Dynamic deployment method of operating system, electronic equipment and computer program product

Abstract

The invention relates to the technical field of computers, in particular to an operating system dynamic deployment method, electronic equipment, a computer readable storage medium and a computer program product. Based on a dynamic deployment strategy, dynamic construction of an operating system mirror image and efficient allocation of resources are achieved through an atomization deployment engine perceived by non-uniform memory access, and cross-domain access delay is reduced through PCIe re-enumeration and memory fragmentation technologies; according to the drive on-demand dynamic injection technology, the mirror image size is simplified, the deployment efficiency and the resource utilization rate are improved, and an optimized deployment environment is provided for efficient operation of the system.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method for dynamically deploying an operating system, electronic devices, computer-readable storage media, and computer program products. Background Technology

[0002] Existing methods for dynamically deploying operating systems based on Baseboard Management Controller (BMC) technology lack intelligent decision-making capabilities and cannot dynamically match hardware characteristics with real-time business needs, resulting in low resource utilization. Summary of the Invention

[0003] This disclosure provides a method for dynamically deploying an operating system, an electronic device, a computer-readable storage medium, and a computer program product. Its main purpose is to address the problems of existing technologies, such as a lack of intelligent decision-making capabilities, inability to dynamically match hardware characteristics with real-time business needs, and low resource utilization.

[0004] According to a first aspect of this disclosure, a method for dynamically deploying an operating system is provided, comprising: Hardware fingerprints are generated by acquiring server hardware device data characteristics through BMC. Based on the hardware fingerprint and historical load data, an AI model is used to predict future load requirements, and a dynamic deployment strategy including hardware configuration information, operating system image version selection, driver dependency information and resource allocation strategy is generated based on the future load requirements. Based on the hardware configuration information, non-uniform memory access-aware memory sharding is performed on the server hardware device through BMC, and the PCIe bus topology is re-enumerated. Based on the operating system image version selection and driver dependency information, and according to the hardware fingerprint matching driver operating system image whitelist, the initial memory file system is reconstructed to generate a customized operating system image. The containerized deployment is combined with lightweight virtual machines for hybrid orchestration, and resources are allocated based on the resource allocation strategy.

[0005] Through this disclosure, the non-uniform memory access-aware atomic deployment engine enables dynamic building of operating system images and efficient allocation of resources, reduces memory access latency, improves deployment efficiency and resource utilization, and provides an optimized deployment environment for subsequent secure and trusted chain construction.

[0006] Preferably, the step of obtaining server hardware device data features through BMC and generating a hardware fingerprint includes: The BMC interface driver obtains static data including CPU microcode version, PCIe bus topology and DRAM SPD information, and generates static features including CPU microcode hash value, PCIe bus topology diagram and DRAM SPD matrix. The BMC sensor interface acquires dynamic data including real-time sensor data and hardware health status, and generates dynamic features including sensor time series and memory health. Based on the static and dynamic features, a globally unique hardware fingerprint is generated.

[0007] Due to the identification conflicts in scenarios where hardware from multiple vendors is deployed in the prior art, the hardware fingerprint (UHI) generated by this implementation method has global uniqueness and high stability, providing a reliable hardware identification basis for subsequent steps and has licensing potential.

[0008] Preferably, the step of using an AI model to predict future load demands based on the hardware fingerprint and historical load data, and generating a dynamic deployment strategy based on the future load demands, including hardware configuration information, operating system image version selection, driver dependency information, and resource allocation strategies, includes: Build AI models, including a spatiotemporal prediction sub-model for predicting future load demand and a fault prediction sub-model for predicting hardware failure probability; Based on the hardware fingerprint and historical load data, the spatiotemporal prediction model is used to predict future load demand. Based on the dynamic data, the fault prediction sub-model is used to predict the probability of hardware failure. Based on the future load demand and the hardware failure probability, a state space is generated that includes a hardware fingerprint, predicts resource demand and hardware failure probability, and includes an action space containing discrete actions with operating system image version selection and continuous actions with CPU quota coefficient. The reward function is calculated based on actual resource utilization, deployment time and hardware failure probability.

