Adaptive data backup and recovery method for hybrid cloud environment

Abstract

The application belongs to the technical field of industrial cloud storage, and discloses a self-adaptive data backup and recovery method for a hybrid cloud environment, which comprises the following steps: collecting the input-output response time gradient of a storage node, and when the input-output response time gradient exceeds a preset degradation threshold, issuing a bus interrupt signal to suspend a file system addressing request, bypassing an operating system file system buffer stack, accessing a non-volatile random access memory based on a physical memory address mapping table, extracting a dirty page bitmap and a storage state register pointer; encapsulating the dirty page bitmap and the storage state register pointer into a state feature packet, calculating an emergency transmission weight value according to a business importance level identifier, and pushing the state feature packet to a backup center through a control plane out-of-band bus. The application solves the problem of data truncation caused by the bottom queue lock of the storage node in the critical degradation period, ensures the physical escape of the core incremental state, and constructs a decoupled fault-tolerant data rescue path.

Description

Technical Field

[0001] This invention belongs to the field of industrial cloud storage technology and relates to an adaptive data backup and recovery method for hybrid cloud environments. It is used for data disaster recovery in continuous chemical production bases or other industrial IoT production control layers, and can ensure the integrity of core incremental data under extreme crash conditions such as the storage node's underlying queue being locked. Background Technology

[0002] Currently, ensuring the continuity and consistency of business data through high-availability architecture and disaster recovery backup mechanisms has become the core path of modern data storage systems. However, existing technologies still have limitations in handling the physical constraints of cross-regional disaster recovery, system-level backup efficiency, and long-sequence recovery under complex failure scenarios. First, there is the disaster recovery takeover technology based on nodes within the cluster. For example, Chinese invention patent CN114281591A discloses a storage node failure handling method, device, equipment, and storage medium. It utilizes striped metadata synchronization and backup between master and slave nodes to achieve business takeover. This solution is essentially a local state feedback within a single cluster. The disaster recovery logic lacks representation of global system-level disasters such as cross-regional and cross-data center disasters. It does not map node-level disaster recovery to the isolation boundary of geographically dispersed physical spaces. As a result, when the system faces extreme disturbances such as data center power outages or network paralysis, it may face the risk that the entire cluster data may be unavailable and irretrievable.

[0003] Second, there is the local hot backup technology based on memory state filtering. For example, Chinese invention patent CN102662799A discloses a data backup method, server, and hot backup system. It filters physical memory without dirty pages by judging the flag bit of the virtual memory area (VMA) and skips invalid traversal to improve efficiency. Its drawback is that this technology focuses on the identification and local optimization of memory running state. Although it improves the efficiency of application-level hot backup, it fails to establish a closed-loop association between memory-level state and the disaster recovery of underlying persistent disk data. It cannot map the pressure on business continuity caused by physical media damage in real time and is difficult to cope with the risk of data overdraft and loss under system-level hardware failure.

[0004] Third, cross-cloud data backup technology based on discrete script scheduling; for example, Chinese invention patent application CN114924909A discloses a cross-cloud multi-combination data backup and recovery method, which writes monthly full, weekly incremental and log backup scripts for cross-cloud nodes and schedules them for execution on a regular basis. Its shortcomings are: existing architectures of this kind usually regard the backup cycle as a discrete or fixed-step strategy unit, ignoring the continuous consistency of the underlying data flow in the long time series and the inherent coupling of the system; when performing long-cycle data recovery, if there is a lack of a globally unified metadata mandatory verification and assembly mechanism, the recovery trajectory is very likely to produce time point offset and assembly delay, causing the output recovery solution to lose its practical value because it exceeds the business tolerance time recovery time (RTO).

[0005] Therefore, how to construct a disaster recovery architecture with global state awareness and cross-level collaboration capabilities, so that data backup and recovery logic can quickly converge to the real high-availability physical security boundary in complex disaster scenarios, has become the technical problem to be solved by this invention. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art, solve the technical problem that storage nodes cannot extract core state data by bypassing the failed file system stack during the critical deterioration period when the input and output are physically locked, and provide an adaptive data backup and recovery method for hybrid cloud environments.

