Enterprise incremental backup method and system based on content awareness and intelligent scheduling

Abstract

The present application belongs to the technical field of industrial control system data storage and backup, and relates to an enterprise-level incremental backup method and system based on content perception and intelligent scheduling, comprising: loading a filtering driver component in the kernel input-output stack, intercepting data write requests and calculating content hash fingerprints; matching a business topology mapping table to establish data weights; constructing a backup task queue in combination with node load states; starting write back pressure logic when the load exceeds a threshold, intercepting physical disk writing actions and feeding back asynchronous confirmation messages; and completing incremental merging and disk writing by relying on fingerprints through a virtualization driver component. The present application effectively solves the resource mutual exclusion contradiction between backup tasks and real-time production instructions through the kernel state in-situ perception and back pressure feedback mechanism, guarantees the business logic continuity under saturated storage bus conditions, prevents write timeout from inducing retry storms, and improves the data availability of industrial control systems.

Description

Technical Field

[0001] This invention belongs to the field of data storage and backup technology for industrial control systems. It relates to an enterprise-level incremental backup method and system based on content awareness and intelligent scheduling, which is used for real-time backup of large-scale data in industrial control environments. It can improve the data availability of industrial control systems and ensure business continuity. Background Technology

[0002] With the widespread application of large-scale storage systems in data centers and cloud computing environments, ensuring the high reliability of data backup and the consistency of the reading process has become a key research focus in the field of data processing. Currently, the industry has proposed a series of technical solutions for data backup and consistency verification.

[0003] In the prior art, for example, Chinese Patent Publication No. CN119311471A discloses a data backup method and apparatus. The technical means of the solution include: installing a kernel-mode driver and starting a user-mode backup application, obtaining the logical volume information of the target server and sending it to the kernel-mode driver, and having the kernel-mode driver register a block device in the kernel that is consistent with the logical volume information to realize the interaction between user mode and kernel mode. The image of the driver layer improves the ability to capture logical volume data. However, in high-concurrency business scenarios, the frequent interaction of the kernel-mode driver will occupy a lot of system bus resources, and the backup operation directly acts on the main storage path, which can easily lead to an increase in the response latency of the business system.

[0004] For example, Chinese Patent Publication No. CN106293504A discloses a system for solving write pressure by using an existing backup device as an additional main memory. When the input buffer of the main memory unit cannot store the current input dataset, the data stream is redirected to the input buffer of the backup memory unit for storage. Although the technical solution alleviates the write pressure of the main memory by introducing an external buffer, the scheduling of data between different memory units on the heterogeneous storage link increases the complexity of metadata management. Especially when processing non-continuous incremental data, the data synchronization status between the backup unit and the main memory unit is difficult to achieve accurate alignment through a simple buffering mechanism.

[0005] Furthermore, regarding the consistency verification of incremental data, for example, Chinese Patent Publication No. CN109614054B discloses a data reading method. This method obtains incremental metadata used to describe incremental data and extracts consistency markers from it. Based on the markers, it determines whether the incremental data has been successfully written to the disk, thereby determining whether to read the corresponding incremental data from the disk. The solution introduces a consistency verification mechanism in the data reading stage. However, in multi-level storage links or distributed architectures, the generation and transmission of consistency markers often lag behind the physical writing of data. In dynamic storage environments where data is frequently updated, relying solely on static consistency markers is insufficient to guarantee the absolute integrity of backup data on any time slice.

[0006] It is evident that while existing technical solutions have made some progress in driver layer mapping, write pressure relief, and incremental data verification, there is still room for improvement in reducing the impact of backup operations on the main system's business performance and in enhancing the collaborative efficiency of data flow and consistency verification in a multi-level storage environment. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art, solve the technical problem of physical mutual exclusion between backup tasks and the underlying input and output of production operations in industrial control environments and the resulting system resource deadlock, and provide an enterprise-level incremental backup method and system based on content awareness and intelligent scheduling.