[0009] This implementation method dynamically adjusts the deployment strategy based on the prediction results, providing precise strategic guidance for the next step of dynamic deployment.

[0010] Preferably, constructing the spatiotemporal prediction sub-model for predicting future load demand includes: A spatiotemporal graph is constructed, which includes hardware components as nodes and physical link latency and historical load correlation coefficients as edge weights. A spatiotemporal graph convolutional network is used to extract local spatiotemporal features, and a Transformer encoder is combined to capture global time dependencies, generating a spatiotemporal prediction sub-model for predicting future load demand.

[0011] According to this implementation method, the accuracy of load prediction is significantly improved by spatiotemporal joint prediction.

[0012] Preferably, constructing the fault prediction sub-model for predicting hardware fault probability includes: Based on long short-term memory networks and multi-granularity attention mechanisms, combined with hardware health indicators, the probability of hardware failure is predicted, and a fault prediction sub-model for predicting the probability of hardware failure is constructed.

[0013] According to this embodiment, the fault detection rate is improved by using a multi-granularity fault prediction mechanism.

[0014] Preferably, the step of combining containerized deployment with lightweight virtual machines for hybrid orchestration and allocating resources based on the resource allocation strategy further includes: During the BMC secure boot phase, the hash value of the customized operating system image is bound to the platform configuration register of the trusted platform module along with the dynamic deployment strategy and the verification logic is performed. If the logic verification passes, the BMC is forced to power down. The process hash is calculated in real time. When a process hash that does not belong to the hardware fingerprint is present, IPMI is triggered to force a power-down.

[0015] According to this implementation, the dynamic trusted chain bound by hardware fingerprint effectively blocks the loading of unauthorized firmware / operating system, and the runtime process hash verification can intercept malicious code injection in real time, enhancing the system's full-stack dynamic security protection capabilities and providing a safe and reliable operating environment for closed-loop optimization.

[0016] Preferably, the step of combining containerized deployment with lightweight virtual machines for hybrid orchestration and allocating resources based on the resource allocation strategy further includes: Local data is obtained from the execution log, and a loss function is calculated based on the actual resource utilization and future load requirements. The gradients of the feature layers that are strongly correlated with the hardware fingerprint are selected to generate the gradients of the local AI prediction model. The hardware performance score is calculated based on the deployment time and hardware failure probability, and the gradient weights of the local AI prediction model are adjusted based on the hardware performance score to obtain the globally optimized model parameters. The local AI prediction model is updated by pushing incremental packages including the globally optimized model parameters through the BMC out-of-band channel, and the dynamic deployment strategy is updated based on the BMC dual-bank flash switching.

[0017] This implementation plan significantly improves model convergence speed, reduces edge node resource over-allocation rate, and enhances resource utilization by adopting a differentiated federated learning aggregation strategy. It also achieves closed-loop optimization of the system and further improves system performance and security.

[0018] According to a second aspect of this disclosure, an operating system dynamic deployment apparatus is provided, comprising: The hardware fingerprint generation module is used to obtain server hardware device data characteristics through BMC and generate hardware fingerprints. The dynamic deployment strategy generation module is used to predict future load demands using an AI model based on the hardware fingerprint and historical load data, and generate a dynamic deployment strategy based on the future load demands, including hardware configuration information, operating system image version selection, driver dependency information and resource allocation strategy. The hardware configuration module is used to perform non-uniform memory access aware memory sharding on the server hardware device through BMC based on the hardware configuration information, and to re-enumerate the PCIe bus topology. The operating system image building module is used to reconstruct the initial memory file system and generate a customized operating system image based on the operating system image version selection and driver dependency information, according to the hardware fingerprint matching driver operating system image whitelist. The hybrid orchestration module is used to perform hybrid orchestration of containerized deployments with lightweight virtual machines and to allocate resources based on the resource allocation strategy.

