Large-model-driven IT system integrated resource intelligent configuration and cooperative scheduling method

Abstract

The invention relates to the technical field of IT system integrated resource scheduling, and discloses a large-model-driven IT system integrated resource intelligent configuration and cooperative scheduling method, which comprises the following steps of: acquiring a logic coupling vector of each execution node in a micro-service architecture; mapping the logic coupling vector to a high-dimensional design space by using a large model, and generating a design constraint tensor representing a resource configuration state; calculating a logic pressure gradient between the nodes according to the calculation density parameter of each execution node, and monitoring an evolution trend of the logic pressure relative to a resource bearing threshold value so as to identify a resource conflict region; and triggering a topology reconstruction operator to generate a scheduling plan, synchronizing a physical state through a resource allocation interface and an execution node, and realizing resource increment compensation, by constructing a logic stress model, stabilizing space-time fluctuation between micro-service components, avoiding a cascade effect generated by resource competition, and guaranteeing the certainty of a system integration architecture.

Description

Technical Field

[0001] This invention relates to a large-model-driven method for intelligent configuration and collaborative scheduling of IT system integration resources, belonging to the field of IT system integration resource scheduling technology. Background Technology

[0002] With the increasing adoption of microservice architecture in system integration, improving development flexibility through distributed components has become an industry consensus. The rational allocation of computing and storage resources is fundamental to ensuring system stability. Existing resource configuration logic often relies on pre-defined rules set by architects or linear predictions based on historical data, treating complex integration architectures as static functional stacking and ignoring the spatiotemporal fluctuations caused by logical coupling between microservice components. This leads to a non-linear cascading amplification effect in resource contention between components during system integration. Monitoring and scheduling methods based on fixed thresholds are essentially post-hoc remedies and cannot prevent performance instability caused by logical topology conflicts at the system design stage.

[0003] While exploring the physical constraints of resource pooling management, even with the introduction of intelligent methods at the software control logic level, current solutions mostly focus on post-system anomaly remediation, lacking a fundamental understanding of the initial logical stress state during integration. For example, Chinese invention patent CN120610872B discloses a microservice fault diagnosis and self-healing system based on a large model. It utilizes a large model to analyze multimodal data and achieves root cause localization and automated repair through a causal inference engine. However, the core logic of such solutions remains anchored in feature capture and compensation strategies after a fault occurs. Their data collection relies on existing logs and communication segments, essentially treating microservices as isolated execution carriers. This observation-diagnosis-self-healing model... The lack of ability to predict deep-seated conflicts in the logical topology during system integration makes it impossible to analyze the instantaneous pressure and uneven stress caused by logical coupling between components. Once the system faces nonlinear cascading effects caused by ultra-high concurrency, relying solely on post-event self-healing often fails to maintain global steady state due to response time delays, and it is impossible to eliminate potential bottleneck nodes at the design level. Although the industry has tried to introduce dynamic sampling and elastic expansion mechanisms, due to the lack of understanding of logical topology semantics, such solutions can only perform passive adjustments after resource conflicts occur. They cannot identify and eliminate potential bottleneck nodes in advance in the logical design space of system construction. Furthermore, the lack of a global optimal judgment criterion for the coordinated scheduling of multi-dimensional resources makes it impossible for the system to maintain global dynamic steady state when facing extremely high performance pressure.

[0004] Therefore, the technical problem to be solved by this invention is how to achieve global dynamic optimization of integrated resource configuration and collaborative scheduling by parsing the communication links and interaction parameters of microservice clusters. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of this invention is as follows: A large-model-driven method for intelligent configuration and collaborative scheduling of IT system integration resources, comprising the following steps: Step S101: Collect the logical coupling vector of each microservice execution node in the microservice architecture. The logical coupling vector includes the communication frequency and the average message length. Step S102: Use the large model to perform semantic decoupling processing on the logical coupling vector, map the logical coupling vector to a high-dimensional design space map, and generate a design constraint tensor representing the resource allocation state. Step S103: Use the large model to extract the computation density parameters of each microservice execution node under full load conditions, and calculate the logical pressure gradient between adjacent nodes in the spatial manifold formed by the computation density parameters. Step S104: The large model performs tensor field equilibrium solution based on logical pressure gradient, monitors the deviation trend of the total logical pressure of multiple microservice execution nodes in the same time slice relative to the preset resource carrying threshold, and locks the potential resource conflict area according to the deviation evolution trend. Step S105: When the deviation evolution trend meets the preset early warning conditions, the topology reconstruction operator is triggered to generate a resource allocation scheme and a collaborative scheduling plan, and the collaborative scheduling plan is sent to the resource allocation execution interface. In step S106, the resource allocation execution interface synchronizes the physical state of the microservice execution node with the collaborative scheduling instructions generated by the collaborative scheduling plan, thereby realizing dynamic incremental compensation of computing resources.