[0008] To achieve the above-mentioned objectives, this invention provides an enterprise-level incremental backup method based on content awareness and intelligent scheduling, comprising the following steps: Step 101, Load kernel filter driver to synchronously intercept write data stream: Load filter driver component in the operating system kernel input / output stack of industrial control node. When industrial control application initiates write request for data block, filter driver component synchronously intercepts data stream before data block enters storage controller queue. Step 102, Calculate the content hash fingerprint and count the frequency of data block changes: The filter driver component calculates the content hash fingerprint of the intercepted data stream and counts the frequency of data block changes within a preset window time. Step 103, Matching the business topology mapping table to establish data importance weight: The filter driver component matches the preset industrial production business topology mapping table according to the file path index carried in the write request to establish the business relevance identifier and importance weight of the data block. Step 104: Monitor real-time load parameters to build an incremental backup task queue: Monitor the CPU utilization, available memory space, and I / O wait time of the industrial control node in real time as real-time load parameters, and combine them with business relevance indicators and importance weights to build an incremental backup task queue containing multiple incremental backup tasks. Step 105, Overload triggers write backpressure interception and write success feedback: When the real-time load parameter exceeds the preset load threshold, the write backpressure logic is triggered, the filter driver component intercepts the physical disk write action against the storage controller, and sends a write success confirmation signal to the industrial control application. Step 106: Incremental merging and disk persistence are completed by cross-node fingerprint comparison: The control virtualization driver component compares data between nodes based on the priority of each incremental backup task in the incremental backup task queue and the content hash fingerprint, and the target storage node completes the merging and disk persistence of the incremental data.

[0009] Step 103 of the present invention includes the following sub-steps: Step 1031, parsing the directory level and file name characteristics corresponding to the file path index; Step 1032, matching the parsing results with the process nodes in the industrial production business topology mapping table to determine the production link to which the data block belongs; Step 1033, querying the preset coefficients corresponding to the production links in the industrial production business topology mapping table, and mapping the preset coefficients to business relevance identifiers.

[0010] The method for calculating the content hash fingerprint in step 102 of the present invention is as follows: call the kernel-mode encryption interface to perform segmented hash operation on the data stream in the non-paged memory area; store the generated content hash fingerprint in the resident lookup table located in the kernel space; compare the hash fingerprint of the current write request with the historical fingerprint in the resident lookup table, identify the data block whose content has changed, and mark the data block as an incremental backup candidate.

[0011] Step 104 of the present invention includes the following sub-steps: Step 1041, extracting the amount of data to be processed corresponding to each incremental backup task in the incremental backup task queue; Step 1042, dynamically adjusting the processing concurrency of the incremental backup task queue according to the changing trend of real-time load parameters.

[0012] After step 105, the present invention further includes the following sub-steps: Step 1051, temporarily storing the data block intercepted by the write backpressure logic in a preset kernel buffer; Step 1052, monitoring the dynamic decline process of the real-time load parameter; Step 1053, when the real-time load parameter is lower than the preset load threshold, releasing the kernel buffer and asynchronously flushing the data block therein to the storage controller.

[0013] In step 101 of this invention, the filter driver component achieves in-situ residency in the following way: during the operating system boot phase, the filter driver component is loaded to the top layer of the protocol stack of the storage driver; by hooking the dispatch function of the underlying storage port driver, it takes over the addressing instructions for sector-level offsets.

[0014] Following step 104, the present invention further includes the following sub-steps: Step 1043, calculating the consumption rate of storage resources based on the generation frequency of the incremental backup task queue; Step 1044, when the consumption rate exceeds the preset storage warning value, starting the data deduplication logic and using content hash fingerprint to perform cross-node redundant block scanning.

[0015] In step 105 of this invention, the operation of sending a confirmation signal of successful write to the industrial control application is only performed on data whose importance weight is lower than a preset safety threshold.

[0016] The process of statistical data block change frequency in step 102 of the present invention includes: step 1021, establishing a scoring counter indexed by the physical address of the data block; step 1022, whenever a change in the content hash fingerprint of the same physical address is detected, accumulating the score of the scoring counter corresponding to the physical address.

[0017] This invention also provides an enterprise-level incremental backup system based on content awareness and intelligent scheduling, which is used to implement an enterprise-level incremental backup method based on content awareness and intelligent scheduling, including: The data stream interception module is loaded into the operating system kernel input / output stack of the industrial control node. It is used to synchronously intercept the data stream before the data block enters the storage controller queue when the industrial control application initiates a write request for the data block. The feature parsing module is used to calculate the content hash fingerprint of the intercepted data stream and to count the frequency of changes of data blocks within a preset window time. The business association mapping module is used to match the preset industrial production business topology mapping table based on the file path index carried in the write request, and to establish the business association degree identifier and importance weight of the data block. The scheduling and management module is used to monitor the CPU utilization, available memory space, and I / O wait time of industrial control nodes in real time as real-time load parameters. Combined with business relevance identifiers and importance weights, it constructs an incremental backup task queue containing multiple incremental backup tasks. The back pressure control module is used to initiate the write back pressure logic when the real-time load parameters exceed the preset load threshold. The data flow interception module intercepts the physical disk write operation for the storage controller and sends a write success confirmation signal back to the industrial control application to maintain the data flow logic of the industrial control application under the storage bus saturation condition. The data synchronization and merging module is used to compare data between nodes based on the priority of each incremental backup task in the incremental backup task queue and the content hash fingerprint, and control the target storage node to complete the merging and disking of incremental data.

