Industrial internet platform micro-service architecture elasticity and resource scheduling system

Abstract

The application relates to the technical field of computer software and industrial internet, and particularly discloses a micro-service architecture elasticity expansion and resource scheduling system of an industrial internet platform, which comprises a micro-service cluster monitoring unit, a topological relation modeling unit, an information entropy calculation unit, a negative entropy flow scheduling decision unit and an elasticity expansion execution unit; through real-time collection of multi-dimensional load data and construction of a dynamic topological atlas, system information entropy is calculated based on the principle of non-equilibrium thermodynamics, task migration and expansion and contraction instructions are generated according to the heat conduction mechanism of the entropy value change trend, and the negative entropy flow is guided to reduce the system disorder degree while guaranteeing the service call affinity. Through the technical scheme, dynamic rebalancing and efficient utilization of cluster resources are realized, and the elasticity response capability and overall robustness of the industrial internet platform in a high-concurrency and strong-coupling scene are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of computer software and industrial internet technology, specifically relating to a microservice architecture elastic scaling and resource scheduling system for industrial internet platforms. Background Technology

[0002] With the deepening evolution of Industrial Internet technology, microservice architecture, with its high scalability and modularity, has become the core architecture supporting complex industrial applications and massive device access. In the context of intelligent manufacturing and the Industrial Internet, Industrial Internet platforms need to handle business flows with strong coupling and dynamic fluctuations, making the underlying elastic scaling and resource scheduling system crucial for ensuring business continuity. Microservices achieve flexible scaling through containerized deployment, and their performance directly determines the resource allocation efficiency and response speed of the industrial platform when dealing with complex production tasks.

[0003] The elastic scheduling mechanism of microservice clusters aims to dynamically adjust the allocation of computing resources based on real-time load. To address the demands of multi-process collaboration and high-concurrency interaction in industrial scenarios, the scheduling system must accurately identify the status of each microservice node and distribute tasks appropriately based on the topological relationships between services. Through refined management of computing resources within the cluster, the system's throughput and computational efficiency in processing industrial big data and complex instruction streams can be significantly improved.

[0004] However, traditional scheduling algorithms often lack effective assessment of the overall system load order when faced with complex call chains and traffic interactions between microservices, leading to severe fragmentation of resource utilization. Existing static polling or random allocation mechanisms struggle to resolve affinity constraints between tasks, easily causing local hotspots to accumulate on specific nodes, thus triggering end-to-end performance blocking. Simultaneously, traditional systems lack the ability to dynamically perceive resource distribution under unbalanced conditions, failing to guide the system from a disordered state to a low-entropy ordered state through effective negative entropy mechanisms. This makes it difficult for the cluster to spontaneously achieve accurate load balancing when the load is uneven. Furthermore, resource mapping models based on linear logic cannot capture the nonlinear coupling relationships between multidimensional load parameters, resulting in significant lag in scheduling decisions under extreme conditions, severely impacting the overall robustness and resource utilization efficiency of industrial internet platforms.

[0005] Therefore, the industrial internet platform is expected to have a microservice architecture with elastic scaling and resource scheduling system. Summary of the Invention

[0006] The purpose of this invention is to provide a microservice architecture elastic scaling and resource scheduling system for industrial internet platforms, which can effectively solve the problems of resource fragmentation and local hotspot accumulation in the background technology.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The microservice architecture elastic scaling and resource scheduling system of the industrial internet platform includes a microservice cluster monitoring unit, a topology modeling unit, an information entropy calculation unit, a negative entropy flow scheduling decision unit, and an elastic scaling execution unit, wherein: The microservice cluster monitoring unit is configured to collect multi-dimensional load status data of each node in the microservice cluster in real time, including CPU utilization, memory usage, network throughput and disk input / output rate, and synchronize the multi-dimensional load status data to the information entropy calculation unit. The topology modeling unit is configured to parse the call chain and data interaction relationship between microservices, construct a dynamic topology graph that reflects the affinity and dependency strength between services, and provide the dynamic topology graph to the negative entropy flow scheduling decision unit as a task migration constraint. The information entropy calculation unit is configured to calculate the information entropy value that characterizes the overall load balance of the system based on the multidimensional load state data and the dynamic topology map, wherein the more uneven the load distribution, the higher the information entropy value. The negative entropy flow scheduling decision unit is configured to evaluate the impact of different task migration schemes on the system orderliness based on the changing trend of the information entropy value, select the task migration path that can reduce the system information entropy, and generate the corresponding resource reallocation instruction. The elastic scaling execution unit is configured to receive the resource reallocation instruction and perform container-level scaling up / down operations or cross-node migration operations on the target microservice instance to achieve dynamic rebalancing of computing resources within the cluster.

[0008] Preferably, the information entropy calculation unit is further configured to treat the microservice cluster as a non-equilibrium thermodynamic system, map the data traffic and call frequency between services to energy flow density, and construct an information entropy model in a multi-dimensional phase space based on statistical thermodynamic principles.