[0006] Preferably, the calculation rule for the density parameter in step S103 is as follows: ,in, The computational density parameters for each microservice execution node under full load conditions; The communication frequency in the logical coupling vector, in units of ; This represents the average length of messages in the logical coupling vector. The initial computing resource weights are preset for microservice execution nodes; the logical pressure gradient is based on the computing density parameters between adjacent microservice execution nodes. The rate of change in three-dimensional space is determined.

[0007] Preferably, in step S101, during the process of collecting the logical coupling vector of each microservice execution node in the microservice architecture, the full resource data under the microservice architecture is obtained through a multi-interface collection method. The full resource data includes computing resource data, storage resource data, network resource data, and logical dependency relationship data between each microservice execution node.

[0008] Preferably, in the process of solving the tensor field equilibrium in step S104, the dynamic evolution trajectory of the design constraint tensor under different time slices is calculated using a large model, and the nonlinear pressure fluctuations caused by the logical coupling between microservice execution nodes are identified.

[0009] Preferably, in the process of generating the resource configuration scheme in step S105, the logical pressure gradient is smoothed by adjusting the logical topology path between the execution nodes of each microservice in the microservice architecture based on the resource gap determined by the tensor field equilibrium solution.

[0010] Preferably, in step S106, during the process of implementing dynamic incremental compensation of computing resources, the resource allocation execution interface calls the computing resource pool through the underlying resource scheduling layer and adjusts the allocation quota of the central processing unit of the target execution carrier according to the collaborative scheduling instructions.

[0011] Preferably, the large model-driven intelligent configuration and collaborative scheduling method for IT system integration resources further includes: real-time monitoring of the running status after the resource allocation execution interface allocates resources, and obtaining real-time feedback data corresponding to the running status; the large model dynamically corrects the design constraint tensor based on the real-time feedback data.

[0012] Preferably, in step S104, during the process of monitoring the deviation trend of the total logical pressure of multiple microservice execution nodes within the same time slice relative to the preset resource carrying capacity threshold, the large model identifies the cumulative rate of the total logical pressure; when the cumulative rate exceeds the preset growth threshold, it is determined that the microservice architecture has a risk of non-linear collapse.

[0013] Preferably, in step S105, during the process of generating the resource configuration scheme, the large model takes minimizing the integration cycle as the objective function and generates a globally optimal configuration list that includes computing resource quotas, storage space allocation, and network bandwidth limits through iterative optimization.

[0014] Preferably, the large model-driven intelligent configuration and collaborative scheduling method for IT system integration resources also includes: updating the high-dimensional design space map based on the execution results of the topology reconstruction operator to achieve predictive architecture alignment between the design and operation phases in the integration process.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the intelligent configuration of IT system integration resources, by parsing the interface definition files and communication links of the microservice cluster, the interaction parameters between components are transformed into logical coupling vectors between nodes, thereby constructing a logical design space map of the system to be integrated. This mechanism reconstructs the traditional resource configuration mode based on static snapshots into an auxiliary design process for complex logical topology systems. By mapping the logical dependencies between microservices into physical vectors in a high-dimensional space, it avoids the defect of the design phase and the operation phase being disconnected in the existing technology, enabling the system to identify the spatiotemporal fluctuations caused by logical coupling in the early stages of integration, thereby ensuring the determinism of resource configuration during large-scale IT system integration.

[0016] 2. By leveraging a large model to perform semantic decoupling on logically coupled vectors, a design constraint tensor representing the resource configuration state is generated. This mechanism no longer relies on architects' pre-defined rules or linear predictions based on historical data. Instead, it identifies potential resource conflict areas by calculating the logical pressure gradient between adjacent nodes. When the system determines that multiple nodes have overlapping resource occupancy within the same time slice, it triggers a topology reconfiguration operator to achieve the equilibrium solution of the tensor field. This transforms the complex operations research optimization problem into a force balance problem similar to that in physical structure design. From a principle perspective, it transforms IT resource configuration from an experience black box into a transparent logical force model, achieving predictive architecture alignment capabilities.

[0017] 3. Relying on the real-time interaction between the execution module and the underlying virtual machine monitor, and based on the collaborative scheduling instructions generated according to the tensor field equilibrium state, real-time hot mounting of computing resources can be achieved. By simulating the squeezing effect of logical traffic on resources at the design level, nonlinear collapse or avalanche effects caused by microservice cascading reactions are avoided. This global dynamic steady-state guarantee mechanism enables the resource scheduling of IT systems to leap from local optimization to overall optimization based on design logic, shortening the system integration cycle, and improving the collaborative utilization rate between microservice components and the overall system performance while reducing the resource configuration conflict rate. Attached Figure Description

[0018] Figure 1 This is a flowchart of the resource scheduling method for balancing logically coupled vectors and tensor fields according to the present invention. Figure 2 This is a schematic diagram of the system architecture of the present invention, which integrates a high-dimensional design space and a state feedback closed loop. Detailed Implementation

[0019] The method claimed in this invention will be described in detail below with reference to the accompanying drawings. It should be noted that the following embodiments are intended to explain and illustrate this invention, and not to limit the scope of protection of this invention.