[0018] Compared with the prior art, the present invention has at least the following beneficial effects: 1. In intelligent scheduling of enterprise-level incremental backup, by sinking the incremental awareness logic to the input / output filtering driver of the operating system kernel, the data feature extraction action is anchored to the transient path of production business disk writing, realizing the synchronous processing of data writing and incremental identification. This eliminates the large number of random read operations caused by the file system's post-scanning in traditional backup technology, avoids the backup process from plundering the production control bus bandwidth from the physical level, and ensures the real-time response of the industrial control system during the backup cycle.

[0019] 2. By using a memory state dictionary to perform in-situ processing of incremental entries and combining the logical importance identifiers determined by the business topology mapping table, a backpressure feedback mechanism with bidirectional constraint capabilities is constructed. When the waiting queue depth of the underlying storage controller reaches the preset deadlock warning threshold, the non-fatal business data load write action is intercepted, and a false write completion confirmation signal is fed back to the upper layer application in real time. This enables industrial control applications to maintain the closed-loop flow of core business logic even under extreme conditions of physical bus saturation, effectively preventing retry storms caused by write timeouts and system-level avalanche-style crashes.

[0020] 3. Because the feature parsing engine directly intercepts the data stream and calculates the content hash value in the kernel input / output stack, the mechanism reduces the high-dimensional file-level comparison task to a lightweight memory pointer comparison and state flag update. This allows the system to consume computing power only during the scattered idle periods of the central processing unit when processing massive, high-frequency concurrent industrial production data. Thus, without changing the underlying storage hardware architecture, it can maximize the use of limited computing resources and provide a near-zero-intrusion high availability guarantee solution for the data base of continuous industrial production lines. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of kernel in-situ sensing and incremental backup in this invention; Figure 2 This is a schematic diagram of the logic block of the intelligent scheduling and back pressure feedback system of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings.

[0023] Example 1: This embodiment discloses an enterprise-level incremental backup method based on content awareness and intelligent scheduling, including the following steps: Step 101, Load kernel filter driver to synchronously intercept write data stream: Load filter driver component in the operating system kernel input / output stack of industrial control node. When industrial control application initiates write request for data block, filter driver component synchronously intercepts data stream before data block enters storage controller queue. Step 102, Calculate the content hash fingerprint and count the frequency of data block changes: The filter driver component calculates the content hash fingerprint of the intercepted data stream and counts the frequency of data block changes within a preset window time. Step 103, Matching the business topology mapping table to establish data importance weight: The filter driver component matches the preset industrial production business topology mapping table according to the file path index carried in the write request to establish the business relevance identifier and importance weight of the data block. Step 104: Monitor real-time load parameters to build an incremental backup task queue: Monitor the CPU utilization, available memory space, and I / O wait time of the industrial control node in real time as real-time load parameters, and combine them with business relevance indicators and importance weights to build an incremental backup task queue containing multiple incremental backup tasks. Step 105, Overload triggers write backpressure interception and write success feedback: When the real-time load parameter exceeds the preset load threshold, the write backpressure logic is triggered, the filter driver component intercepts the physical disk write action against the storage controller, and sends a write success confirmation signal to the industrial control application. Step 106: Incremental merging and disk persistence are completed by cross-node fingerprint comparison: The control virtualization driver component compares data between nodes based on the priority of each incremental backup task in the incremental backup task queue and the content hash fingerprint, and the target storage node completes the merging and disk persistence of the incremental data.

[0024] Step 103 in this embodiment includes the following sub-steps: Step 1031, parse the directory level and file name characteristics corresponding to the file path index; Step 1032, match the parsing results with the process nodes in the industrial production business topology mapping table to determine the production link to which the data block belongs; Step 1033, query the preset coefficients corresponding to the production links in the industrial production business topology mapping table and map the preset coefficients to the business relevance identifier.