[0009] Furthermore, the negative entropy flow scheduling decision unit is configured to simulate a heat conduction mechanism. After identifying high-load nodes as high-temperature areas, it prioritizes migrating some tasks to low-load nodes that have strong topological connections with them and whose current load is below a preset threshold, thereby introducing negative entropy flow while maintaining service call efficiency.

[0010] Furthermore, the topology modeling unit is configured to periodically update the call frequency matrix and data interaction intensity matrix between microservices, and identify potential service clusters based on the community detection algorithm in graph theory, so that subsequent scheduling decisions can optimize the global load distribution while maintaining the locality of intra-cluster communication.

[0011] Preferably, the elastic scaling execution unit is configured to verify whether the resource reserves of the target node meet the minimum resource requirements of the microservice to be migrated before performing the task migration, and to start the rapid pulling of the container image and instance reconstruction process after confirming that they meet the requirements, so as to ensure that the impact of the migration process on business continuity is controlled within a predetermined time.

[0012] Furthermore, the microservice cluster monitoring unit is configured to filter abnormally fluctuating load indicators, eliminate instantaneous spike interference, and generate a smoothed load trend curve by combining historical data within the sliding time window for the information entropy calculation unit to perform stability assessment.

[0013] Furthermore, the negative entropy flow scheduling decision unit is configured to trigger a global resource reorganization mechanism when the system information entropy is continuously higher than a preset threshold and there is no effective migration path. This mechanism temporarily freezes the scaling requests of non-critical services and centrally schedules idle resources to alleviate the overload pressure on the core service links.

[0014] Preferably, the information entropy calculation unit and the negative entropy flow scheduling decision unit work together to form a closed-loop feedback control structure, wherein the actual load distribution after each scheduling operation is re-collected and used to correct the information entropy prediction model for the next cycle, thereby continuously improving the adaptability and foresight of the scheduling strategy.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The microservice architecture elastic scaling and resource scheduling system of the industrial internet platform provided by this invention, by introducing non-equilibrium thermodynamics theory, transforms the resource scheduling problem of microservice clusters into a negative entropy flow guidance process to maintain the low-entropy ordered state of the system, fundamentally solving the problem of local hotspot accumulation and resource fragmentation caused by the lack of global order measurement in traditional scheduling algorithms.

[0016] 2. Based on dynamic topology and multidimensional load data, this invention can accurately identify high-entropy disordered states and proactively implement a task migration strategy that simulates heat conduction, thereby ensuring service call affinity while achieving efficient utilization of cluster computing power.

[0017] 3. This invention enhances the elastic response capability and overall robustness of the industrial internet platform when dealing with high concurrency and strongly coupled business flows, avoiding full-link performance blockage caused by single-point overload, thereby providing a highly reliable and intelligent resource scheduling infrastructure for complex industrial application scenarios. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2This is a schematic diagram of the core principle framework of information entropy calculation and negative entropy flow scheduling based on a non-equilibrium thermodynamic model in this invention; Figure 3 This is a flowchart of the main stages of the process based on multidimensional load data acquisition, smoothing processing and state assessment in this invention. Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the information entropy evaluation, scheduling decision and feedback adjustment mechanism in this invention; Figure 5 This is a flowchart illustrating the logical flow of task migration path selection and resource reallocation in this invention, which simulates the heat conduction mechanism. Detailed Implementation

[0019] Example 1: To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0020] Please refer to Figures 1 to 5 This embodiment provides a microservice architecture elastic scaling and resource scheduling system for an industrial internet platform, including a microservice cluster monitoring unit, a topology relationship modeling unit, an information entropy calculation unit, a negative entropy flow scheduling decision unit, and an elastic scaling execution unit.

[0021] The microservice cluster monitoring unit is used to collect multi-dimensional load status data of each node in the microservice cluster in real time. The microservice cluster monitoring unit includes a distributed data acquisition module, a load feature extraction module, and a load data filtering module. The distributed data acquisition module is deployed on each physical node or virtual machine node of the microservice cluster and is configured to periodically extract the resource consumption of each microservice instance within that node by reading kernel-level metrics and container engine interfaces. The multi-dimensional load status data includes not only CPU utilization and memory usage, but also network throughput, disk I / O rate, and process-level context switching frequency, reflecting the real-time interactive characteristics of the Industrial Internet.

[0022] The load data filtering module is connected to the distributed data acquisition module and internally stores a preset adaptive smoothing operator. The load data filtering module is configured to denoise the acquired raw indicators. Specifically, this module identifies instantaneous spikes in the load indicators and, based on historical distribution characteristics within a sliding time window, removes or averages outliers exceeding a preset confidence interval, generating a smoothed load trend curve. The load feature extraction module is configured to extract higher-order features such as the rate of change, load variance, and peak period from the smoothed curve, and synchronizes this processed structured load data to the information entropy calculation unit, providing accurate underlying data support for subsequent orderliness assessment.