[0019] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0020] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0021] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0023] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 A flowchart illustrating an operating system dynamic deployment method provided in an embodiment of this disclosure; Figure 2 A flowchart illustrating another method for dynamically deploying an operating system provided in this embodiment of the disclosure; Figure 3 A flowchart illustrating yet another method for dynamically deploying an operating system provided in this disclosure embodiment; Figure 4 This is a schematic diagram of the structure of an operating system dynamic deployment device provided in an embodiment of this disclosure. Detailed Implementation

[0024] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0025] The following description, with reference to the accompanying drawings, outlines a method, apparatus, electronic device, computer-readable storage medium, and computer program product for dynamically deploying an operating system according to embodiments of the present disclosure.

[0026] Figure 1 This is a flowchart illustrating an operating system dynamic deployment method provided in an embodiment of the present disclosure.

[0027] like Figure 1 As shown, the method includes the following steps: Step 101: Obtain server hardware device data characteristics through BMC and generate hardware fingerprint; Specifically, server hardware device data includes Baseboard Management Controller (BMC) sensor data, firmware configuration, hardware topology, and Trusted Platform Module (TPM) chip metrics.

[0028] Step 102: Based on the hardware fingerprint and historical load data, use an AI model to predict future load requirements, and generate a dynamic deployment strategy based on the future load requirements, including hardware configuration information, operating system image version selection, driver dependency information and resource allocation strategy. Step 103: Based on the hardware configuration information, perform non-uniform memory access aware memory sharding on the server hardware device through BMC, and re-enumerate the PCIe bus topology. Specifically, based on the hardware configuration information, the BMC control layer performs non-uniform memory access aware memory sharding on the server hardware device according to the Redfish API operation sequence, and sends an IPMI command to force a refresh of the PCIe bus topology.

[0029] Step 104: Based on the operating system image version selection and driver dependency information, and according to the hardware fingerprint matching driver operating system image whitelist, reconstruct the initial memory file system and generate a customized operating system image. Specifically, through the operating system deployment layer, based on the operating system image version selection and driver dependency information, and according to the hardware fingerprint matching driver operating system image whitelist, the initial memory file system is reconstructed to generate a customized operating system image.

[0030] Step 105: Perform hybrid orchestration of containerized deployment and lightweight virtual machines, and allocate resources based on the resource allocation strategy.

[0031] Specifically, resources are allocated based on predicted resource requirements by combining a software orchestration layer with containerized deployment and lightweight virtual machines (MicroVMs) in a hybrid orchestration manner.

[0032] In summary, the method provided by the embodiments of this disclosure generates a dynamic deployment strategy based on hardware fingerprints and future load requirements. Based on the dynamic deployment strategy, it achieves dynamic construction of the operating system image and efficient allocation of resources through an atomic deployment engine with non-uniform memory access awareness. Through PCIe re-enumeration and memory sharding technology, it reduces cross-domain access latency. The driver on-demand dynamic injection technology simplifies the image size, improves deployment efficiency and resource utilization, and provides an optimized deployment environment for the efficient operation of the system.

[0033] It should be noted that the embodiments of this disclosure may include multiple steps. For ease of description, these steps are numbered, but these numbers are not a limitation on the execution time slots or execution order between the steps; these steps can be implemented in any order, and the embodiments of this disclosure do not limit this.

[0034] In some embodiments, step 101 above may be implemented by the following steps: Step 1011: Obtain static data including CPU microcode version, PCIe bus topology and DRAM SPD information through the BMC interface driver, and generate static features including CPU microcode hash value, PCIe bus topology diagram and DRAM SPD matrix; Specifically, the CPU microcode version is obtained through IPMI commands and its SHA3-256 hash value is calculated and truncated to 128 bits; the Redfish API is called to obtain the PCIe bus topology and construct a weighted directed graph; DRAM SPD EEPROM data is read and encoded into a 16×16 matrix.

[0035] Step 1012: Acquire dynamic data including real-time sensor data and hardware health status through the BMC sensor interface, and generate dynamic features including sensor time series and memory health. The sensor data includes temperature, power consumption, and voltage jitter rate. The formula for calculating memory health is as follows:

[0036] ECC_Rate: A memory health metric that represents the relative frequency of memory errors within a specific time window.

[0037] Correctable_Errors: The number of correctable errors, referring to the number of errors detected and successfully corrected during memory operations.

[0038] Uncorrectable_Errors: The number of uncorrectable errors, referring to the number of errors detected during memory operations that cannot be corrected.