[0025] In step 102 of this embodiment, the method for calculating the content hash fingerprint is as follows: call the kernel-mode encryption interface to perform segmented hash operation on the data stream in the non-paged memory area; store the generated content hash fingerprint in the resident lookup table located in the kernel space; compare the hash fingerprint of the current write request with the historical fingerprint in the resident lookup table to identify the data block whose content has changed, and mark the data block as an incremental backup candidate.

[0026] Step 104 in this embodiment includes the following sub-steps: Step 1041, extract the amount of data to be processed corresponding to each incremental backup task in the incremental backup task queue; Step 1042, dynamically adjust the processing concurrency of the incremental backup task queue according to the changing trend of real-time load parameters.

[0027] Following step 105 in this embodiment, the following sub-steps are also included: Step 1051, temporarily storing the data block intercepted by the write backpressure logic in a preset kernel buffer; Step 1052, monitoring the dynamic decline process of the real-time load parameters; Step 1053, when the real-time load parameters are lower than the preset load threshold, releasing the kernel buffer and asynchronously flushing the data block therein to the storage controller.

[0028] In step 101 of this embodiment, the filter driver component achieves in-situ residency in the following way: during the operating system boot phase, the filter driver component is loaded to the top layer of the protocol stack of the storage driver; by hooking the dispatch function of the underlying storage port driver, it takes over the addressing instructions for sector-level offsets.

[0029] Following step 104 in this embodiment, the following sub-steps are also included: Step 1043, calculate the consumption rate of storage resources based on the generation frequency of the incremental backup task queue; Step 1044, when the consumption rate exceeds the preset storage warning value, start the data deduplication logic and use content hash fingerprint to perform cross-node redundant block scanning.

[0030] In step 105 of this embodiment, the operation of sending a confirmation signal of successful write to the industrial control application is only performed on data whose importance weight is lower than a preset safety threshold.

[0031] The process of statistical data block change frequency in step 102 of this embodiment includes: step 1021, establishing a scoring counter indexed by the physical address of the data block; step 1022, whenever a change in the content hash fingerprint of the same physical address is detected, accumulating the score of the scoring counter corresponding to the physical address.

[0032] This embodiment also provides an enterprise-level incremental backup system based on content awareness and intelligent scheduling, including: The data stream interception module is loaded into the operating system kernel input / output stack of the industrial control node. It is used to synchronously intercept the data stream before the data block enters the storage controller queue when the industrial control application initiates a write request for the data block. The feature parsing module is used to calculate the content hash fingerprint of the intercepted data stream and to count the frequency of changes of data blocks within a preset window time. The business association mapping module is used to match the preset industrial production business topology mapping table based on the file path index carried in the write request, and to establish the business association degree identifier and importance weight of the data block. The scheduling and management module is used to monitor the CPU utilization, available memory space, and I / O wait time of industrial control nodes in real time as real-time load parameters. Combined with business relevance identifiers and importance weights, it constructs an incremental backup task queue containing multiple incremental backup tasks. The back pressure control module is used to initiate the write back pressure logic when the real-time load parameters exceed the preset load threshold. The data flow interception module intercepts the physical disk write operation for the storage controller and sends a write success confirmation signal back to the industrial control application to maintain the data flow logic of the industrial control application under the storage bus saturation condition. The data synchronization and merging module is used to compare data between nodes based on the priority of each incremental backup task in the incremental backup task queue and the content hash fingerprint, and control the target storage node to complete the merging and disking of incremental data.

[0033] Example 2: In this embodiment, in an automated production workshop with a large-scale distributed control system, industrial control nodes continuously process sampled data from sensor arrays and write it to the underlying memory. When a sudden change in equipment status occurs on the production line, causing a transient surge in control commands and alarm logs, the underlying bus of the storage controller often suffers from physical resource contention because it cannot simultaneously handle high-frequency data write-to-disk and incremental feature scanning actions. By loading a filter driver component into the input / output stack of the operating system kernel of the industrial control node, when the industrial control application initiates a write request for a data block, the filter driver component synchronously intercepts the data stream before the data block enters the storage controller queue, calculates the content hash fingerprint of the data stream in situ, parses the file path index carried by the write request and matches it with a preset industrial production business topology mapping table, establishes the business relevance identifier and importance weight of the data block, and the in-situ perception mechanism anchors the feature extraction action to the transient path of production business write-to-disk, eliminating the additional read overhead generated by the backup perception process.