[0023] The topology modeling unit is used to parse the call chains and data interaction relationships between microservices. The topology modeling unit includes a link tracing subunit, a dynamic graph construction subunit, and a service affinity assessment subunit. The link tracing subunit is configured to capture all remote procedure call logs and message queue interaction records generated by the microservices during operation, and extract the call depth, call frequency, and data volume carried in a single request between each pair of microservices.

[0024] The dynamic graph construction subunit is connected to the link tracing subunit and is configured to construct a dynamic topology graph reflecting the dependencies between services in the entire cluster based on the extracted call information and using a graph theory model. In the dynamic topology graph, each microservice instance is mapped to a node, and the call relationships between services are mapped to weighted directed edges. The weight calculation logic is configured to: perform a weighted sum of call frequency and data interaction intensity to characterize the tightness of coupling between two service nodes.

[0025] The service affinity assessment subunit is connected to the dynamic graph construction subunit and is configured to periodically run a community detection algorithm to identify frequently interacting service clusters in the topology graph. The service affinity assessment subunit provides the identified cluster characteristics and the real-time status of the dynamic topology graph to the negative entropy flow scheduling decision unit as key constraints during task migration, ensuring that local communication efficiency between strongly coupled services is preserved to the maximum extent possible during resource reallocation.

[0026] The information entropy calculation unit is used to calculate the information entropy value, which characterizes the overall load balance of the system, based on the multidimensional load state data and the dynamic topology map. The information entropy calculation unit integrates a thermodynamic mapping logic module, a phase space construction submodule, and an entropy quantification and evaluation submodule. The thermodynamic mapping logic module is configured to abstract the microservice cluster as a non-equilibrium thermodynamic system, where the total computing resources of each node are defined as the total heat capacity of the system, and the real-time load consumption is mapped to the current internal energy state of the node. Data traffic and call frequency between services are mapped to the energy flow density within the system.

[0027] The phase space construction submodule is connected to the thermodynamic mapping logic module and is configured to construct a multi-dimensional phase space based on CPU utilization, memory usage, network throughput, and topological coupling. Within this space, the current operating state of the system is described as a probability distribution function. The entropy quantification and evaluation submodule is configured to quantify the probability distribution function based on statistical thermodynamics principles. The specific calculation logic is as follows: for each node, the local probability density of its load distribution in the multi-dimensional space is calculated. This probability density is multiplied by its natural logarithm, and the results for all nodes in the cluster are summed and their negatives are taken to obtain the information entropy value representing the overall disorder of the system. A larger information entropy value indicates a more uneven load distribution, with the system in a high-entropy chaotic state; a smaller information entropy value indicates a more balanced load distribution, with the system in a low-entropy ordered state.

[0028] The negative entropy flow scheduling decision unit is used to evaluate the impact of different task migration schemes on the system's orderliness based on the changing trend of the information entropy value. The negative entropy flow scheduling decision unit includes a migration path simulation module, a negative entropy effect evaluation module, a scheduling scheme optimization module, and a global reorganization triggering module. The migration path simulation module is connected to the information entropy calculation unit and the topology relationship modeling unit, and is configured to simulate a heat conduction mechanism for decision-making. Specifically, this module identifies nodes with significantly higher-than-average loads and marks them as high-temperature regions; simultaneously, it identifies nodes with lower loads that are topologically connected to these high-temperature regions and marks them as low-temperature regions.

[0029] The migration path simulation module generates multiple potential task migration path candidate sets to simulate the process of transferring a specific microservice task from a high-temperature region to a low-temperature region. The negative entropy effect evaluation module is configured to predict the change in system information entropy, i.e., entropy change, after each migration path is implemented. Only when the predicted entropy change is negative is the path considered a negative entropy flow that contributes to the system's orderliness.

[0030] The scheduling scheme optimization module is configured to select the optimal path from all predicted entropy-negative schemes that maximizes the reduction in information entropy while simultaneously satisfying topology constraints and migration cost constraints, and then generate the corresponding resource reallocation instruction. The global reorganization trigger module is configured to initiate a global resource reorganization mechanism when the overall system information entropy continuously exceeds a preset safety threshold and local migration fails to achieve effective entropy reduction. This mechanism temporarily freezes resource requests from non-core services and forces cross-physical partition resource reclamation and reallocation, alleviating overload pressure on core links at a macro level and forcibly introducing negative entropy to restore system stability.

[0031] The elastic scaling execution unit is used to receive the resource reallocation instruction and perform container-level scaling operations or cross-node migration operations on the target microservice instance. The elastic scaling execution unit includes a resource pre-inspection subunit, a container orchestration control subunit, and an execution feedback monitoring subunit. The resource pre-inspection subunit is configured to, upon receiving the instruction, first verify whether the available resource balance of the target node is greater than the minimum resource threshold required by the service to be migrated. The container orchestration control subunit is connected to the resource pre-inspection subunit and integrates a communication interface with the underlying container cluster management system. After passing the pre-inspection, this subunit issues instructions to execute the smooth decommissioning of the original node instance, the rapid pulling of the target node image, and the startup process of the new instance.

