalignments, propagation direction deviations, or overlapping congestion periods, thereby establishing a time mapping pair between data packets and control signals.

[0064] After completing the timing mapping between control signals and data packets, a consistency comparison operation needs to be performed on the mapping pairs to check whether there is a significant deviation between the logical layer triggering order and the physical layer transmission order. Specifically, the execution trigger time of the control signal should be used as the baseline time point to determine whether the reception time of the corresponding data packet strictly follows the control command trigger, with a reasonable threshold range set with a certain time tolerance. If a data packet arrives at the target node before the control signal, or if a data packet enters the task processing flow before the control signal has been acknowledged, it should be marked as a timing inconsistency event. Simultaneously, combining the link bandwidth changes and node processing capacity status in the path evolution baseline, it is assessed whether there is abnormal acceleration or lag in the arrival time of data packets during path congestion, further determining whether it is asynchronous behavior generated during path switching. In this comparison phase, it is necessary not only to determine the logical relationship between a single control command and its corresponding data packet, but also to perform cross-command propagation order verification between multiple instruction sequences of the same task to confirm whether there is time reversal in the control sequence or a long-term delay of a certain instruction in the control cluster, in order to form the preliminary skeleton structure of a causal conflict chain.

[0065] After identifying consistency discrepancies between control commands and data packets, a multi-level causal verification chain should be generated based on the constructed command and data packet timing pairs, covering the network transport layer, scheduling triggering layer, and task execution layer. This verification chain uses each control signal and corresponding data packet pair as a node, with the timing offset between the control triggering order and the data receiving order as an edge attribute, spanning multiple task scheduling stages. During construction, potential state transitions during task execution, such as suspension before migration, saving during migration, and resumption after migration, should be considered, and the contextual changes of task processing nodes should be marked in the verification chain. Furthermore, for scenarios where the same node receives multiple control signals within the same time period but only responds to one, parallel branches should be established in the verification chain to express the multi-source interference impact on the task decision path. By establishing this cross-level verification chain, the timing logic between each level—from control command generation to execution and from data packet encapsulation to decapsulation—can be clearly demonstrated, and the causal relationship between logical triggering and physical implementation can be explicitly expressed.

[0066] After constructing the causal verification chain and identifying the logical-physical layer timing offset between instructions and data packets, time redirection is required for all scheduling commands and data packets exhibiting inconsistent behavior to restore a unified execution order. During time redirection, the execution priority of control signals is adjusted within the time domain based on the expected execution time of the control signals and the actual arrival time of the data packets. For task data packets arriving earlier than expected, their entry into the processing flow should be delayed, while the activation time of the control commands is moved forward to the moment before the task is ready, ensuring synchronous overlap between logical layer triggering and physical layer response. In cases where a delay in the control signal confirmation response forces task processing to start automatically, the task should be marked as asynchronously active, and the confirmation flag of the control commands should be associated with the data packet processing flow in parallel to simulate a delay compensation feedback mechanism in a synchronous control scenario. Furthermore, for the phenomenon of packet processing stage jumps caused by incorrect ordering of multiple control commands, the order of control signal issuance should be reconstructed through the instruction sorting mechanism, and their timing nodes in the evolution baseline should be rearranged to ensure that a logically traceable and predictably executed control behavior flow is finally formed on the redirected timeline, thereby restoring stability and consistency in the scheduling process.

[0067] The pre-switching module constructs a reversible queue fence and shadow channel switching sequence based on the verification results of the causal check chain. Before the main path switching, it performs a consistency pre-switching through the shadow path and performs the main path switching operation after verifying the stability of the control signal.