[0007] To achieve the above-mentioned objectives, the present invention provides an adaptive data backup and recovery method for hybrid cloud environments, comprising the following steps: Step S1, Collect I / O response delay gradient and determine degradation threshold: Collect the input and output response delay gradient of the storage node and determine whether the input and output response delay gradient exceeds the preset degradation threshold; Step S2, issue an interrupt to suspend the file system and perform physical addressing to retrieve data: When the input / output response delay gradient exceeds the degradation threshold, issue a bus interrupt signal to suspend the addressing request for the storage node file system, bypass the operating system file system buffer stack of the storage node, and perform physical addressing of the non-volatile random access memory based on the physical memory address mapping table to extract the dirty page bitmap and storage status register pointer of the data to be backed up. Step S3, Concatenate fields to construct a feature message and calculate the emergency transmission weight: Concatenate the dirty page bitmap and the storage status register pointer by bit fields to construct a data packet structure with atomic recovery attributes, obtain a status feature message consisting of a header segment and a payload segment, and calculate the emergency transmission weight of the status feature message based on the service importance level identifier of the data to be backed up. Step S4: Enable the out-of-band bus and push messages to the backup center according to the weights: Close the data transmission channel of the storage node, enable the control plane out-of-band bus of the storage node, and push status feature messages to the backup center through the control plane out-of-band bus according to the transmission priority order determined by the emergency transmission weights.

[0008] The step S1 of the present invention for collecting the input-output response delay gradient of the storage node includes the following sub-steps: Step S11, within a preset sampling period, monitor in real time the input-output queue depth of the non-volatile random access memory controller in the storage node and the interrupt response duration within a unit clock cycle; Step S12, calculate the rate of change of the interrupt response duration within adjacent sampling periods, and combine it with the weighted average of the input-output queue depth to generate the input-output response delay gradient.

[0009] The physical addressing of the non-volatile random access memory in step S2 of the present invention includes: locking the hardware memory controller of the non-volatile random access memory, and reading the original binary stream data stored in the non-volatile memory area according to the base address in the physical memory address mapping table, under the condition of bypassing the operating system kernel interrupt, in order to identify the dirty page bitmap marked as unsynchronized.

[0010] The bit field concatenation in step S3 of the present invention includes: according to a preset protocol mapping template, encapsulating the dirty page bitmap into the payload segment of the status feature message, and embedding the storage status register pointer bit-left by bit into the header segment of the status feature message.

[0011] The step S4 of the present invention, which enables the control plane out-of-band bus of the storage node, includes: waking up the baseboard management controller in the storage node and sending a switching signal to the baseboard management controller to switch the physical layer communication path from the Ethernet controller to a dedicated hardware bus based on the intelligent platform management interface protocol.

[0012] The method described in this invention further includes the following recovery steps: Step S5, the backup center receives a status feature message and locates the register address of the target recovery node based on the storage status register pointer in the header segment; Step S6, the corresponding physical block incremental data is pushed to the target recovery node according to the dirty page bitmap in the payload segment.

[0013] The degradation threshold described in this invention is determined by: statistically analyzing the historical input / output response latency distribution of storage nodes under normal operating conditions, and calculating the standard deviation of the historical input / output response latency; using three times the mean of the standard deviation as the initial benchmark for the degradation threshold, and dynamically correcting the initial benchmark based on the real-time load fluctuation of the storage nodes.

[0014] In step S2 of the present invention, before the operating system file system buffer stack of the bypass storage node, the method further includes: sending a suspend instruction to the input / output scheduler of the storage node to freeze the write transactions currently in the committed state and isolate the damaged memory page table entries to prevent the dirty page bitmap from being overwritten during the addressing process.