[0032] The execution feedback monitoring subunit is configured to track key time points during the migration process in real time, including image pull duration, instance readiness duration, and request takeover success rate. After the migration is complete, the execution feedback monitoring subunit will feed back the actual resource distribution to the microservice cluster monitoring unit, forming a complete closed-loop control loop. In this way, the system can continuously use actual execution results to revise the information entropy prediction model, thereby making subsequent negative entropy flow scheduling decisions more forward-looking and accurate.

[0033] Example 2: Based on the microservice architecture elastic scaling and resource scheduling system of the industrial internet platform described in Example 1, this example provides a variant of the distributed hierarchical scheduling architecture for ultra-large-scale heterogeneous computing environments. In this example, the various units of the system are further refined and enhanced in terms of physical deployment and logical hierarchy.

[0034] An edge preprocessing proxy layer is introduced into the microservice cluster monitoring unit. For microservices deployed at edge nodes in industrial sites, this edge preprocessing proxy layer is configured to first perform preliminary aggregation and feature compression of high-frequency sampled data locally, and only upload key feature vectors representing load trends to the information entropy calculation unit in the cloud. This layered monitoring mode greatly reduces the bandwidth consumption of monitoring data on the backbone network in scenarios with large-scale device access, while ensuring the real-time performance of monitoring.

[0035] In this embodiment, the topology modeling unit is configured to have multi-level topology resolution capabilities. Besides identifying logical service call relationships, this unit also acquires physical network topology information, including switch cascading levels, fiber optic link delays, and subnet isolation status. The dynamic graph construction subunit uses the remaining bandwidth of physical links as a correction factor for edge weights, constructing a multi-dimensional constrained composite topology graph by overlapping and mapping the logical and physical topologies. This enables subsequent scheduling decisions to effectively avoid network congestion points and prevent new congestion caused by physical network bottlenecks during the introduction of negative entropy flows.

[0036] In this embodiment, the information entropy calculation unit employs a hierarchical entropy calculation model. The system divides the entire microservice cluster into several logical availability zones, with a local information entropy calculation submodule running within each availability zone. This local information entropy calculation submodule is configured to calculate the local load balancing within its area. The top-level information entropy calculation unit is responsible for collecting the local entropy values ​​from each availability zone and, combined with cross-zone traffic distribution, calculating the global system information entropy. This hierarchical calculation mechanism allows the system to first attempt to resolve hotspot issues within the availability zones through local negative entropy flow. Only when local adjustments fail does it escalate to a large-scale global resource rebalancing, thereby significantly improving system response efficiency and reducing scheduling oscillations.

[0037] In this embodiment, the negative entropy flow scheduling decision unit incorporates a predictive heat conduction simulation module. This module no longer relies solely on current real-time load data but incorporates a service traffic prediction model based on a long short-term memory network. The predictive heat conduction simulation module is configured to identify nodes that may evolve into high-temperature zones in advance based on service traffic forecasts for a preset time period, and to pre-calculate preventative negative entropy injection paths. This proactive scheduling mechanism allows the system to guide resources towards an orderly flow before load imbalance actually occurs, thereby maintaining greater robustness in the face of dynamically fluctuating industrial production task flows.

[0038] In this embodiment, the elastic scaling execution unit is configured to support heterogeneous resource scheduling logic. When the instructions generated by the decision unit involve the migration of nodes with different instruction set architectures, the container orchestration control subunit can automatically select and deploy an image version that matches the target node architecture. Simultaneously, the resource pre-inspection subunit will also verify whether the target node possesses the corresponding specific hardware resources, addressing common hardware accelerator requirements in industrial applications, to ensure that the migrated service can maintain its original high-performance computing capabilities.

[0039] Through the aforementioned layered and heterogeneous enhancement design, the system described in this embodiment can more effectively support the needs of industrial internet platforms across regions and multiple data centers, while ensuring a globally low-entropy and orderly state and significantly optimizing the system's adaptability in complex physical environments.

[0040] Example 3: Building upon Examples 1 and 2, this example details a microservice architecture elastic scaling and resource scheduling system for an industrial internet platform integrating proactive fault self-healing and performance guarantee mechanisms. This system is particularly enhanced in its ability to elastically reorganize resources in response to partial hardware failures or extreme load surges.

[0041] In this embodiment, the microservice cluster monitoring unit adds a hardware health monitoring module. This module is configured to generate a node health score by monitoring changes in node power consumption, fan speed fluctuations, processor core temperature, and disk addressing error rate. This health score is input into the information entropy calculation unit as a special energy conversion efficiency parameter. When a node's hardware indicators show abnormalities but are not completely failed, this module automatically reduces the node's weight in the resource scheduling model, making it appear as a region with higher impedance in the information entropy model, thereby guiding tasks naturally to nodes with higher health.

[0042] In this embodiment, the topology modeling unit is configured to perform risk propagation path analysis. The service affinity assessment subunit simulates how this disordered state propagates upstream and downstream along the call chain when a core service node in the topology experiences overload and congestion. By calculating this entropy increase diffusion coefficient, the system can identify the critical paths in the cluster that have the greatest impact on overall stability. The topology modeling unit marks the vulnerability characteristics of these critical paths in a dynamic graph for the decision-making unit to perform priority protection scheduling.