[0068] After completing the timing consistency comparison and time redirection between scheduling instructions and data packets through a causal verification chain, to ensure that control signals do not experience abnormal drift, redundant triggering, or failure interruption during path switching, an execution structure with rollback capabilities and a pre-simulation mechanism is constructed before the main path switch to guarantee the controllability of behavior and the continuity of task states during the switching phase. The specific steps are as follows:

[0069] Based on the control signal trigger points, path change windows, and time redirection results marked in the causal verification chain, a buffer structure with state preservation capabilities should be established to temporarily encapsulate task scheduling instructions and related data messages that are about to enter the path switching phase. This structure serves as the behavioral boundary for the critical scheduling period. Scheduling commands should be arranged in the order of triggering and configured with unique identifiers to ensure that each command has a clear behavioral attribution and response expectation before execution. Simultaneously, to avoid state conflicts or execution overlaps during path switching, a rollback flag needs to be introduced into this buffer structure. For each instruction, its source path number, control cycle number, and whether it has passed consistency comparison status information should be recorded. When any relay node or target node in the scheduling path returns an inconsistent response or execution anomaly signal during shadow path testing, the system can quickly terminate the execution of the current instruction based on the rollback flag and restore to the previous stable node in the buffer state, thereby achieving reversible control of scheduling behavior.

[0070] After constructing the instruction buffer structure with rollback capability, the time redirection information generated in the causal verification chain should be used as input to select a set of paths similar in structure to the main path but physically isolated, as the shadow path execution channel. This shadow path must be consistent with the main path in topology, node type, and protocol configuration to ensure that instruction distribution, data packet forwarding, and response processing in the shadow path are structurally equivalent to the actual execution behavior of the main path. After selecting the shadow path, the instruction stream from the aforementioned buffer structure should be injected into the shadow path sequentially, and the propagation trajectory of each instruction, relay node response time, task state change process, and final result reception time should be recorded. By comparing the timing of data packets returned by each node in the shadow path, instruction state feedback, and control behavior trajectory, a set of indicators for evaluating the stability of control signal propagation in the switching path should be established to determine whether the conditions for formal path switching are met.

[0071] During the shadow path execution phase, key attention should be paid to whether behavioral deviations, excessive processing delays, or inconsistent feedback occur during the propagation of control signals between different nodes. For each control command propagating in the shadow path, a complete timeline from the initiating node to each relay node and the final execution node must be recorded, and the transmission time, command confirmation time, and task status feedback delay at each hop must be calculated. In the shadow path structure, if the command time deviates significantly from the main path baseline, or if the command order is inconsistent with the logical order defined in the causal verification chain, the path should be marked as unstable and prohibited from participating in the main path switching process. Conversely, if the propagation behavior of the command in the shadow path is consistent with the timing defined in the verification chain of the previous stage, and the response feedback of each node is stable, it indicates that the control signal has the ability to propagate stably in this path and can be used as a candidate channel for main path switching. In this process, an intermediate task state verification mechanism should also be introduced to compare the temporary results generated during execution with the main path task state to confirm the continuity of behavior and data consistency.

[0072] After completing the shadow path consistency rehearsal and confirming that all indicators meet the switching requirements, control signals should be released from the buffer structure to the actual main path according to the redirected scheduling timeline, and path switching operations should be executed sequentially. During the release process, it is necessary to ensure that the current task scheduling node is in a state where no other scheduling commands are active, to prevent state conflicts caused by the simultaneous activation of the old and new paths. During the main path switching, for all relay nodes and target nodes participating in the switching, the relay behavior and command response templates recorded in the shadow path should be pre-loaded to reduce execution deviations caused by inconsistent node states during the path switching process. At the same time, to prevent interruptions or command misalignments during the switching process, the original shadow path channel should be continuously monitored until all control signals have completed confirmation feedback in the main path and entered a stable execution phase before the shadow path link can be closed. After the control signals have been executed stably, the task scheduling table and time series labels are updated according to the status information returned by the nodes, marking the completion of the path switching operation and achieving a stable transition of path change and seamless connection of task execution.

[0073] The interference suppression module, based on the pre-simulation results of the shadow channel switching sequence and the buffer isolation mechanism provided by the reversible queue fence, performs phase conjugate timing back-turn suppression operation and polarization channel isolation linkage operation. It completes path drift and control signal deduplication processing within the peak period of network congestion, constructs a dynamic control closed loop, and realizes real-time suppression of link interference and stability assurance of task migration.