[0015] After pushing the corresponding physical block incremental data to the target recovery node in step S6 of the present invention, the method further includes: comparing the current state hash value of the target recovery node with the verification hash value carried in the state feature message, and releasing the input / output lock state of the target recovery node when the current state hash value and the verification hash value are consistent.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: 1. In adaptive data backup in a hybrid cloud environment, the core data of storage nodes is physically rescued during the critical deterioration period. By monitoring the I / O response latency gradient of storage nodes in real time, and forcibly bypassing the standard file system buffer stack of the operating system when the latency gradient exceeds a preset threshold, this mechanism directly performs physical addressing of non-volatile random access memory to extract dirty page bitmaps. This extraction method, which sinks to the hardware control level, breaks the implicit dependence of traditional backup methods on the integrity of operating system kernel interrupts and file system read and write calls. It ensures that even under extreme crash conditions such as the underlying I / O queue being locked and the standard file system process being suspended, the incremental state with the highest business importance can still be deterministically captured, eliminating the fault tolerance blind spot of industrial production control nodes at the edge of hardware failure.

[0017] 2. Physical decoupling of the data rescue path from the main business transmission protocol stack is constructed. By bit-concatenating the dirty page bitmap and the state machine register offset pointer at the microcode layer, a miniaturized hijacking vector is generated, and the priority algebra weight is calculated. This method can compress the system state to be backed up to a smaller size. Then, the transmission link of the main data network is cut off, and an independent out-of-band management bus interface is activated to perform broadcast tasks. This path of transmitting microdata using a low-bandwidth, high-reliability physical link avoids the TCP / IP protocol stack and Ethernet driver layer, which are extremely prone to collapse in the instant of hardware degradation. This ensures that the escape process of the core control state does not occupy the backbone network bandwidth and does not depend on the file system entity, thereby improving the system's survivability in extreme network congestion scenarios. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the adaptive backup execution process triggered by the critical degradation of the storage node in this invention; Figure 2 This is a schematic diagram of the hardware and software collaborative rescue topology of the bypass file system stack of this invention. Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings.

[0020] Example 1: This embodiment discloses an adaptive data backup and recovery method for hybrid cloud environments, including the following steps: Step S1, Collect I / O response delay gradient and determine degradation threshold: Collect the input and output response delay gradient of the storage node and determine whether the input and output response delay gradient exceeds the preset degradation threshold. Step S2, issue an interrupt to suspend the file system and perform physical addressing to retrieve data: When the input / output response delay gradient exceeds the degradation threshold, issue a bus interrupt signal to suspend the addressing request for the storage node file system, bypass the operating system file system buffer stack of the storage node, and perform physical addressing of the non-volatile random access memory based on the physical memory address mapping table to extract the dirty page bitmap and storage status register pointer of the data to be backed up. Step S3, Concatenate fields to construct a feature message and calculate the emergency transmission weight: Concatenate the dirty page bitmap and the storage status register pointer by bit fields to construct a data packet structure with atomic recovery attributes, obtain a status feature message consisting of a header segment and a payload segment, and calculate the emergency transmission weight of the status feature message based on the service importance level identifier of the data to be backed up. Step S4: Enable the out-of-band bus and push messages to the backup center according to the weights: Close the data transmission channel of the storage node, enable the control plane out-of-band bus of the storage node, and push status feature messages to the backup center through the control plane out-of-band bus according to the transmission priority order determined by the emergency transmission weights.

[0021] In this embodiment, step S1, which involves collecting the input / output response delay gradient of the storage node, includes the following sub-steps: Step S11, within a preset sampling period, real-time monitoring of the input / output queue depth of the non-volatile random access memory controller in the storage node and the interrupt response duration within a unit clock cycle; Step S12, calculating the rate of change of the interrupt response duration within adjacent sampling periods, and combining it with the weighted average of the input / output queue depths to generate the input / output response delay gradient.

[0022] In step S2 of this embodiment, physical addressing of the non-volatile random access memory includes: locking the hardware memory controller of the non-volatile random access memory, and reading the original binary stream data stored in the non-volatile memory area according to the base address in the physical memory address mapping table, while bypassing the operating system kernel interrupt, to identify dirty page bitmaps marked as unsynchronized.

[0023] In step S3 of this embodiment, bit field concatenation includes: according to a preset protocol mapping template, encapsulating the dirty page bitmap into the payload segment of the status feature message, and then embedding the storage status register pointer bit-left into the header segment of the status feature message.