[0020] This invention provides a large-model-driven intelligent configuration and collaborative scheduling method for IT system integration resources. It comprises a data acquisition module, a logic analysis module, a decision scheduling module, and an execution feedback module working collaboratively. By parsing the interface definition files and communication links of the microservice cluster, it transforms the interaction parameters between components into physical vectors in a high-dimensional design space graph. Using a large model, it generates design constraint tensors representing the resource configuration state. By calculating the logic pressure gradient and monitoring the evolution trend of logic pressure relative to the resource carrying capacity threshold, it identifies resource conflict areas and ultimately triggers topology reconfiguration and sends resource adjustment signals to the underlying virtual machine monitor, achieving real-time hot mounting of computing resources. Addressing the technical problem that the highly dynamic component interactions in microservice architecture cause traditional static configuration methods to fail, the method claimed in this invention adopts... This dataset contains the logical coupling vectors of each microservice execution node in a microservice architecture. These vectors include the communication frequency N and the average message length L. Using a service mesh edge proxy component deployed in the microservice cluster, the number of interface calls per unit sampling time is extracted in real time as the communication frequency N. Simultaneously, the message length generated by each call is calculated and its arithmetic mean is taken as the average message length L. To ensure the accuracy of data collection, the sampling period is set to 10ms to 100ms. For example, for the order processing microservice node, the number of calls received per second is collected as 500, i.e., the communication frequency N is 500Hz. At the same time, the average message data of each call is calculated to be 1024 bytes, i.e., the average message length L is 1024 bytes. This type of quantified vector input provides a data foundation for constructing a logical design space map.

[0021] Because logical dependencies between microservices can lead to non-linear performance degradation under complex operating conditions, a large model is used to perform semantic decoupling on logical coupling vectors. This maps the logical coupling vectors to a high-dimensional design space graph, generating design constraint tensors representing resource allocation states. In this process, the pre-built large model uses an attention mechanism to analyze the strength of logical connections between execution nodes, defining each execution node as a design node in the high-dimensional space. The logical coupling vectors are converted into logical coupling directions and magnitudes between nodes, outputting a multi-dimensional tensor matrix. The elements of this tensor matrix represent the resource interaction pressure between nodes at a specific time slice, transforming abstract logical traffic into a computable physical tensor model. To address the risk of insufficient disclosure, the generation procedure for the design constraint tensor includes: extracting communication frequencies using the hidden layer of the large model. With average message length The high-dimensional feature mapping is used to calculate the cosine similarity matrix between each microservice execution node by executing a multi-head attention operator. The cosine similarity matrix is ​​then multiplied element-wise with the feature magnitude matrix composed of logical coupling vectors to output the specific component values ​​of the design constraint tensor. Semantic decoupling processing is performed using a large model to generate the design constraint tensor: business logic tags are extracted from the interface definition files of the microservice execution nodes, and these tags are then compared with the communication frequencies in the logical coupling vectors. and average message length Semantic feature sequences are concatenated and input into the embedding layer of a large model to generate high-dimensional feature vectors. The multi-head attention operator in the hidden layer of the large model calculates the logical association strength matrix between nodes. The logical association strength matrix and the feature magnitude of the logical coupling vector feature are multiplied element-wise to output the specific component values ​​of the design constraint tensor. The elements of the design constraint tensor represent the pressure components of resource interaction between nodes under a specific time slice. The business communication semantics are converted into a high-dimensional computable physical tensor model to achieve nonlinear pressure fluctuation quantitative identification.

[0022] In traditional resource monitoring, fixed threshold triggering mechanisms are insufficient to predict instantaneous resource collapse caused by logical cascading reactions. This paper utilizes a large model to extract the computational density parameter D of each microservice execution node under full load conditions, and calculates the logical pressure gradient between adjacent nodes within the spatial manifold formed by the computational density parameter D. The formula for calculating the computational density parameter D is as follows: ,in, This refers to the computational density parameters for microservice execution nodes under full load conditions. This refers to the communication frequency, measured in Hz. Here, C represents the average message length in bytes, and C is the preset initial computational resource weight for the microservice execution node. For example, if the initial computational resource weight C of a certain execution node is set to 0.8, and the communication frequency N is 1000Hz and the average message length L is 2048 bytes, the computational density parameter D of this node is 2560000. The initial computational resource weight C is determined based on the idle baseline response latency calibration of the microservice execution node. During the sampling period, 1000 stateless baseline requests are sent to the microservice execution node, and the average response latency is recorded. (unit According to the formula Calculate the initial computational resource weights According to the formula Calculate the computational density parameters for each microservice execution node under full load conditions. , To calculate the density parameter, Communication frequency (unit) ), Average message length (in bytes). To initially calculate resource weights, calculate density parameters. This characterizes the unit hardware load pressure of the logical coupling vector on the microservice execution node, establishes an objective and quantitative correlation between logical interaction parameters and underlying physical resource requirements, and calculates the computational density parameters between adjacent microservice execution nodes. The rate of change in three-dimensional space determines the logical pressure gradient, which reflects the transmission rate of resource demand in the topological network.