[0039] DRAM_Capacity_GB: Memory capacity (in GB), representing the total dynamic random access memory (DRAM) capacity of the system.

[0040] Time_Window_sec: The time window (in seconds) represents the time range considered when calculating memory health.

[0041] Step 1013: Generate a globally unique hardware fingerprint based on the static and dynamic features.

[0042] Specifically, a Graph Attention Network (GAT) is used to encode the PCIe bus topology graph, and static and dynamic features are fused together to generate a unique hardware fingerprint (UHI) using the BLAKE3 hash algorithm.

[0043] Furthermore, when hardware changes (such as PCIe device insertion or removal), only the weights of the affected subgraphs are recalculated to ensure fingerprint stability.

[0044] In summary, the method provided by the embodiments of this disclosure encodes the PCIe topology graph through a graph attention network (GAT), dynamically weights and fuses static features and dynamic sensor data to generate a globally unique hardware fingerprint (UHI), which solves the identification conflict problem in scenarios with mixed hardware deployment from multiple vendors and improves the uniqueness and stability of the hardware fingerprint.

[0045] In some embodiments, step 102 above may be implemented by the following steps: Step 2011: Construct AI models, including a spatiotemporal prediction sub-model for predicting future load demands and a fault prediction sub-model for predicting hardware failure probabilities. In one embodiment, a spatiotemporal graph is constructed, which includes hardware components (such as CPU, memory, NVMe) as nodes and physical link latency and historical load correlation coefficients (e.g., Pierre correlation coefficient > 0.7) as edge weights. A spatiotemporal graph convolutional network (e.g., convolutional kernel size 3×3, number of channels 64→128) is used to extract local spatiotemporal features, and a Transformer encoder (e.g., number of heads 8, vector dimension size 256) is combined to capture global temporal dependencies, generating a spatiotemporal prediction sub-model for predicting future load demand.

[0046] In another embodiment, based on long short-term memory networks and multi-granularity attention mechanisms, combined with hardware health indicators, the probability of hardware failure is predicted, and the fault prediction sub-model for predicting the probability of hardware failure is constructed.

[0047] Among them, the multi-granularity attention mechanism:

[0048] : Attention weight at time step t, representing the attention intensity at the current time step.

[0049] σ: Activation function (such as Sigmoid or ReLU) used to map the output of a linear combination to a specific range.

[0050] Short window weight matrix, used to adjust the hidden state of the short window. The importance of.

[0051] : The LSTM short window (60 seconds) hidden state at time step t, representing the short-term features of the current time step.

[0052] Long window weight matrix, used to adjust the long window feature pooling results. The importance of.

[0053] Long window (300 seconds) feature pooling results represent the summary information of long-term features.

[0054] Hardware health weight matrix, used to adjust hardware health indicators. The importance of.

[0055] Hardware health metrics (such as SMART logs) indicate the health status of the hardware.

[0056] Step 2012: Based on the hardware fingerprint and historical load data, predict future load demand using the spatiotemporal prediction model; Step 2013: Based on the dynamic data, use the fault prediction sub-model to predict the probability of hardware failure; Step 2014: Based on the future load demand and the hardware failure probability, generate a state space including a hardware fingerprint, predicting resource demand and hardware failure probability, including a discrete action space containing operating system image version selection and a continuous action space containing CPU quota coefficient, wherein the reward function is calculated based on actual resource utilization, deployment time and hardware failure probability.

[0057] The state space S is a vector that contains multiple features describing the current state of the system:

[0058] : Hardware fingerprint embedding vector, representing the encoded information of hardware features.

[0059] Pred_CPU: Predicted CPU utilization (percentage).

[0060] Pred_Mem: Predicted memory utilization (percentage).

[0061] Hardware failure probability (between 0 and 1).

[0062] QoS historical percentile: Historical percentile value of Quality of Service (QoS), a statistical indicator used to evaluate system performance.

[0063] Action space A contains all possible actions that the agent can take. It is specifically divided into two categories: Discrete Actions: OS Image Version Selection: Select the version of the operating system image, such as "Linux Kernel 6.6 with RT patch".