[0024] In step S4 of this embodiment, enabling the control plane out-of-band bus of the storage node includes: waking up the baseboard management controller in the storage node and sending a switching signal to the baseboard management controller to switch the physical layer communication path from the Ethernet controller to a dedicated hardware bus based on the intelligent platform management interface protocol.

[0025] The method described in this embodiment further includes the following recovery steps: Step S5, the backup center receives the status feature message and locates the register address of the target recovery node according to the storage status register pointer in the header segment; Step S6, according to the dirty page bitmap in the payload segment, pushes the corresponding physical block incremental data to the target recovery node.

[0026] The degradation threshold described in this embodiment is determined in the following way: statistically analyze the historical input-output response latency distribution of the storage node under normal operating conditions, and calculate the standard deviation of the historical input-output response latency; use three times the mean of the standard deviation as the initial benchmark for the degradation threshold, and dynamically correct the initial benchmark according to the real-time load fluctuation of the storage node.

[0027] In step S2 of this embodiment, before the operating system file system buffer stack of the bypass storage node, the method further includes: sending a suspend instruction to the input / output scheduler of the storage node to freeze the write transactions currently in the committed state and isolate the damaged memory page table entries to prevent the dirty page bitmap from being overwritten during the addressing process.

[0028] In step S6 of this embodiment, after pushing the corresponding physical block incremental data to the target recovery node, the method further includes: comparing the current state hash value of the target recovery node with the verification hash value carried in the state feature message, and releasing the input / output lock state of the target recovery node when the current state hash value and the verification hash value are consistent.

[0029] Example 2: In this embodiment, within a continuous chemical production base comprising an industrial IoT production control layer and a hybrid cloud architecture, the core data disaster recovery and maintenance system handles instantaneous read / write requests generated by a large-scale sensor cluster. Private cloud distributed storage nodes and public cloud archive data centers engage in cross-domain data interaction. The physical disks of each storage node are subject to hardware degradation, experiencing a critical degradation period on the order of microseconds before physical failure. During this period, the operating system kernel of the storage node suspends its file system buffer stack due to the locking of the underlying input / output queue, preventing the backup program based on the standard software protocol stack from acquiring incremental data blocks. When the system detects that the input / output response latency gradient of the storage node exceeds a preset degradation threshold, the controller issues a bus interrupt signal, suspending all addressing requests to the storage node's file system and bypassing the storage node's operating system file system buffer stack. The controller then accesses the non-volatile random access memory (NVM) according to the physical memory address mapping table. RAM addressing locks the base address of the hardware memory controller of the non-volatile random access memory. By bypassing operating system kernel interrupts, the raw binary stream data is extracted, thereby identifying the dirty page bitmap marked as unsynchronized and the corresponding memory status register pointer. The controller reads the reserved hardware descriptor sectors in the non-volatile memory (NVRAM) to obtain the page descriptor metadata synchronously written to the hardware layer by the memory management unit (MMU). Each bit in this metadata corresponds one-to-one with the synchronization status of the physical memory page. When a physical page is written but not yet written to disk, the hardware controller automatically sets the corresponding descriptor bit. This invention directly locks the physical offset address of the descriptor sector through the physical address mapping table and segments the read raw binary stream according to a preset bit width. This allows for the direct extraction of the dirty page bitmap reflecting the data update status at the hardware level without relying on operating system file system metadata parsing.