[0043] In this embodiment, the information entropy calculation unit introduces the concept of potential energy difference to optimize the entropy quantification logic. The system not only calculates the current load entropy but also the remaining potential energy of each node relative to its rated capacity. The entropy quantification evaluation submodule is configured such that when the load difference between two nodes causes their potential energy difference to exceed a preset threshold, this potential energy difference is converted into a driving force that propels the spontaneous flow of negative entropy. This logic simulates the spontaneous transfer of heat driven by temperature differences in the real physical world, enabling the scheduling algorithm to converge faster when handling large-scale load bursts.

[0044] In this embodiment, the negative entropy flow scheduling decision unit integrates a self-healing strategy generation module. When the system detects a sharp increase in information entropy in a local area at a non-linear rate, indicating a potential avalanche overload, this module immediately enters emergency mode. In emergency mode, the scheduling scheme optimization module relaxes the restrictions on migration costs, prioritizing a large-span negative entropy injection strategy. This involves directly allocating resources from a remote ultra-low load pool to offset computing power, and even triggering temporary cross-cloud resource expansion. Simultaneously, this module collaborates with the elastic scaling execution unit to implement second-level rapid scaling for core service links, quickly diluting overload pressure by increasing instance redundancy and forcibly reducing system entropy.

[0045] In this embodiment, the elastic scaling execution unit particularly strengthens the image pre-distribution and warm-start logic. The instance deployment module is configured to push commonly used microservice image packages to the local caches of various possible low-temperature region nodes in advance, based on the potential migration paths predicted by the negative entropy flow scheduling decision unit. When the formal reallocation instruction is issued, the container orchestration control subunit can directly start the container from the local cache, eliminating the latency of pulling images from the network and ensuring that the entire negative entropy introduction process can be completed in linear time, minimizing industrial business interruptions caused by scheduling delays.

[0046] Furthermore, the execution feedback monitoring subunit in this embodiment also has a scheduling effect backtracking function. It continuously records the actual curve of the system entropy value decrease after each round of negative entropy injection and compares it with the prediction model. If the prediction deviation is found to exceed a preset ratio, the system will automatically trigger a parameter self-calibration process, adjusting the thermal conductivity coefficient or entropy value calculation weight to enable the system to adapt to different types of industrial load characteristics, such as periodic task flows or random burst command flows.

[0047] This deeply optimized design, which integrates hardware awareness, risk prediction, potential energy driving and self-healing response, makes the system described in this embodiment not only a load balancer, but also an intelligent resource governance framework with self-organizing and self-evolving characteristics.

[0048] Example 4: This example further illustrates a microservice architecture elastic scaling and resource scheduling system for an industrial internet platform based on hardware offloading acceleration and high-reliability dual closed-loop regulation. In this example, the system utilizes specific hardware resources to carry complex entropy calculation and scheduling logic, ensuring that the system's own management overhead does not become a new bottleneck under extremely high loads.

[0049] In this embodiment, a dedicated computing acceleration module is introduced into the information entropy calculation unit. This module, built on a programmable gate array (GGA) or graphics processing unit (GPU), is configured to process real-time load data streams from thousands of nodes in parallel. Since the quantification of information entropy involves numerous logarithmic operations, multiply-accumulate operations, and multidimensional matrix transformations, the dedicated computing acceleration module, by implementing a streaming computing architecture at the hardware level, can increase the update frequency of the global entropy value to the millisecond level. This high-frequency update ensures that the negative entropy flow scheduling decision unit can capture extremely brief traffic transients in the Industrial Internet, thereby enabling more precise resource fine-tuning.

[0050] In this embodiment, the negative entropy flow scheduling decision unit is configured as a two-layer decision structure. The bottom layer decision is executed by the fast response logic embedded in the hardware acceleration card, which is used to handle micro-task scheduling and instantaneous spike cancellation within a single node; the top layer decision is executed by the main control logic running on the management node, which is responsible for global topology evolution and long-cycle negative entropy guidance. The fast response logic can dynamically rearrange the task chain within the node within microseconds by monitoring local bus data. This localized, hardware-based decision mechanism greatly reduces the round-trip latency of scheduling signaling transmission in the network.

[0051] In this embodiment, the microservice cluster monitoring unit integrates a high-precision clock synchronization module. This module ensures that the sampling timestamps of all nodes within the cluster are consistent at the nanosecond level. Based on the synchronized timestamps, the load feature extraction module can construct an accurate system state snapshot. This snapshot reflects the true energy distribution of the entire cluster at the same moment, eliminating sampling timing deviations caused by network transmission delays. The entropy calculation unit, based on this high-precision snapshot, can more realistically reflect the system's equilibrium state, avoiding scheduling misjudgments caused by expired data.