[0064] Continuous Actions: CPU Quota Coefficient: Sets the CPU resource quota coefficient, calculated as CPU_Quota=Pred_CPU×[0.8,1.2].

[0065] The reward function R is used to evaluate the immediate reward of an agent after taking a certain action in a specific state. The specific formula is as follows:

[0066] Actual_CPU: Actual CPU utilization (percentage).

[0067] Pred_CPU: Predicted CPU utilization (percentage).

[0068] Deploy_Time: Deployment time (seconds).

[0069] Threshold: Deployment time threshold (e.g., 30 seconds).

[0070] Hardware failure probability (between 0 and 1).

[0071] In summary, the method provided by the embodiments of this disclosure, with its spatiotemporal graph convolutional network (ST-GCN) + Transformer hybrid architecture, significantly improves the accuracy of load prediction, enhances the fault detection rate through a multi-granularity fault prediction mechanism, provides precise strategy guidance for dynamic deployment, and optimizes resource allocation and fault prevention.

[0072] Furthermore, in one possible implementation of this embodiment, Figure 2 This is a flowchart illustrating another method for dynamically deploying an operating system provided in an embodiment of this disclosure. Figure 2 As shown, based on the above embodiments, the method further includes the following steps: Step 106: During the BMC secure boot phase, the hash value of the customized operating system image is bound to the platform configuration register (PCR0) of the Trusted Platform Module (TPM) along with the dynamic deployment strategy and the verification logic is performed. If the logic verification passes, the BMC is forced to power down. Specifically, the current value of the PCR0 register is read from the TPM. If the value of PCR0 does not match the expected hash value, it indicates that there may be a security problem during system startup, triggering the BMC to force a power-off to protect the system.

[0073] Step 107: Calculate the process hash during runtime in real time. If there is a process hash that does not belong to the hardware fingerprint, trigger IPMI to force power down.

[0074] Specifically, the process hash value is calculated based on the executable file content of the process, the hardware fingerprint, and the current value of the PCR0 register.

[0075] In summary, the methods provided by the embodiments of this disclosure, including Trusted Platform Module (TPM) PCR dynamic extension and runtime process hash verification, enhance the full-stack dynamic security protection capabilities of the system, effectively block the loading of unauthorized firmware / operating system (OS), intercept malicious code injection in real time, and ensure the safe and reliable operation of the system.

[0076] Furthermore, in one possible implementation of this embodiment, Figure 3This is a flowchart illustrating another method for dynamically deploying an operating system provided in this disclosure. Based on the above embodiments, the method further includes the following steps: Step 106: Obtain local data based on the execution log, calculate the loss function based on the actual resource utilization and future load requirements, filter the feature layer gradients that are strongly correlated with the hardware fingerprint, and generate the local AI prediction model gradient. Specifically, the loss function:

[0077] Loss: Total loss value, used to measure the accuracy of model predictions and the performance of fault prediction.

[0078] MAE: Mean Absolute Error, used to measure the difference between predicted CPU utilization and actual CPU utilization.

[0079] Pred_CPU: Predicted CPU utilization (percentage).

[0080] Actual_CPU: Actual CPU utilization (percentage).

[0081] FocalLoss: An improved cross-entropy loss function for handling class imbalance problems, particularly suitable for fault prediction.

[0082] Actual_Fault: The actual fault state (0 indicates no fault, 1 indicates a fault).

[0083] Step 107: Calculate the hardware performance score based on the deployment time and hardware failure probability, and adjust the gradient weights of the local AI prediction model based on the hardware performance score to obtain the globally optimized model parameters. Specifically, the weight calculation formula is as follows:

[0084] Global_Grad: Global gradient, used to update model parameters on the central server.

[0085] Deploy_Time[i]: Deployment time (in seconds) for the i-th edge node.

[0086] Local_Grad[i]: The local gradient of the i-th node, representing the direction and magnitude of the model parameter update at that node.

[0087] Step 108: Push incremental packets including the globally optimized model parameters through the BMC out-of-band channel to update the local AI prediction model, and update the dynamic deployment strategy based on the BMC dual-bank flash switching.