[0030] To extract the core state under physical lock-up conditions, the controller concatenates the extracted dirty page bitmap with the memory state register pointer using bit fields to generate a data packet structure with atomic-level recovery properties. This results in a state feature message consisting of a header segment and a payload segment, and the emergency transmission weight is calculated according to the following formula: ,in, For emergency transmission weights; This is the quantized value of the input-output response delay gradient; The remaining timeout period set for the controller's built-in hardware watchdog. A preset quantization constant is used to identify the business importance level of the data to be backed up; before the controller is substituted into the formula for calculation, the execution cycle of a single instruction from the main hardware controller is extracted as the timeliness baseline constant. Compare the remaining timeout periods and Numerical size; in Greater than Calculate the emergency transmission weight in time ,exist Less than or equal to The division operation path is truncated at the time, and the microprocessor supports assigning the maximum single-precision floating-point value to the emergency transmission weight. The system wakes up the baseboard management controller and sends a switching signal, switching the physical layer communication path from the Ethernet controller to a dedicated hardware bus based on the intelligent platform management interface protocol. This enables the out-of-band bus of the storage node's control plane. Before the main data network protocol stack fails, the baseboard management controller sends the status feature message to the out-of-band management front-end deployed in the industrial IoT LAN via the dedicated hardware bus. The out-of-band management front-end parses the message payload, adds the public cloud archive data center communication endpoint identifier, and calls the standard WAN routing protocol stack to push the status feature message to the backup center. After receiving the status feature message, the backup center locates the register address of the target recovery node according to the storage status register pointer in the header segment, and performs an algebraic reconstruction operation independent of the file system entity in the redundant erasure code pool in combination with the dirty page bitmap in the payload segment. This restores the consistency state of the damaged data within the window after the node's physical collapse. This process moves the data acquisition intervention point from the operating system file system layer to the register micro-element layer of the hardware controller, eliminating the risk of industrial data becoming unextractable at the moment of node collapse and maintaining the data integrity of the storage system under hardware failure conditions.

[0031] Example 3: In this embodiment, the verification platform is deployed in a distributed test cluster comprising four storage nodes equipped with baseboard management controllers and a backup center. The data acquisition port is located at the non-volatile random access memory controller, whose hardware clock frequency is 2.4 GHz, and the sampling period is... The setting is used to balance the degradation detection sensitivity and the main control processor load. When a change in the interrupt response delay dispersion is detected, the sampling period is adjusted. The time was set to 10ms. At this time, the main control processor utilization rate was 2.8%, and wideband noise with a signal-to-noise ratio of 18dB and 50Hz power frequency harmonic interference were superimposed on the control bus.

[0032] The experiment was divided into control group 1, control group 2, and experimental group. Control group 1 used a standard file system interface backup method, control group 2 used the Ethernet main channel during transmission, and the experimental group used the method provided in this invention. During the process of simulating physical head degradation leading to an increase in input / output response delay from 0.85ms to 12.6ms, control group 1 experienced a driver stack deadlock 4.2ms after the onset of degradation, with an incremental data recovery rate of 0.0%. Control group 2's status characteristic messages caused signaling blockage in the Ethernet main channel, with a message transmission success rate of 14.8%. The experimental group... When the degradation threshold is triggered, the file system buffer stack is bypassed by issuing a bus interrupt signal. The base address of the hardware memory controller is locked from the physical memory address mapping table, and a dirty page bitmap of size 512 bytes is extracted. During the transmission phase, the test group calculates the emergency transmission weight. The value is 8.64, with the business importance level identifier being... Set to version 1.0, the baseboard management controller switches the physical layer path to a dedicated hardware bus based on the intelligent platform management interface protocol. The test group completes message push within a 15.2ms window before the physical failure of the storage node. The backup center achieves 98.6% recovery of critical incremental data through algebraic reconstruction operations. As the degradation threshold increases, the prediction timing becomes delayed. When the predicted node is later than the kernel lock-up time, the data recovery rate drops to 35.4%. This trend reflects the nonlinear constraint relationship between the threshold value and the rescue window. The experiment verifies the synergistic effect of direct addressing non-volatile random access memory and control plane out-of-band bus under the condition of operating system kernel failure. The data proves that by completing data encapsulation at the hardware register level, the backup vacuum of the industrial storage system at the moment of node failure can be eliminated, and the data integrity can be continuously maintained during the physical degradation period.