[0052] In this embodiment, the topology modeling unit adds dynamic link quality awareness. It monitors not only application-layer calls but also congestion window changes in the transport layer protocol stack and the bit error rate of the physical links. When constructing the dynamic topology graph, the dynamic graph construction subunit incorporates real-time link quality loss into the migration cost model. If two migration paths have similar logical entropy reduction effects, the system prioritizes the path with better physical link quality and lower packet loss rate. This approach of deeply integrating communication physical characteristics into the topology model ensures that the negative entropy flow is not interrupted during the guidance process due to underlying network jitter.

[0053] In this embodiment, the elastic scaling execution unit employs an advanced execution mechanism called zero-copy migration. For nodes supporting shared storage or memory mirroring technology, the container orchestration control subunit can directly map the running state of the original instance on the target node without re-pulling the complete image. Combined with the execution feedback monitoring subunit in this embodiment, the system can achieve millisecond-level seamless migration of critical microservice instances. This means that even in production environments with extremely high loads and a sharp decline in orderliness, the system can instantly smooth out load fluctuations through this rapid negative entropy injection method, ensuring that the microservice architecture of the industrial internet platform always maintains a low-entropy, ordered, high-performance operating range.

[0054] This embodiment transforms thermodynamic scheduling theory into a practical solution with strong engineering feasibility through hardware acceleration and physical characteristic awareness. In large-scale, high-concurrency Industry 4.0 production environments that are extremely sensitive to latency, the system demonstrates stability and response speed that surpasses traditional software-defined scheduling schemes, providing a collaborative optimization perspective from the underlying hardware to the high-level logic for complex resource balancing problems.

[0055] Example 5: This example further provides a microservice architecture elastic scaling and resource scheduling system for an industrial internet platform based on multidimensional multi-step prediction and game theory optimization, which aims to solve the problem of non-cooperative entropy increase caused by resource competition among different business flows in a large-scale multi-tenant environment.

[0056] In this embodiment, the microservice cluster monitoring unit is equipped with business attribute identification capabilities. The load feature extraction module is configured to identify the business priority, real-time requirements, and data consistency constraints of traffic using deep packet inspection technology. These business attributes are converted into potential energy weights in the information entropy calculation unit. For control-type services with high real-time requirements, the system will artificially increase the potential energy of its node, making it more sensitive when calculating entropy values. This means that even if its absolute load is not high, once fluctuations occur, the system will identify it as a high-temperature center that needs to be cooled, thus prioritizing the introduction of negative entropy flow for protection.

[0057] In this embodiment, the information entropy calculation unit is configured to calculate the interaction correlation entropy. This unit not only focuses on the load of the node itself, but also on the mutual interference between nodes caused by resource sharing (such as shared L3 cache, shared disk bus). The phase space construction submodule maps this cross-node resource competition relationship into interaction forces within the phase space. When two tightly coupled service instances run in physically adjacent but resource-constrained environments, the system will identify that this potential competition will lead to a rapid increase in the overall entropy value, thereby generating an early warning isolation scheduling suggestion.

[0058] In this embodiment, the negative entropy flow scheduling decision unit integrates a multi-agent game optimization module. Since industrial internet platforms often run microservices from different suppliers or production lines simultaneously, these services exhibit non-cooperative characteristics when competing for cluster resources. The multi-agent game optimization module is configured to treat each microservice cluster as a game participant. When generating negative entropy scheduling paths, this module seeks a Nash equilibrium point that maximizes the reduction of the overall system information entropy while maintaining the performance indicators of each service cluster within the preset service level agreement range. This method effectively solves the problem of traditional scheduling sacrificing the performance of specific core services by solely pursuing global load balancing.

[0059] In this embodiment, the topology modeling unit is configured to predict evolution trends. The dynamic graph construction subunit analyzes the changes in the topology over multiple periods to identify which service nodes are gradually becoming new communication hubs. By predicting the evolution trend of the topology, the system can reserve sufficient elastic resource buffer zones around these potential communication hubs in advance. This topology-guided negative entropy layout strategy enables the system to maintain a low-entropy state of the overall architecture with less energy cost.

[0060] In this embodiment, the elastic scaling execution unit supports fine-grained vertical resource scaling. In addition to horizontal migration across nodes, the container orchestration control subunit can also directly adjust the resource limit settings of running containers. When the negative entropy flow scheduling decision unit determines that only minor adjustments are needed to reduce the local entropy value, the system will prioritize this in-situ scaling approach. Since no instance restart is required, this operation has almost zero impact on the continuity of industrial control services, and also reduces the network oscillation entropy increase caused by large-scale container migration.

[0061] The execution feedback monitoring subunit is also configured as a continuously evolving knowledge base system. It records the optimal negative entropy flow path characteristics under various operating conditions (such as day shift high-load production mode, night shift maintenance mode, and material switching peak mode). When the monitoring unit identifies a match between the current production mode and historical records, the decision-making unit can directly call mature strategies from the library, thereby saving the need for large-scale real-time calculations.