[0023] To address the risk of system-wide paralysis caused by microservice avalanche effects, a large-scale model is used to solve the tensor field equilibrium problem based on logical pressure gradients. This model monitors the deviation trend of the sum of logical pressures from multiple microservice execution nodes within the same time slice relative to a preset resource carrying capacity threshold. Based on this deviation trend, potential resource conflict areas are identified. During this process, the large-scale model calculates the dynamic evolution trajectory of the design constraint tensor under different time slices, identifies nonlinear pressure fluctuations caused by logical coupling, and monitors the cumulative rate of the sum of logical pressures, i.e., the change in pressure deviation per unit time. When the cumulative rate exceeds a preset growth threshold, the microservice architecture is deemed to have a nonlinear collapse risk. The specific sampling strategy and risk judgment logic are set as follows: the system maintains 1... A sliding window buffer with a length of 20 sampling points is used, with a fixed sampling frequency of 100Hz, meaning that logical pressure data is collected once every 10ms. The cumulative rate is calculated by taking the arithmetic mean of the logical pressure of the last 10 sampling points in the current sliding window and subtracting the arithmetic mean of the logical pressure of the first 10 sampling points. If the difference exceeds the growth threshold of 15 units for three consecutive sampling periods, the system determines that there is a risk of nonlinear collapse in the current area and issues a scheduling warning. For example, if the resource carrying capacity threshold is set to 100 units, when it is detected that the total logical pressure of a certain microservice cluster increases from 60 units to 95 units within 10ms and the slope of the trend line continues to increase, the large model marks the area as a high-risk conflict zone, providing an early warning for preventive scheduling.

[0024] To ensure the determinism of IT system integration resource scheduling, in the method claimed in this invention, when the deviation evolution trend meets the preset early warning conditions, a topology reconfiguration operator is triggered to generate a resource configuration scheme and a collaborative scheduling plan, and the collaborative scheduling plan is sent to the resource allocation execution interface. The large model takes minimizing the integration cycle as the objective function, and generates a globally optimal configuration list including computing resource quotas, storage space allocation, and network bandwidth limits through iterative optimization. At the same time, based on the resource gap determined by the tensor field balance solution, the logical topology path weights between execution nodes in the microservice architecture are adjusted to smooth the logical pressure gradient. The collaborative scheduling instructions generated according to the collaborative scheduling plan are synchronized with the physical state of the microservice execution nodes to achieve dynamic incremental compensation of computing resources. The resource allocation execution interface calls the computing resource pool through the underlying resource scheduling layer, adjusts the allocation quota of the central processing unit of the target execution carrier, and sends control instructions to the virtualization management program to adjust the number of virtual cores of the target microservice instance from 4 cores to 8 cores, thus smoothing the logical pressure at the physical level. The topology reconfiguration operator is triggered to generate a resource configuration scheme and a collaborative scheduling plan: the resource gap is determined based on the tensor field equilibrium solution, the incremental quota of virtual cores of the target microservice execution node is calculated, and resource allocation is performed. The execution interface sends traffic weight adjustment instructions to the service mesh edge agent, calls the underlying resource management layer control interface to adjust the allocation share of the target carrier's central processing unit, uses the controller to modify the target process processor time slice allocation value, matches the actual physical resource allocation status with the design constraint tensor to define the logical pressure balance point, monitors the resource allocation execution interface to obtain real-time feedback data after resource allocation, dynamically corrects the design constraint tensor based on the real-time feedback data, calculates the actual resource utilization rate and the predicted pressure residual to adjust the internal semantic feature weights inversely, and updates the high-dimensional design space map based on the execution results of the topology reconstruction operator to achieve integrated flow. The system aligns predictive architecture between the design and operation phases. To ensure the continued effectiveness of resource allocation schemes, the system monitors the operational status of resources allocated by the resource allocation execution interface in real time, obtains real-time feedback data corresponding to the operational status, and dynamically corrects the design constraint tensor based on the real-time feedback data. By calculating the residual between the actual resource utilization rate and the predicted pressure, the semantic feature weights within the large model are adjusted in reverse. The high-dimensional design space map is updated based on the execution results of the topology reconstruction operator, thereby achieving predictive architecture alignment between the design and operation phases in the integration process and ensuring that the IT system maintains global dynamic steady state when facing high concurrency conditions.