[0033] Example 4: In this embodiment, within an industrial automation control system including a distributed storage cluster, the system manages multiple storage nodes located on different physical racks to maintain real-time backups of production command images. The physical disk response characteristics of each storage node differ due to varying degrees of hardware degradation. During the initialization phase, the system completes a calibration procedure to determine a degradation threshold benchmark for a specific hardware environment. This process involves continuously collecting input / output response delay samples for sixty cycles to form a time series, calculating the first difference of this series to extract the delay fluctuation gradient, and introducing a benchmark deviation factor characterizing the degree of hardware aging. This factor is generated by a linear mapping of the error correction code retry frequency of the storage node's non-volatile random access memory controller. The formula for calculating the degradation threshold is as follows: ,in, This is the degradation threshold; This is the arithmetic mean of the gradients of the collected samples; The standard deviation of the sample gradient; The confidence adjustment coefficient is determined based on the benchmark deviation factor. Through this calibration procedure, the system generates a dynamic judgment limit corresponding to the physical health status of each storage node, avoiding technical implementation deviations caused by using uniform empirical values.

[0034] The input-output response delay gradient of a specific node is identified. breakthrough In an instant, the controller initiates a microcode-level instruction fetching stream. The controller requests physical bus control from the board management controller via direct memory access mode, reads the control registers of the non-volatile random access memory to locate the starting physical offset of the dirty page bitmap, and performs an atomic scan of the memory page table at a granularity of 64 bytes per group. It then performs bit-level concatenation of the physical block addresses marked as not yet written to disk with the corresponding memory status register pointers. This process is combined with the remaining timeout based on the hardware watchdog timer. The dynamic truncation logic restricts message encapsulation to the window before the operating system kernel physical deadlock. After receiving the status feature message pushed through the out-of-band bus, the backup center adjusts the substitution reconstruction operator in the redundancy erasure code pool according to the hardware status information in the message, bypassing the logical volume mounting obstacle caused by the node operating system crash, and realizing the restoration of data consistency. This method eliminates the uncertainty factors in the implementation process by calibrating the degradation threshold and programmatically limiting the data extraction instruction stream, and ensures the reproducibility accuracy of the technical solution under different hardware architectures.

[0035] Example 5: In this example, in a mixed business flow environment that includes multi-dimensional industrial production instructions and real-time monitoring data, the system determines the business importance level by extracting features from the data to be backed up. The process involves identifying the data source tags of the data packets to be backed up to distinguish data types. Specifically, real-time control commands involving inverter drive logic are classified as Level 1 services, production ledgers as Level 2 services, and general environmental monitoring data as Level 3 services. The controller extracts the original score of the corresponding service level from a preset level weight mapping table and performs normalization calculations based on the bit length of the data to be backed up. The result is determined to be a dimensionless value within the range of 0.5 to 2.5, thus providing the emergency transmission weight. The calculations provide physical parameters that match the urgency of the business.

[0036] When the system faces storage node hardware replacement or physical link environment changes, the controller executes a standardized pre-debugging process to calibrate the confidence adjustment coefficient. The process involves collecting one hundred interrupt response duration samples via the out-of-band bus while the node is in a silent state, and then analyzing the statistical distribution characteristics of the sample set to obtain the hardware intrinsic response variance. Based on the intrinsic response variance of the hardware The confidence adjustment coefficient is determined by the ratio of the variance to the preset ideal environmental baseline value. The controller will adjust the calibrated confidence level based on the given value. Write to the non-volatile configuration area of ​​the baseboard management controller of the storage node to set the degradation threshold. The judgment logic shields the response noise introduced by hardware architecture differences, maintaining the consistency of degradation judgment results in heterogeneous hardware clusters.