[0088] In summary, the method provided by the embodiments of this disclosure, based on hardware fingerprint (UHI) driven gradient weighting and edge node hot update mechanism, significantly improves the model convergence speed, reduces the edge node resource over-provisioning rate, improves resource utilization, realizes closed-loop optimization of the system, and further improves the system performance and security.

[0089] Based on the above embodiments, in one possible implementation of this embodiment, the method includes the following steps: Step 101: Obtain server hardware device data characteristics through BMC and generate hardware fingerprint; Step 102: Based on the hardware fingerprint and historical load data, use an AI model to predict future load requirements, and generate a dynamic deployment strategy based on the future load requirements, including hardware configuration information, operating system image version selection, driver dependency information and resource allocation strategy. Step 103: Based on the hardware configuration information, perform non-uniform memory access aware memory sharding on the server hardware device through BMC, and re-enumerate the PCIe bus topology. Step 104: Based on the operating system image version selection and driver dependency information, and according to the hardware fingerprint matching driver operating system image whitelist, reconstruct the initial memory file system and generate a customized operating system image. Step 105: Perform hybrid orchestration of containerized deployment and lightweight virtual machines, and allocate resources based on the resource allocation strategy; Step 106: During the BMC secure boot phase, the hash value of the customized operating system image is bound to the platform configuration register of the trusted platform module along with the dynamic deployment strategy and the verification logic is performed. If the logic verification passes, the BMC is forced to power down. Step 107: Calculate the process hash during runtime in real time. If there is a process hash that does not belong to the hardware fingerprint, trigger IPMI to force power down.

[0090] Step 108: Obtain local data based on the execution log, calculate the loss function based on the actual resource utilization and future load requirements, filter the feature layer gradients that are strongly correlated with the hardware fingerprint, and generate the local AI prediction model gradient. Step 109: Calculate the hardware performance score based on the deployment time and hardware failure probability, and adjust the gradient weights of the local AI prediction model based on the hardware performance score to obtain the globally optimized model parameters. Step 110: Push an incremental package containing the globally optimized model parameters through the BMC out-of-band channel to update the local AI prediction model, and update the dynamic deployment strategy based on the BMC dual-bank flash switching.

[0091] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0092] According to embodiments of this disclosure, this disclosure also provides an operating system dynamic deployment apparatus.

[0093] For example, Figure 4 This is a schematic diagram of the structure of an operating system dynamic deployment device provided in an embodiment of this disclosure. The operating system dynamic deployment device 400 includes: a hardware fingerprint generation module 401, a dynamic deployment strategy generation module 402, a hardware configuration module 403, an operating system image building module 404, and a hybrid orchestration module 405; wherein, The hardware fingerprint generation module 401 is used to obtain server hardware device data characteristics through BMC and generate a hardware fingerprint. The dynamic deployment strategy generation module 402 is used to predict future load demand using an AI model based on the hardware fingerprint and historical load data, and generate a dynamic deployment strategy based on the future load demand, including hardware configuration information, operating system image version selection, driver dependency information and resource allocation strategy. Hardware configuration module 403 is used to perform non-uniform memory access aware memory sharding on server hardware devices through BMC based on the hardware configuration information, and to re-enumerate the PCIe bus topology. The operating system image building module 404 is used to reconstruct the initial memory file system and generate a customized operating system image based on the operating system image version selection and driver dependency information, according to the hardware fingerprint matching driver operating system image whitelist. The hybrid orchestration module 405 is used to perform hybrid orchestration of containerized deployments and lightweight virtual machines, and to allocate resources based on the resource allocation strategy.

[0094] Furthermore, the hardware fingerprint generation module 401 is used to obtain server hardware device data characteristics through the BMC, and specifically for generating a hardware fingerprint when doing so: The BMC interface driver obtains static data including CPU microcode version, PCIe bus topology and DRAM SPD information, and generates static features including CPU microcode hash value, PCIe topology map and DRAM SPD matrix. The BMC sensor interface acquires dynamic data including real-time sensor data and hardware health status, and generates dynamic features including sensor time series and memory health. Based on the static and dynamic features, a globally unique hardware fingerprint is generated.