[0074] After completing the consistency pre-simulation of the shadow channel and confirming its stable control signal transmission capability, combined with the previously constructed buffer structure with state rollback capability, dynamic path fine-tuning and signal behavior optimization operations can be performed under high network load conditions. This forms a real-time suppression mechanism for link superposition interference, ensuring scheduling continuity and computational stability during task migration. The specific steps are as follows:

[0075] The propagation trajectory data formed during the rehearsal of the shadow channel switching sequence should be used as a reference benchmark. Combined with the instruction states and propagation path records cached in the buffer structure, a phase conjugate timing reversal suppression operation based on time axis reverse mapping should be performed. Specifically, at the path switching node, the propagation timing of each control signal in the shadow path is extracted, and its propagation behavior between nodes is calibrated by time period. By calculating the relative phase difference between each signal, instruction groups forming high-density propagation intersections within a certain time window are identified. After identifying the instruction set with a tendency for propagation phase overlap, a set of mirror timelines is constructed based on the original transmission time and reception confirmation time recorded in the signal buffer structure. These signals are then reverse-mapped according to their time mirror direction to identify time periods that may form multiple activations within the node's reception window. For these highly overlapping time periods, a control signal propagation interval adjustment operation is performed, resetting the release timing of instructions with excessively close times and inserting time suppression flags in the buffer, thereby avoiding concentrated triggering and repeated execution of signals at the path switching point.

[0076] Based on the propagation phase analysis and time window integration, polarization channel isolation linkage operations need to be implemented at locations in the link topology where interference-susceptible nodes exist. This physically reduces the risk of crosstalk between signal propagation channels. Specifically, the sets of nodes involved in the handover process between the main path and the shadow path are classified. Based on the bandwidth stability of the relay links connected to the nodes, the historical fluctuation curve of the transmission load, and the topology density of adjacent nodes, the network is divided into multiple propagation channels, and polarization signal tags are configured for high-density links. These signal tags use frequency domain modulation characteristics and path fluctuation trends as references, setting mutually exclusive transmission directions and signal content feature identification rules. During propagation, similar control commands propagating in physically close channels within the same transmission cycle are restricted from entering the same node's processing queue. By implementing hierarchical separation and scheduling of propagation channels, a signal flow isolation mechanism is established in the physical distribution of channels, thereby significantly reducing control command crosstalk caused by network link jitter.

[0077] After completing the time-domain compression of the propagation rhythm and the spatial isolation of the channel, path drift control is performed on links entering a congestion cycle, combined with the network status monitoring mechanism. In this stage, using path stability assessment data recorded in the shadow channel rehearsal and the load monitoring indicators currently fed back by the main path, the availability of all possible scheduling paths for task migration is ranked, and the second-best path with the lowest latency, fewest hops, and optimal bandwidth under the current network conditions is selected as the drift candidate path. Upon confirming that the current path has experienced high latency, high packet loss rate, or confirmation feedback timeout, a micro-periodic drift operation of the scheduling path is automatically triggered, transferring the current scheduling task hop-by-hop to the candidate path. At the completion of each hop migration, the control signal state saved in the previously set buffer structure is invoked to ensure that the signal can continue to propagate from the previous confirmation node in the new transmission path without being restarted. During the path drift process, the signal execution state between the starting and ending nodes of each migration stage must be compared and confirmed once to ensure that the task state is not misaligned or interrupted.

[0078] After path drift is completed, deduplication of control signals should be performed synchronously to address potential issues of duplicate signal distribution or mirror execution during path migration. This operation uses the fingerprint characteristics of control signals, combined with historical propagation records in the path evolution baseline and time consistency data in the causal verification chain, to uniquely verify each control command entering the scheduling target node. Each control signal contains five key elements during its formation phase: path number, task identifier, instruction type, execution timestamp, and confirmation feedback status. By performing a full comparison of these elements, duplicate commands with the same task identifier and execution target are identified and automatically masked, retaining only the version that completed the confirmation feedback earliest for scheduling. Simultaneously, using a pre-defined reversible queue fence structure, the execution status of the valid instructions formed after deduplication is written back and updated, ensuring that only one consistent control behavior sequence exists in the task scheduling table. This constructs a dynamic control closed loop based on shadow paths for verification and propagation control as the core support, achieving real-time extinguishing of link interference and ensuring the stability of task migration.