[0025] Example 1: This example is a further refinement and application derivation of the above specific implementation methods. The technical terms, parameter names, and formula variables mentioned in this example are mirror images of the above specific implementation methods. In the operation of the large model-driven intelligent configuration and collaborative scheduling method for IT system integration resources provided by this invention, when the system faces pulse-like load conditions generated by high-frequency small-amount payment settlement business, logical coupling occurs between the order processing microservice node and the account settlement microservice node in the integration environment, resulting in a cascading amplification effect caused by resource contention in the local topology path. At this time, the objective obstacle faced by the system is that the lag in physical resource utilization cannot represent the instantaneous pressure at the logical level. For the above conditions, the service mesh edge proxy component deployed in the microservice cluster extracts the number of interface calls per unit sampling time in real time as the communication frequency. Simultaneously, the message length generated by each call is counted and the arithmetic mean is calculated to obtain the average message length. In the specific operational data of this embodiment, the communication frequency of the order processing microservice node is set. 2500Hz, average message length The initial computational resource weight is 512 bytes. Calculate the density parameter with a value of 1.0. The calculation formula is as follows: ,in, To calculate the density parameter, This refers to the communication frequency, measured in Hz. This represents the average message length in bytes. The initial resource weights are calculated; the computational density parameters of this node are then calculated using the formula described above. The value is 1,280,000. Based on this, the large model uses an attention mechanism to analyze the strength of logical connections between execution nodes and adjusts the communication frequency. With average message length The mapping is transformed into node interaction pressure in a high-dimensional design space graph. The generated design constraint tensor describing the resource configuration status transforms the originally imperceptible logical interaction load into a quantifiable topological stress model, identifying that the logical pressure gradient of the order processing microservice node relative to the adjacent settlement node exceeds a preset threshold.

[0026] Because this method changes the criterion for resource allocation from a single load utilization rate to a tensor field balance solution of the logical design space, the topology reconstruction operator intervenes before the physical resources actually collapse. By adjusting the logical topology path weights between execution nodes in the microservice architecture, it smooths out the logical pressure gradient. The decision scheduling module generates a collaborative scheduling plan containing resource quota adjustment instructions and sends it to the resource allocation execution interface. The resource allocation execution interface sends control signals to the underlying virtual machine monitor to adjust the virtual computing core quota of the target order processing microservice node from 8 cores to 16 cores. Through real-time hot mounting of computing resources, dynamic resource increment compensation is completed, reducing the real-time logical pressure of the order processing microservice node from 95 units to 60 units, and restoring the global logical pressure distribution of the system to the preset steady-state range.

[0027] Example 2: To verify the stability of the large-model-driven IT system integration resource intelligent configuration and collaborative scheduling method in dealing with cascading conflicts in a large-scale distributed architecture, an integration test platform consisting of 12 microservice execution nodes was established. The computing resource pool of the integration test platform adopts a configuration of dual-socket 32-core CPUs and 256GB synchronous dynamic random access memory per node. Data acquisition is achieved through probe components deployed on the service mesh edge agent. The sampling period is set to balance the real-time performance of data perception with the additional computational overhead generated by system monitoring, and is determined to be 50ms. In the test scheme, Gaussian white noise with a signal-to-noise ratio of 20dB is injected into the microservice interaction link during simulated communication conditions. The test is set up with three test groups, namely... The experiment included a control group using a fixed threshold triggering mechanism, a partially missing control group with the tensor field equilibrium solution removed, and an experimental group using the complete technical method of this invention. For payment transaction processing, the initial computational resource weight C was set to 1.0, the communication frequency N ranged from 100Hz to 5000Hz, and the average message length L ranged from 128 bytes to 2048 bytes. The experiment lasted 3600 seconds. By monitoring the evolution trend of the logical pressure of each execution node under varying load intensity, the method's ability to lock resource conflict areas was quantitatively analyzed. During the experimental run, the system collected the logical coupling vectors of each execution node, used a large model to map the flow parameters to a high-dimensional design space map, and calculated the computational density parameters characterizing the node features. And the logical pressure gradient formed by the pressure difference between adjacent nodes; see Table 1, which records the logical pressure gradient as a function of communication frequency under injected environmental disturbance noise. With average message length Gradient increases, density parameter calculation The corresponding evolution relationship with real-time logic pressure: Table #1: Key Intermediate Data Recordings for Test Groups Under Different Load Intensities