[0037] Example 6: In this example, in an industrial cloud storage node cluster that includes heterogeneous storage hardware versions, the system performs a pre-deployment calibration process to establish the business importance level identifier. The quantification conversion criteria for physical transmission parameters involve establishing a weighted mapping table with three service levels to quantify service urgency. The initial score for Level 1 services is set to 2.0, for Level 2 services to 1.2, and for Level 3 services to 0.6. The system calculates the ratio of the size of the data packet to be backed up to the standard data block size as an adjustment factor. The product of the initial score and the adjustment factor is used as the identifier for the determined service importance level. The resulting value was then limited to the range of 0.5 to 2.5, and this quantification result was used to calculate the emergency transmission weight. The input parameters enable the emergency transmission weight. Reflecting the physical constraints of the data to be backed up in the production process, when the controller first mounts non-volatile random access memory (NVRAM) on the storage node, the system executes a register base address self-probing procedure to obtain the starting physical offset of the dirty page bitmap. The controller scans the physical address space by sending a set of byte-aligned probe pulses to the memory bus and monitors changes in the control bit state in the physical memory address mapping table. Here, the probe pulses refer to a set of read request level signals of a specific frequency generated by the controller, conforming to the memory bus timing specification. By progressively incrementing the physical address lines, the controller probes the data characteristics fed back from the bus. When the fed-back data signal stream meets preset mask sequence characteristics—in a continuous 64-bit binary sequence, the prefix 16 bits conform to the static signature defined by the NVRAM hardware manufacturer, and subsequent bits exhibit periodically changing flag bit attributes—the controller determines... The current scan position is set as the starting physical offset of the dirty page bitmap. This mask sequence is pre-stored in the firmware register of the controller and is used to achieve blind localization of the bitmap structure in the physical address space through feature matching. When the probe pointer points to the mask sequence with bitmap features, the controller records the logical offset value of the current segment register and converts it into a physical address offset. Then, the physical address offset is determined as the reference starting address for executing the microcode-level extraction instruction stream. This calibration procedure converts the memory layout into a deterministic hardware physical address pointer, ensuring that the dirty page bitmap can be accurately extracted through atomic addressing during the node crash critical period.

[0038] In industrial storage nodes that include multi-head concurrent access architectures, the system executes a composite mapping procedure of latency metrics to generate input-output response latency gradients. This process involves controlling the input / output queue depth of the non-volatile random access memory controller. Perform a weighted moving average calculation, where the weights are assigned according to a time decay factor. The rule calculates the interruption response time within adjacent sampling periods based on the highest weight of the current instantaneous sampled value. The first-order difference is then multiplied by the weighted mean of the input and output queue depths, and the result is determined as the input-output response delay gradient. ,in, The input-output response delay gradient, The depth of the input / output queue. Interruption response duration, For time parameters, The calibration procedure uses a preset attenuation constant to couple the originally decoupled queuing pressure and interrupt jitter into a single-dimensional measure that reflects the degradation of the physical medium, thus eliminating the nondeterministic shift of the judgment parameters under sudden hardware noise interference.

[0039] Upon receiving a status characteristic message, the backup center executes an asymmetric data reconstruction procedure based on hardware register pointers to recover damaged data. This process locks the physical block image in the distributed hash table by parsing the storage status register pointer in the header segment, and performs a bitwise XOR operation on the checksum blocks in the private cloud redundant erasure coding pool using the dirty page bitmap in the payload segment as a logical mask bit. In this recovery procedure, the dirty page bitmap serves as a precise addressing index for the physical blocks, guiding the backup center to retrieve the stored physical block data corresponding to the dirty page location from the redundant erasure coding pool. The system performs a bitwise XOR operation on these stored physical blocks and the cloud redundant checksum blocks to remove the old data items that originally participated in the verification. Finally, the system retrieves the data from the status characteristic message... The incremental state extracted from the report is algebraically synthesized again with the XOR result to deduce the new physical block payload that failed to be written to disk at the moment of node failure. This procedure utilizes the reversible property of XOR operation to transform the bitmap index into a dynamic data selection switch, realizing the closed-loop reconstruction of logical state and physical payload. Thus, under the condition of bypassing the parsing of the operating system file system log of the failed node, logical block data that is completely aligned with the moment before the node crash can be reconstructed on the backup end. This procedure eliminates the metadata synchronization gap caused by the deadlock of the file system kernel by establishing an algebraic mapping relationship between the register state at the hardware microcode level and the global redundancy pool, and realizes the maintenance of storage consistency state under hardware physical failure conditions.