[0034] The system monitors the CPU utilization, available memory, and I / O latency of industrial control nodes in real time as real-time load parameters. Combined with the business relevance indicators and importance weights established in previous steps, an incremental backup task queue is constructed. When the I / O latency exceeds a preset load threshold and bus pressure approaches saturation, the system initiates write backpressure logic. This filters and drives the physical disk write operation of low-importance data blocks targeting the storage controller, and sends a standard write success confirmation signal back to the industrial control application. This kernel-mode interception mechanism with state deception addresses the underlying I / O physical limitations. To resolve conflicts between the flow and the upper-layer industrial control business logic, dynamic avoidance of bandwidth for writing critical production instructions to disk is achieved without triggering write timeout exceptions in the industrial control software. After the real-time load parameters fall back to a safe range, the virtualization driver component compares data between nodes based on the priority of each incremental backup task in the incremental backup task queue and the hash fingerprint of the content residing in memory. It then controls the target storage node to complete the merging and writing of incremental data to disk. This method reduces the high-dimensional file-level comparison task to a lightweight memory state tag update, ensuring the logical disaster recovery capability and system stability of core control data under extreme concurrency conditions.

[0035] Example 3: This embodiment uses a distributed physical test platform consisting of 10 industrial control nodes to deploy a Fibre Channel storage area network environment with a bandwidth of 32Gb / s. It simulates the I / O throughput of distributed processing units in a continuous chemical plant production line, verifying the effectiveness of the kernel-level in-situ sensing mechanism in ensuring the latency of core production command writes to disk under transient concurrency surges. The sampling period is [not specified in the original text]. Based on the minimum response interval of the industrial control node to the sensor pulse signal, the sampling period is determined by balancing the CPU cycle loss caused by millisecond-level instruction real-time performance and kernel-mode characteristic calculations. The value is 1ms. When the bus bandwidth utilization of the monitored storage controller is in the high range of over 90%, the value is determined according to the integrity requirements of the signal frequency envelope of the Nyquist sampling theorem to ensure that the statistics of data block change frequency have physical closed-loop characteristics. The system injects Gaussian white noise with a signal-to-noise ratio of 20dB into the input and output stack of the industrial control node to simulate bus transmission jitter caused by electromagnetic interference in the industrial field. The test scheme is divided into a test group using the method of this invention and a control group that removes the write backpressure logic. The filter driver component in the test group performs in-situ interception on the raw data stream that enters the input and output queue in real time. According to the preset industrial production business topology mapping table, the data block containing the production equipment status code is assigned an importance weight of 0.95, and the system image log used for later auditing is assigned an importance weight of 0.15. The content hash fingerprint is calculated synchronously to anchor the data change characteristics. The process is implemented by hooking the dispatch function of the underlying storage port driver. The feature extraction action does not depend on the metadata query of the file system.

[0036] When the simulation experiment enters the peak data concurrency phase, the real-time load parameters... The waiting time increased from the normal 2.4ms to 18.6ms. In the control group, due to the lack of underlying backpressure regulation, the latency of core production command writes simultaneously increased to 45.2ms, exceeding the 30ms safety response threshold set by the industrial control system and inducing a logic retry. The experimental group detected... When the waiting time exceeds the preset load threshold of 15ms, the write backpressure logic is automatically activated. This filters and physically intercepts low-weight log data blocks with an importance weight of 0.15, and sends a standard write success confirmation signal back to the industrial control application. Test data shows that the disk write latency for critical instructions with an importance weight of 0.95 in the experimental group remains at 8.4ms, noise-induced latency jitter is smoothed, and the noise suppression rate reaches 72.5%. This confirms the effectiveness of the state spoofing mechanism in resolving the conflict between physical bus saturation and business logic coherence. To determine the performance boundary of the write backpressure logic, this experiment constructs a gradient comparison system with bus loads ranging from 50% to 120%. Within the 80% range, the data write-to-disk performance of the experimental group and the control group was the same. When the load crossed the 80% performance inflection point and entered the 110% overload zone, the system throughput of the control group decreased due to frequent addressing conflicts, and the critical instruction loss rate rose to 12.8%. However, the experimental group, by dynamically prioritizing the incremental backup task queue and temporarily storing the fingerprint comparison status in memory to delay the actual write-to-disk of low-priority data, kept the throughput of critical services at a saturated output state of over 95% until the load pressure dropped back to the safe range. The nonlinear response characteristics defined the optimal working window of the preset load threshold, ensuring the logical availability and operational stability of industrial data management under the distributed architecture.