[0028] in, To calculate the density parameter, This refers to the communication frequency, measured in Hz. This represents the average message length in bytes. To initially calculate resource weights, calculate density parameters. Satisfy the formula The unit of real-time logic pressure is defined as a percentage relative to the hardware's carrying capacity; as can be seen from the data evolution trend in Table 1, when calculating the density parameter... When the preset critical threshold is exceeded, the real-time logic pressure exhibits a non-linear growth trend; in test sequence 4, the logic pressure reached 108.4 units, exceeding the preset 100-unit bearing threshold; the large model identifies stress concentration phenomena in high-dimensional manifolds through tensor field equilibrium solutions, locking the logical topology between the order processing node and the settlement node as a conflict region; by comparing the performance indicators of different groups, the synergistic effect of this method in resolving resource conflicts is confirmed; under the same pulsed load conditions, the control group using a fixed threshold triggering mechanism triggers adjustments after a cascading response occurs when the physical resource utilization rate reaches 95%, and the system integration cycle... The volatility exceeded 45%, while the experimental group triggered the topology reconfiguration operator when the logic pressure reached the warning condition. By adjusting the logical topology path weights, incremental compensation of resources was achieved. The resource configuration conflict rate of the experimental group was reduced by 72.5% compared with the control group, and the integration cycle was shortened by 31.8%. When the logic pressure exceeded 110% of the hardware carrying capacity threshold, the throughput growth curve of the system output tended to flatten and entered the saturation region. This data supports the rationality of the resource carrying capacity threshold range setting and proves that this method reconfigures IT resource scheduling into predictive configuration based on logical stress analysis, enabling the integration architecture to maintain a global dynamic balance under extreme conditions.

[0029] Example 3: This example combines Figures 1 to 2 This section describes a method for intelligent resource allocation and collaborative scheduling in IT system integration driven by a large model, such as... Figure 1As shown, step S101 involves collecting the logical coupling vectors of each microservice execution node in the microservice architecture. These logical coupling vectors include communication frequency and average message length. Step S102 then uses a large model to perform semantic decoupling processing on the logical coupling vectors, mapping them to a high-dimensional design space graph and generating a design constraint tensor representing the resource configuration state. Next, in step S103, the large model extracts the computational density parameters of each microservice execution node under full load conditions and calculates the logical pressure gradient between adjacent nodes in the spatial manifold formed by the computational density parameters. Finally, step S104 involves the large model calculating the logical pressure gradient based on the logical pressure gradient. The force gradient is used to solve the tensor field equilibrium problem. The deviation trend of the total logical pressure of multiple microservice execution nodes in the same time slice relative to the preset resource carrying threshold is monitored, and potential resource conflict areas are identified accordingly. When the deviation trend meets the preset warning conditions, the process enters step S105 to trigger the topology reconstruction operator to generate a resource configuration scheme and a collaborative scheduling plan. The collaborative scheduling plan is then sent to the resource allocation execution interface. Finally, step S106 is executed, where the collaborative scheduling instructions generated by the resource allocation execution interface according to the collaborative scheduling plan are synchronized with the physical state of the microservice execution nodes, thereby realizing dynamic incremental compensation of computing resources.

[0030] like Figure 2 As shown, the edge proxy component located in the microservice runtime environment collects interface call frequency and message length data, and transmits it to the logical coupling vector acquisition and quantization module to generate a logical coupling vector. This vector is then input to the semantic decoupling and tensor generation module, which receives mapping parameters from the A1 high-dimensional design space graph and semantic feature weight correction data from the state feedback and model dynamic correction module. After processing, it outputs a design constraint tensor. The logical pressure gradient calculation module performs calculations based on the design constraint tensor, and transmits the calculated density parameters and logical pressure gradient to the tensor field equilibrium solution and conflict identification module. The threshold parameters from the A2 resource carrying capacity threshold library are analyzed, and the resource conflict warning and deviation trend are output to the topology reconstruction and scheduling plan generation module. After receiving the signal, the topology reconstruction and scheduling plan generation module performs multi-channel output, sending update map signals to the A1 high-dimensional design space map, sending collaborative scheduling plan and topology reconstruction instructions to the resource allocation instruction issuing module, and generating physical resource adjustment instructions that directly act on the virtual machine monitor and the underlying resource layer. The real-time running status feedback data generated by the underlying resource layer is finally sent back to the status feedback and model dynamic correction module to complete the dynamic correction closed loop of the design constraint tensor.