[0095] Furthermore, the dynamic deployment strategy generation module 402 is used to predict future load demands using an AI model based on the hardware fingerprint and historical load data, and to generate a dynamic deployment strategy including hardware configuration information, operating system image version selection, driver dependency information, and resource allocation strategy based on the future load demands. Specifically, it is used for: Build AI models, including a spatiotemporal prediction sub-model for predicting future load demand and a fault prediction sub-model for predicting hardware failure probability; Based on the hardware fingerprint and historical load data, the spatiotemporal prediction model is used to predict future load demand. Based on the dynamic data, the fault prediction sub-model is used to predict the probability of hardware failure. Based on the future load demand and the hardware failure probability, a state space is generated that includes a hardware fingerprint, predicts resource demand and hardware failure probability, and includes an action space containing discrete actions with operating system image version selection and continuous actions with CPU quota coefficient. The reward function is calculated based on actual resource utilization, deployment time and hardware failure probability.

[0096] Furthermore, when 402 is used to construct the spatiotemporal prediction sub-model for predicting future load demand, it is specifically used for: A spatiotemporal graph is constructed, which includes hardware components as nodes and physical link latency and historical load correlation coefficients as weights. A spatiotemporal graph convolutional network is used to extract local spatiotemporal features, and a Transformer encoder is combined to capture global time dependencies, generating a spatiotemporal prediction sub-model for predicting future load demand.

[0097] Further, 402 is used to construct the fault prediction sub-model for predicting hardware fault probabilities, specifically for " Based on long short-term memory networks and multi-granularity attention mechanisms, combined with hardware health indicators, the probability of hardware failure is predicted, and a fault prediction sub-model for predicting the probability of hardware failure is constructed.

[0098] Furthermore, the operating system dynamic deployment device 400 also includes: The first security and trust assurance module 406 is used to bind the hash value of the customized operating system image and the dynamic deployment strategy to the platform configuration register of the trusted platform module and verify the logic during the BMC secure startup phase. If the logic verification is successful, the BMC is forced to power down. The second security and trust assurance module 407 is used to calculate the process hash during runtime in real time. When there is a process hash that does not belong to the hardware fingerprint, it triggers IPMI to force power down.

[0099] Furthermore, the operating system dynamic deployment device 400 also includes: The first feedback optimization and incremental learning module 408 is used to obtain local data based on the execution log, calculate the loss function based on the actual resource utilization and future load demand, filter the feature layer gradients that are strongly correlated with the hardware fingerprint, and generate the local AI prediction model gradient. The second feedback optimization and incremental learning module 409 is used to calculate the hardware performance score based on the deployment time and hardware failure probability, and adjust the gradient weight of the local AI prediction model based on the hardware performance score to obtain the globally optimized model parameters. The third feedback optimization and incremental learning module 410 is used to push incremental packages including the globally optimized model parameters through the BMC out-of-band channel, update the local AI prediction model, and update the dynamic deployment strategy based on the BMC dual-bank flash switching.

[0100] It should be noted that the description of the features in the embodiment corresponding to the operating system dynamic deployment device can be found in the relevant description of the embodiment corresponding to the operating system dynamic deployment method, and will not be repeated here.

[0101] Embodiments of this disclosure also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above embodiments of the operating system dynamic deployment method.

[0102] Embodiments of this disclosure also provide a computer-readable storage medium storing a computer program configured to execute the steps in any of the above-described embodiments of the operating system dynamic deployment method.

[0103] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0104] Embodiments of this disclosure also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the operating system dynamic deployment method.

[0105] Embodiments of this disclosure also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in any of the above-described embodiments of the operating system dynamic deployment method.

[0106] Those skilled in the art will further recognize that the modules and algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0107] The target detection method provided in this disclosure has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this disclosure. The descriptions of the embodiments above are only for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this disclosure without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this disclosure.