[0079] This invention constructs a refined path evolution baseline and control signal feature set through path modeling and fingerprinting analysis, significantly improving the ability to detect potential link conflicts and abnormal control signal behavior. It restores logical consistency between control commands and data flows through causal verification mechanisms and time redirection strategies. Combined with shadow path pre-simulation and rollback mechanisms, it ensures that command execution during main path switching has verification support and fault tolerance capabilities. Finally, it establishes a dynamic real-time extinguishing mechanism for link interference using propagation timing compression and channel physical isolation technologies. In summary, this invention can achieve accuracy in control signal scheduling, continuity in path switching, and stability in computing power migration in dynamic network environments, thereby comprehensively improving the reliability and robustness of the cloud-edge-device integrated scheduling system.

[0080] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A cloud-edge-device collaborative computing power scheduling system that supports dynamic task migration, characterized in that: It includes a path modeling module, a conflict identification module, a timing verification module, a pre-simulation handover module, and an interference suppression module: The path modeling module, based on the dynamic topology of cloud-edge transmission links, constructs a route convergence time profile and link playback spectrum, models the overlap window of old and new paths and the life cycle of control signals, and generates a path evolution baseline for subsequent conflict identification. The conflict identification module generates a fingerprint feature set of control signals based on the path evolution baseline. By analyzing the timing deviation and transmission trajectory of the control signals, it constructs a control command contradiction relationship diagram to identify control command misalignment sources and false triggering intervals. The timing verification module establishes a causal verification chain across network protocol layers based on the control command contradiction relationship diagram. By comparing the consistency and monotonicity of the timestamps of control signals and data packets, it completes the time redirection of scheduling commands and data packets. The pre-switching module constructs a reversible queue fence and shadow channel switching sequence based on the verification results of the causal check chain. Before the main path switching, it performs a consistency pre-switching through the shadow path and performs the main path switching operation after verifying the stability of the control signal. The interference suppression module, based on the pre-simulation results of the shadow channel switching sequence and the buffer isolation mechanism provided by the reversible queue fence, performs phase conjugate timing back-flip suppression operation and polarization channel isolation linkage operation to complete path drift and control signal deduplication processing within the peak period of network congestion.

2. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 1, characterized in that, The steps for generating the path evolution baseline are as follows: Based on the physical communication structure and connection node relationship between the cloud and the edge, the connection status changes of edge nodes, scheduling strategy execution information, task migration request response time and cloud resource feedback information are collected in continuous time periods. The connection status and key transmission parameters of each path at each moment are recorded to construct a routing convergence time sequence profile. The connection state is associated with the task scheduling behavior. Based on the number of hops, transmission delay and access node information of each hop path, the complete transmission path of the control command from the sending node to the receiving node is reconstructed, the link playback spectrum is constructed and the overlapping areas of the path are marked. Based on the propagation trajectory of control commands in the link playback spectrum, the start and end times of the propagation of control signals between the old and new paths, the path intersection time window, and the command execution feedback time are marked, and a life cycle description model of control signals is established. Combining the aforementioned time-series profile, link playback spectrum, and lifecycle model, a unified mapping is applied to the time-labeled sequence to label the active path set, control signal set, and path overlap relationship at each moment, thus constructing a path evolution baseline.

3. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 2, characterized in that, When constructing the path evolution baseline, the active path set, control signal set, and path overlap relationship corresponding to each time label are jointly labeled with network congestion level, bandwidth fluctuation frequency, and link switching response delay to enhance the adaptability of the path evolution baseline to dynamic changes in links.

4. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 2, characterized in that, The steps for constructing a control command contradiction diagram are as follows: Based on the propagation trajectory of control signals in the path evolution baseline, the generation time, forwarding time and arrival time of control commands in each path segment are collected to construct a tag set; Based on the control signal tag set, its propagation timing, path overlap and feedback behavior are extracted to generate a unique fingerprint structure. The signal feature vector is then normalized by combining historical task behavior to construct a fingerprint feature set. Based on fingerprint feature sets, control signal pairs with consistent targets and propagation conflicts are selected, a control command conflict relationship graph is constructed, and the behavioral conflict paths and propagation time windows are marked. Based on the information of conflict edges in the contradiction relationship graph, the propagation path of the control signal is traced back, and the switching time and link behavior in the path evolution baseline are combined to identify the source of instruction misalignment and the false triggering interval.

5. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 4, characterized in that, Each conflict edge in the control command conflict relationship graph is accompanied by a propagation time window, path intersection segment identifier, and conflict level index to help identify high-incidence segments of propagation timing disorder and the location of command misalignment sources.

6. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 4, characterized in that, Based on the control command contradiction graph, a causal verification chain is established across network protocol layers. The consistency and monotonicity of the timestamps of control signals and data packets are compared and checked. The time redirection steps are as follows: Based on the control command conflict relationship diagram, conflicting control command pairs are extracted and associated with the transmission time behavior of their corresponding data packets to establish a time mapping pair between control signals and data packets. Based on the time mapping pair, the consistency comparison between the control signal trigger time and the data packet reception time is performed to check whether there is a deviation in the order of the two at the logic layer and the transport layer, and to construct a preliminary causal conflict chain. Based on consistency deviation information, a causal verification chain covering the network transmission layer, scheduling triggering layer and task execution layer is generated, and the task state switching points and control branch relationships are marked. By referencing the causal verification chain and path evolution baseline, time redirection is performed on control signals and data messages with sequence deviations to adjust their execution timing and achieve consistent restoration of task scheduling behavior.

7. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 6, characterized in that, Time redirection involves advancing the activation time of the control signal to the synchronization point before the arrival of the task data packet, and delaying the processing of task data packets that are triggered earlier than the control signal until the control signal is confirmed.

8. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 6, characterized in that, Based on the causal verification chain results, a reversible queue fence and shadow channel switching sequence are constructed. After performing a consistency rehearsal through the shadow path, the main path switching operation steps are as follows: Based on the time redirection results of the causal verification chain, a buffer structure with state preservation capability is constructed to temporarily encapsulate scheduling instructions and data packets and configure rollback flags; Based on the control instructions in the buffer structure, a physically isolated path consistent with the main path structure is selected as the shadow path, and the instruction stream is injected in sequence, and the propagation trajectory and response behavior of each node are recorded. By evaluating the timing consistency of the control signal propagation behavior in the shadow path, it is determined whether it meets the stable propagation conditions, and intermediate state consistency is compared. Release the control commands in the buffer structure to the main path, perform a path switching operation and load the response template in the shadow path until the control signal completes the confirmation feedback and updates the task scheduling table.

9. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 8, characterized in that, Before the control signal is released to the main path, the relay behavior and command response results in the shadow path are templated, and the corresponding template is loaded in the main path node to match the execution status, ensuring the consistency and continuity of the control signal response during path switching.

10. The cloud-edge-device collaborative computing power scheduling system supporting dynamic task migration according to claim 8, characterized in that, Based on the pre-simulation results of the shadow channel switching sequence and the buffer isolation mechanism of the reversible queue fence, the following steps are performed to carry out path drift and control signal deduplication during the peak period of network congestion: Phase conjugate timing backflip suppression operation and polarization channel isolation linkage operation are executed. Based on the shadow channel switching sequence, a pre-simulation trajectory and buffer isolation mechanism are used to perform phase conjugate timing back-turn suppression operation, reverse mapping of dense propagation instructions and reset release time; Based on the location of vulnerable nodes in the path topology, polarization channel isolation linkage operation is implemented to group and isolate the propagation channels and set signal labels to avoid crosstalk. During network congestion periods, micro-periodic path drift operations are performed using the stability assessment results of shadow paths, and hop-by-hop task transfers are completed by calling the control signal states recorded in the cache structure. After the path drift is completed, the control signal deduplication process is performed based on the fingerprint characteristics of the control signal, and the valid instruction status is written back through the reversible queue fence to build a dynamic control closed loop.

Images