Claims

1. An adaptive data backup and recovery method for hybrid cloud environments, characterized in that, Includes the following steps: Step S1: Collect the input and output response delay gradient of the storage node and determine whether the input and output response delay gradient exceeds the preset degradation threshold. Step S2: When the input / output response delay gradient exceeds the degradation threshold, a bus interrupt signal is sent to suspend the addressing request for the storage node file system, bypass the operating system file system buffer stack of the storage node, and perform physical addressing of the non-volatile random access memory based on the physical memory address mapping table to extract the dirty page bitmap and storage status register pointer of the data to be backed up. Step S3: The dirty page bitmap and the storage status register pointer are concatenated by bit fields to construct a data packet structure with atomic recovery attributes, resulting in a status feature message consisting of a header segment and a payload segment. The urgent transmission weight of the status feature message is calculated based on the service importance level identifier of the data to be backed up. Step S4: Close the data transmission channel of the storage node, enable the control plane out-of-band bus of the storage node, and push status feature messages to the backup center through the control plane out-of-band bus according to the transmission priority order determined by the emergency transmission weight.

2. The adaptive data backup and recovery method for hybrid cloud environments according to claim 1, characterized in that, The step S1 of collecting the input and output response delay gradient of the storage node includes the following sub-steps: Step S11, within the preset sampling period, monitor the input and output queue depth of the non-volatile random access memory controller in the storage node and the interrupt response duration within a unit clock cycle in real time. Step S12: Calculate the rate of change of interrupt response duration within adjacent sampling periods, and combine it with the weighted average of the input and output queue depths to generate the input and output response delay gradient.

3. The adaptive data backup and recovery method for hybrid cloud environments according to claim 1, characterized in that, Step S2, which involves physically addressing the non-volatile random access memory, includes: locking the hardware memory controller of the non-volatile random access memory, and reading the raw binary stream data stored in the non-volatile memory area based on the base address in the physical memory address mapping table, while bypassing operating system kernel interrupts, to identify dirty page bitmaps marked as unsynchronized.

4. The adaptive data backup and recovery method for hybrid cloud environments according to claim 1, characterized in that, Step S3, bit field concatenation, includes: according to the preset protocol mapping template, encapsulating the dirty page bitmap into the payload segment of the status feature message, and embedding the storage status register pointer bit-left into the header segment of the status feature message.

5. The adaptive data backup and recovery method for hybrid cloud environments according to claim 1, characterized in that, The step S4 of enabling the control plane out-of-band bus of the storage node includes: waking up the baseboard management controller in the storage node and sending a switching signal to the baseboard management controller to switch the physical layer communication path from the Ethernet controller to a dedicated hardware bus based on the intelligent platform management interface protocol.

6. The adaptive data backup and recovery method for hybrid cloud environments according to claim 1, characterized in that, The method also includes the following recovery steps: Step S5, the backup center receives the status feature message and locates the register address of the target recovery node according to the storage status register pointer in the header segment; Step S6, according to the dirty page bitmap in the payload segment, pushes the corresponding physical block incremental data to the target recovery node.

7. The adaptive data backup and recovery method for hybrid cloud environments according to claim 1, characterized in that, The degradation threshold is determined as follows: the historical input / output response latency distribution of the storage node under normal operating conditions is statistically analyzed, and the standard deviation of the historical input / output response latency is calculated; three times the mean of the standard deviation is used as the initial benchmark for the degradation threshold, and the initial benchmark is dynamically adjusted according to the real-time load fluctuation of the storage node.

8. The adaptive data backup and recovery method for hybrid cloud environments according to claim 1, characterized in that, Before the operating system file system buffer stack of the bypass storage node in step S2, the method further includes: sending a suspend instruction to the input / output scheduler of the storage node to freeze the write transactions that are currently in the committed state and to isolate the damaged memory page table entries to prevent the dirty page bitmap from being overwritten during the addressing process.

9. The adaptive data backup and recovery method for a hybrid cloud environment according to claim 6, characterized in that, After pushing the corresponding physical block incremental data to the target recovery node in step S6, the method further includes: comparing the current state hash value of the target recovery node with the verification hash value carried in the state feature message, and releasing the input / output lock state of the target recovery node when the current state hash value and the verification hash value are consistent.

Images