Claims

1. A method for dynamically deploying an operating system, characterized in that, include: Hardware fingerprints are generated by acquiring server hardware device data characteristics through BMC. Based on the hardware fingerprint and historical load data, an AI model is used to predict future load requirements, and a dynamic deployment strategy including hardware configuration information, operating system image version selection, driver dependency information and resource allocation strategy is generated based on the future load requirements. Based on the hardware configuration information, non-uniform memory access-aware memory sharding is performed on the server hardware device through BMC, and the PCIe bus topology is re-enumerated. Based on the operating system image version selection and driver dependency information, and according to the hardware fingerprint matching driver operating system image whitelist, the initial memory file system is reconstructed to generate a customized operating system image. The containerized deployment is combined with lightweight virtual machines for hybrid orchestration, and resources are allocated based on the resource allocation strategy.

2. The method according to claim 1, characterized in that, The step of obtaining server hardware device data features through BMC and generating hardware fingerprints includes: The BMC interface driver obtains static data including CPU microcode version, PCIe bus topology and DRAM SPD information, and generates static features including CPU microcode hash value, PCIe bus topology diagram and DRAM SPD matrix. The BMC sensor interface acquires dynamic data including real-time sensor data and hardware health status, and generates dynamic features including sensor time series and memory health. Based on the static and dynamic features, a globally unique hardware fingerprint is generated.

3. The method according to claim 2, characterized in that, The dynamic deployment strategy, which uses an AI model to predict future load demands based on the hardware fingerprint and historical load data, and generates a strategy based on these future load demands that includes hardware configuration information, operating system image version selection, driver dependency information, and resource allocation strategies, includes: Build AI models, including a spatiotemporal prediction sub-model for predicting future load demand and a fault prediction sub-model for predicting hardware failure probability; Based on the hardware fingerprint and historical load data, the spatiotemporal prediction model is used to predict future load demand. Based on the dynamic data, the fault prediction sub-model is used to predict the probability of hardware failure. Based on the future load demand and the hardware failure probability, a state space is generated that includes a hardware fingerprint, predicts resource demand and hardware failure probability, and includes an action space containing discrete actions with operating system image version selection and continuous actions with CPU quota coefficient. The reward function is calculated based on actual resource utilization, deployment time and hardware failure probability.

4. The method according to claim 3, characterized in that, Constructing the spatiotemporal prediction sub-model for predicting future load demand includes: A spatiotemporal graph is constructed, which includes hardware components as nodes and physical link latency and historical load correlation coefficients as edge weights. A spatiotemporal graph convolutional network is used to extract local spatiotemporal features, and a Transformer encoder is combined to capture global time dependencies, generating a spatiotemporal prediction sub-model for predicting future load demand.

5. The method according to claim 3, characterized in that, Constructing the fault prediction sub-model for predicting hardware failure probability includes: Based on long short-term memory networks and multi-granularity attention mechanisms, combined with hardware health indicators, the probability of hardware failure is predicted, and a fault prediction sub-model for predicting the probability of hardware failure is constructed.

6. The method according to claim 1, characterized in that, The process of combining containerized deployment with lightweight virtual machines for hybrid orchestration, and allocating resources based on the resource allocation strategy, further includes: During the BMC secure boot phase, the hash value of the customized operating system image is bound to the platform configuration register of the trusted platform module along with the dynamic deployment strategy and the verification logic is performed. If the logic verification passes, the BMC is forced to power down. The process hash is calculated in real time. When a process hash that does not belong to the hardware fingerprint is present, IPMI is triggered to force a power-down.

7. The method according to claim 1, 3, or 6, characterized in that, The process of combining containerized deployment with lightweight virtual machines for hybrid orchestration, and allocating resources based on the resource allocation strategy, further includes: Local data is obtained from the execution log, and a loss function is calculated based on the actual resource utilization and future load requirements. The gradients of the feature layers that are strongly correlated with the hardware fingerprint are selected to generate the gradients of the local AI prediction model. The hardware performance score is calculated based on the deployment time and hardware failure probability, and the gradient weights of the local AI prediction model are adjusted based on the hardware performance score to obtain the globally optimized model parameters. The local AI prediction model is updated by pushing incremental packages including the globally optimized model parameters through the BMC out-of-band channel, and the dynamic deployment strategy is updated based on the BMC dual-bank flash switching.

8. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, The computer program is executed by a processor using the method described in any one of claims 1 to 7.

Images