[0031] Example 4: In a distributed integration architecture comprising 50 microservice components, including order management, inventory synchronization, payment settlement, and logistics tracking, the system faces link blockage obstacles caused by uneven resource pre-allocation during the initial integration phase. The logical coupling relationships between execution nodes are complex, and monitoring methods cannot provide identification criteria for deep-seated logical topology conflicts. The system determines the initial computational resource weights through performance stress testing. In an unloaded environment, 1000 standard stateless requests are sent to the target microservice execution node, and the average response latency is recorded. To ensure consistency in calibration, the specific boundary conditions of the experiment were set as follows: the payload of a single request was fixed at 64 bytes of JSON string; the number of concurrent threads in the load testing tool was set to 10; and the physical baseline of the load testing environment was uniformly set to a 2-core CPU and 4GB of memory. The system recorded the average response latency of 1000 requests in milliseconds, and calculated according to the formula... Determine the weights, where, To initially calculate resource weights, The average response time is expressed in milliseconds (ms); where the average response time is for the order management node. The initial calculation of resource weights takes 2ms. The value is set at 0.5.

[0032] The system collects the logical coupling vector of the order management node through the telemetry interface of the service mesh, at which time the communication frequency... for Hz, average message length The model is 1024 bytes long; the large model uses an attention mechanism to parse logical coupling vectors, calculates the correlation coefficient matrix between nodes to generate design constraint tensors, maps flow data to node force vectors in a high-dimensional design space map, and calculates density parameters. The calculation formula is as follows: ,in, This refers to the computational density parameters for microservice execution nodes under full load conditions. This refers to the communication frequency, measured in Hz. This represents the average message length in bytes. To initially calculate resource weights, substitute the above calibration data into the formula to obtain the computational density parameters of the order management node. The value is 6,144,000. The large model determines the logical pressure gradient by calculating the ratio of the difference in computational density between the node and its downstream inventory synchronization nodes to the topological distance. When the deviation of the total logical pressure of the order management node from the preset resource carrying capacity threshold shows a monotonically increasing trend and the cumulative rate exceeds 15 units per sampling period, the large model triggers the topology reconstruction operator to perform tensor field balance solution, and iteratively reduces the topological path weight between high-pressure nodes to smooth the logical pressure gradient. The resource allocation execution interface receives the collaborative scheduling instruction, adjusts the central processing unit quota of the order management node from the current value to 20%, and synchronously updates the node position in the high-dimensional design space map until the residual between the actual resource utilization rate and the predicted logical pressure converges to within 5%. The system reduces the request drop rate of the integrated architecture from 12.5% ​​to 0.8% during load peaks through a predictive configuration method based on physical scaling and tensor evolution.

[0033] Example 5: In a cloud-native integrated environment that includes microservice components for order management, payment settlement, and inventory synchronization, the interaction weights between execution nodes are uncalibrated during system topology initialization. The system selects synchronous calls, asynchronous messages, and distributed transaction interaction templates to simulate communication frequencies under a controlled environment. With average message length The load flow; the attention mechanism layer of the large model calculates the weight distribution of the load flow on the topology path, and establishes a logical coupling vector relative to the computational density parameter in the high-dimensional design space. The physical mapping benchmark is used to convert the interaction mode into the benchmark node position in the high-dimensional design space map and determine the initial mapping coefficient of the logical flow pressure under different communication protocols.

[0034] When the system faces cascading load conditions triggered by high-concurrency payment transactions, the system implements quantitative calibration of resource carrying capacity thresholds during the debugging phase; the system gradually increases the communication frequency of microservice execution nodes within the test cluster. Monitor the computation density parameters of the microservice execution node. The evolution trend of response latency; when the response latency reaches the preset service quality protocol upper limit, the logical pressure value output by the large model is recorded and used as the resource carrying threshold of the microservice execution node under full load conditions; the large model performs tensor field balance solution based on the acquired real-time feedback data and resource carrying threshold, and locks the logical topology conflict area before the physical load accumulates, so that the integrated architecture maintains a global dynamic balance state when facing instantaneous peak impact.

[0035] Example 6: In a distributed integration architecture that handles heterogeneous interaction protocols, the order management microservice and the payment settlement microservice interact through a mutual exclusion protocol. The system executes a pre-deployment semantic consistency calibration procedure and uses the service mesh edge proxy to extract the average communication frequency of the synchronous call link. 1000Hz, average message length Set the initial computational resource weight to 1024 bytes. Set the value to 1.0 to determine the synchronous semantic transformation coefficient. Given a value of 1.25, calculate the density parameter. Satisfy the physical mapping formula ;in, To calculate the density parameter, This refers to the communication frequency, measured in Hz. This represents the average message length in bytes. These are semantic transformation coefficients. To initially calculate resource weights, the system uses calibrated semantic transformation coefficients to normalize heterogeneous traffic loads into node force vectors in a high-dimensional design space graph, and determines the mapping coefficients of logical traffic pressure under different communication protocols.