[0062] This comprehensive approach, based on business awareness, game theory balancing, and evolutionary prediction, enables the system described in this embodiment to handle extremely complex industrial multi-task hybrid scenarios. Through highly refined control of energy flow (data flow) and system state (load distribution), it ensures that the industrial internet platform, while pursuing maximum resource utilization, possesses extremely strong business resilience and self-organizing balancing capabilities.

Claims

1. A microservice architecture elastic scaling and resource scheduling system for an industrial internet platform, characterized in that, include: The microservice cluster monitoring unit is configured to collect multi-dimensional load status data of each node in the microservice cluster in real time. The multi-dimensional load status data includes CPU utilization, memory usage, network throughput, and disk input / output rate. The topology modeling unit is configured to parse the call chain and data interaction relationship between microservices, construct a dynamic topology graph that reflects the affinity and dependency strength between services, and use the dynamic topology graph as a task migration constraint. The information entropy calculation unit is connected to the microservice cluster monitoring unit and the topology modeling unit, and is configured to calculate the information entropy value that characterizes the overall load balance of the system based on the multidimensional load status data and the dynamic topology map, wherein the unevenness of load distribution is positively correlated with the information entropy value. The negative entropy flow scheduling decision unit is connected to the information entropy calculation unit and is configured to evaluate the impact of different task migration schemes on the system orderliness based on the changing trend of the information entropy value, select the task migration path that can reduce the system information entropy, and generate the corresponding resource reallocation instruction. The elastic scaling execution unit is connected to the negative entropy flow scheduling decision unit and is configured to receive the resource reallocation instructions to perform container-level scaling operations or cross-node migration operations on the target microservice instance, so as to achieve dynamic rebalancing of computing resources within the cluster.

2. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The microservice cluster monitoring unit includes a distributed data acquisition module, a load feature extraction module, and a load data filtering and processing module. The distributed data acquisition module is deployed on each physical node or virtual machine node of the microservice cluster. It is configured to periodically extract the resource consumption of each microservice instance within the node by reading kernel-mode metrics and container engine interfaces. The resource consumption includes CPU utilization, memory usage, network throughput, disk I / O rate, and process-level context switching frequency. The load data filtering and processing module is connected to the distributed data acquisition module. It stores a preset adaptive smoothing operator and is configured to denoise the collected raw indicators, identify instantaneous spikes in the load indicators, and based on the historical distribution characteristics within the sliding time window, remove or mean-replace outliers exceeding the preset confidence interval, thereby generating a smoothed load trend curve. The load feature extraction module is connected to the load data filtering and processing module, and is configured to extract the rate of change, load variance and peak period from the load trend curve, and synchronize the processed structured load data to the information entropy calculation unit.

3. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The topology modeling unit includes a link tracing subunit, a dynamic graph construction subunit, and a service affinity assessment subunit; The link tracing subunit is configured to capture remote procedure call logs and message queue interaction records generated during the operation of microservices, and extract the call depth, call frequency and data volume carried in a single request between each pair of microservices. The dynamic graph construction subunit is connected to the link tracing subunit and is configured to construct a dynamic topology graph reflecting the dependencies between services in the entire cluster based on the extracted call information using a graph theory model. Each microservice instance is mapped to a node, and the call relationship between services is mapped to a weighted directed edge. The weight of the directed edge is determined by the weighted sum of the call frequency and the intensity of data interaction, thereby characterizing the degree of coupling between two service nodes. The service affinity assessment subunit is connected to the dynamic topology graph construction subunit and is configured to periodically run the community discovery algorithm to identify frequently interacting service clusters in the dynamic topology graph. The identified cluster characteristics and the real-time status of the dynamic topology graph are provided to the negative entropy flow scheduling decision unit as key constraints in the task migration process.

4. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The information entropy calculation unit includes a thermodynamic mapping logic module, a phase space construction submodule, and an entropy value quantification evaluation submodule. The thermodynamic mapping logic module is configured to abstract the microservice cluster as a non-equilibrium thermodynamic system, define the total computing resources of each node as the total heat capacity of the system, map the real-time load consumption to the current internal energy state of the node, and map the data flow and call frequency between services to the energy flow density inside the system. The phase space construction submodule is connected to the thermodynamic mapping logic module and is configured to construct a multi-dimensional phase space based on the CPU utilization, memory usage, network throughput and topological coupling, and describe the current operating state of the system as a probability distribution function within the multi-dimensional phase space. The entropy quantification evaluation submodule is connected to the phase space construction submodule and is configured to calculate the local probability density of the load distribution in the multidimensional phase space for each node based on the principle of statistical thermodynamics. The local probability density is multiplied by its own natural logarithm, and the product results generated by all nodes in the cluster are accumulated and the opposite number is taken to obtain the information entropy value representing the overall disorder of the system.

5. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The negative entropy flow scheduling decision unit includes a migration path simulation module, a negative entropy effect evaluation module, a scheduling scheme optimization module, and a global reorganization triggering module. The migration path simulation module is connected to the information entropy calculation unit and is configured to identify nodes whose current load is higher than the average level and mark them as high-temperature areas. At the same time, it identifies nodes whose load is lower than a preset threshold and have a topological connection with the high-temperature areas and marks them as low-temperature areas, generating multiple candidate sets of task migration paths to transfer specific microservice tasks from high-temperature areas to low-temperature areas. The negative entropy effect evaluation module is connected to the migration path simulation module. It is configured to predict the change in system information entropy after the implementation of each task migration path, and select migration paths that make the predicted change negative as negative entropy flow paths. The scheduling scheme optimization module is connected to the negative entropy effect evaluation module and is configured to select the optimal path from the negative entropy flow path that maximizes the decrease in information entropy and satisfies topological constraints and migration cost constraints, and generate resource reallocation instructions. The global reorganization trigger module is configured to initiate a global resource reorganization mechanism when the system information entropy continuously exceeds a preset security threshold and local migration fails. This mechanism temporarily freezes resource requests for non-core services and forces resource reclamation and reallocation across physical partitions.

6. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The elastic scaling execution unit includes a resource pre-inspection subunit, a container orchestration control subunit, and an execution feedback monitoring subunit; The resource pre-inspection subunit is configured to verify, upon receiving a resource reallocation instruction, whether the available resource balance of the target node is greater than the minimum resource threshold required for the service to be migrated; the container orchestration control subunit is connected to the resource pre-inspection subunit, and integrates a communication interface with the underlying container cluster management system. It is configured to execute the offline process of the original node instance, the pull process of the target node image, and the startup process of the new instance after passing the resource pre-inspection. The execution feedback monitoring subunit is configured to track the image pull time, instance readiness time, and request takeover success rate in real time during the migration process, and to feed back the actual resource distribution to the microservice cluster monitoring unit after the migration is completed, forming a closed-loop control loop to correct the prediction model of the information entropy calculation unit.

7. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The microservice cluster monitoring unit also includes an edge preprocessing proxy layer, which is configured to perform preliminary aggregation and feature compression of high-frequency sampled data of microservices deployed on the edge nodes of the industrial site, and upload key feature vectors representing load trends to the information entropy calculation unit. The topology modeling unit is configured to have multi-level topology resolution capabilities to obtain physical network topology information, including switch cascading levels, fiber optic link delays, and subnet isolation status. The dynamic graph construction sub-unit configuration is used to use the remaining bandwidth of the physical link as a weight correction factor, and to construct a multi-dimensional constrained composite topology graph by overlapping and mapping the logical call topology with the physical network topology, so as to avoid physical network congestion points.

8. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The information entropy calculation unit is configured to use a hierarchical entropy value calculation model to divide the microservice cluster into several logical availability zones, and to run a local information entropy calculation submodule within each logical availability zone. The local information entropy calculation submodule is used to calculate the local load balancing degree within the zone. The information entropy calculation unit is configured to summarize the local entropy values ​​of each logical availability zone and calculate the global system information entropy in combination with the cross-zone traffic distribution, guiding the negative entropy flow in the order of local first and then global. The negative entropy flow scheduling decision unit also includes a predictive heat conduction simulation module. The predictive heat conduction simulation module, combined with a service traffic prediction model based on a long short-term memory network, is configured to identify potential high-temperature area nodes in advance based on service traffic forecasts within a preset time period, and to pre-calculate preventive negative entropy injection paths.

9. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The microservice cluster monitoring unit also includes a hardware health monitoring module, which is configured to generate a node health score by monitoring changes in node power consumption, fan speed fluctuations, processor core temperature, and disk addressing error rate. The health score is input to the information entropy calculation unit as an energy conversion efficiency parameter. When the hardware indicators of a node are abnormal, the weight of that node in the resource scheduling model is reduced, so that it appears as a high-resistance region in the information entropy model, thereby guiding the task flow to a node with higher health. The topology modeling unit is also configured to have risk propagation path analysis function. It calculates the entropy increase diffusion coefficient caused by the overload of core service nodes through the service affinity assessment subunit, identifies the critical path with the greatest impact on stability in the cluster, and marks the vulnerability characteristics of the critical path in the dynamic topology map.

10. The microservice architecture elastic scaling and resource scheduling system for the industrial internet platform according to claim 1, characterized in that, The information entropy calculation unit integrates a dedicated computing acceleration module, which is built on a programmable gate array or a graphics processing unit. It is configured to process real-time load data streams from cluster nodes in parallel and implements a streaming computing architecture at the hardware level to perform logarithmic operations, multiply-accumulate operations, and multidimensional matrix transformations, thereby improving the update frequency of the global entropy value. The negative entropy flow scheduling decision unit is configured as a two-layer decision structure, including fast response logic embedded in the hardware acceleration card and main control logic running on the management node. The fast response logic is configured to complete the dynamic rearrangement of the task chain inside the node within microseconds. The elastic scaling execution unit supports a zero-copy migration mechanism, and the container orchestration control subunit is configured to migrate microservice instances between nodes that support shared storage or memory mirroring technology by mapping the running state of the original instance.

Images