[0036] When the integrated system detects that the cumulative rate of deviation of the logical pressure from the resource carrying capacity threshold reaches the critical point of 10 units per sampling period, the large model triggers the topology reconstruction operator to implement the path weight correction process. This process adjusts the logical topology weights between adjacent microservice execution nodes through iterative calculations, resulting in updated path weights. Compared with the path weight before the update Logic pressure deviation and resource carrying capacity threshold They satisfy a linear correction relationship ;in, For the updated path weights, The path weights before the update. This represents the deviation from the logical pressure. The resource carrying capacity threshold is set; the topology reconstruction operator sends a collaborative scheduling plan containing resource hot-mount quota to the resource allocation execution interface, adjusting the number of virtual cores of the settlement node from 8 cores to 12 cores, reducing the resource response latency of the integrated architecture under load pulse conditions from 150ms to 45ms, and the global resource configuration state converges to the preset steady-state range.

[0037] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A large-model-driven method for intelligent allocation and collaborative scheduling of IT system integration resources, characterized in that, Includes the following steps: Step S101: Collect the logical coupling vector of each microservice execution node in the microservice architecture. The logical coupling vector includes the communication frequency and the average message length. Step S102: Use the large model to perform semantic decoupling processing on the logical coupling vector, map the logical coupling vector to a high-dimensional design space map, and generate a design constraint tensor representing the resource allocation state; Step S103: Use the large model to extract the computation density parameters of each microservice execution node under full load conditions, and calculate the logical pressure gradient between adjacent nodes in the spatial manifold formed by the computation density parameters. Step S104: The large model performs tensor field equilibrium solution based on logical pressure gradient, monitors the deviation trend of the total logical pressure of multiple microservice execution nodes in the same time slice relative to the preset resource carrying threshold, and locks the potential resource conflict area according to the deviation evolution trend. Step S105: When the deviation evolution trend meets the preset early warning conditions, the topology reconstruction operator is triggered to generate a resource allocation scheme and a collaborative scheduling plan, and the collaborative scheduling plan is sent to the resource allocation execution interface. In step S106, the resource allocation execution interface synchronizes the physical state of the microservice execution node with the collaborative scheduling instructions generated by the collaborative scheduling plan, thereby realizing dynamic incremental compensation of computing resources.

2. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model as described in claim 1, characterized in that, The calculation rules for the density parameter in step S103 are as follows: ,in, The computational density parameters for each microservice execution node under full load conditions; The communication frequency in the logical coupling vector, in units of ; This represents the average length of messages in the logical coupling vector. The initial computing resource weights are preset for microservice execution nodes; the logical pressure gradient is based on the computing density parameters between adjacent microservice execution nodes. The rate of change in three-dimensional space is determined.

3. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model as described in claim 1, characterized in that, In step S101, during the process of collecting the logical coupling vector of each microservice execution node in the microservice architecture, the full resource data under the microservice architecture is obtained through a multi-interface collection method. The full resource data includes computing resource data, storage resource data, network resource data, and logical dependency relationship data between each microservice execution node.

4. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model as described in claim 1, characterized in that, In step S104, during the tensor field equilibrium solution process, the dynamic evolution trajectory of the design constraint tensor under different time slices is calculated using a large model, and the nonlinear pressure fluctuations caused by the logical coupling between microservice execution nodes are identified.

5. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model as described in claim 1, characterized in that, In step S105, during the process of generating the resource configuration scheme, the logical pressure gradient is smoothed by adjusting the logical topology path between the execution nodes of each microservice in the microservice architecture based on the resource gap determined by the tensor field equilibrium solution.

6. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model according to claim 1, characterized in that, In step S106, during the process of implementing dynamic incremental compensation of computing resources, the resource allocation execution interface calls the computing resource pool through the underlying resource scheduling layer and adjusts the allocation quota of the central processing unit of the target execution carrier according to the collaborative scheduling instructions.

7. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model according to claim 1, characterized in that, The large-model-driven intelligent configuration and collaborative scheduling method for IT system integration resources also includes: real-time monitoring of the running status after the resource allocation execution interface allocates resources, and obtaining real-time feedback data corresponding to the running status; the large model dynamically corrects the design constraint tensor based on the real-time feedback data.

8. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model according to claim 1, characterized in that, In step S104, during the process of monitoring the deviation trend of the total logical pressure of multiple microservice execution nodes within the same time slice from the preset resource carrying threshold, the large model identifies the cumulative rate of the total logical pressure; when the cumulative rate exceeds the preset growth threshold, it is determined that the microservice architecture has a risk of non-linear collapse.

9. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model according to claim 1, characterized in that, In step S105, during the generation of the resource configuration scheme, the large model takes minimizing the integration cycle as the objective function and generates a globally optimal configuration list that includes computing resource quotas, storage space allocation, and network bandwidth limits through iterative optimization.

10. The method for intelligent allocation and collaborative scheduling of IT system integration resources driven by a large model according to claim 1, characterized in that, The method also includes updating the high-dimensional design space map based on the execution results of the topology reconstruction operator.

Images