[0037] Example 4: In this embodiment, under high-concurrency conditions in a real-time scheduling center, the system calibrates the initial operating environment by conducting a 300-second baseline stress test on industrial control nodes under idle conditions. The test involves cyclically issuing random write commands of 4KB size and statistically analyzing the average... Waiting time and the standard deviation of waiting time According to the formula Determine the preset load threshold, where, To preset the load threshold, For average Waiting time To accommodate the standard deviation of the waiting time, in this approach, the filter driver component is loaded into the storage protocol stack during the operating system kernel startup phase. By hooking the entry address of the dispatch function of the underlying disk driver, it takes over direct access requests to sector offsets, ensuring that all data flows entering the physical addressing phase are controlled. To overcome the architectural barrier that the underlying physical addressing phase cannot directly obtain the semantics of the upper-layer file system, the filter driver component adopts a cross-layer context tracing mechanism. It synchronously registers a callback monitoring program at the virtual file system layer. When the application layer issues a write command containing a file path, the monitoring program intercepts the file path identifier and converts it into a numerical index. Using the reserved stack unit of the input / output request packet, the index is passed down as a custom parameter, enabling the sector interceptor at the protocol layer to accurately identify the sector offset. The system identifies the business affiliation of the current data block. The industrial production business topology mapping table is stored in the kernel's resident memory using a three-level hash index structure. The system intercepts write requests and parses the file path index carried within them. If the path prefix matches the core control database directory of the scheduling center, the corresponding business relevance identifier is set to the highest level, and an importance weight of 0.95 is assigned based on a preset logical correspondence. If the path matches the regular system operation log directory, an importance weight of 0.15 is assigned. This in-situ perception process, combined with real-time monitoring of the central processing unit occupancy and available memory space of the industrial control nodes, dynamically constructs an incremental backup task queue and binds the content hash fingerprint with the importance weight in binary serialization, eliminating the addressing overhead caused by file system metadata queries.

[0038] Under peak concurrency conditions, data throughput and latency events on industrial control buses often exhibit a long-tail distribution. The system first performs a data smoothing preprocessing procedure based on the log-normal distribution principle, extracting the natural logarithm from the collected discrete I / O latency sample values. Based on this, the average latency and standard deviation are calculated. For the specific operating condition of concurrent writing of multiple types of heterogeneous data into a single shared database, a topology mapping table embeds a physical sector distribution benchmark for the database tablespace. The system reads the logical block offset of intercepted data blocks and compares it with the distribution benchmark to identify the business source. During the synchronous feature extraction operation, a non-cryptographic hash algorithm is invoked, and a segmented hash benchmark is set. The computation block size is a standard 4KB operating system memory page, limiting the computational overhead to within the total CPU cycle safety boundary. In kernel mode, the system performs hash calculations by directly calling the hardware-level cryptographic coprocessor interface that supports single-instruction-multiple-data instruction sets. It uses the processor's built-in cache to read data pages in a lock-free circular queue, thereby compressing the computation time of the 4KB data block to an extremely short hardware clock cycle level. This ensures that the resident latency introduced by this feature extraction process is much smaller than the physical addressing response time of the underlying bus, thus preventing congestion or timeouts in the input / output queues due to waiting for CPU calculations.

[0039] When the real-time monitored I / O latency exceeds the preset load threshold calculated in the previous steps... When the filter driver component enables write backpressure logic, it checks the importance weight attribute of each issued I / O request packet. For data blocks with an importance weight lower than 0.30, their physical write operation is intercepted and redirected to a preset kernel circular buffer. The system modifies the return status register of the I / O request packet, forcibly setting its flag to success, and sends a write success confirmation signal back to the application layer software. This method, by simulating the response of physical storage completion in kernel mode, prevents write timeout retry storms triggered by underlying bus congestion in upper-layer applications, and achieves priority release of physical addressing bandwidth. To prevent the interruption of data persistence between the application layer state machine and the physical storage medium caused by abnormal physical power supply interruption, the kernel circular buffer is constructed using NVRAM non-volatile random access memory or memory chips with independent backup capacitor power supply paths. Synchronizing with the return of the write success signal to the application layer, the system establishes a write success confirmation signal in kernel space. The system redirects read requests to non-volatile memory addresses using a mapping record. If an industrial control application initiates a read request for the same data block address, the filtering driver component compares the mapping record and extracts temporary data from the kernel circular buffer, sending it back to the application layer to maintain strong consistency between read and write views. This redirection extraction process is implemented by constructing a logical sector address translation table (LSA) in resident memory. When the underlying layer intercepts a read instruction carrying a logical block address request that matches the key value in the LSA, the system intercepts the addressing action sent to the physical disk and instead performs a direct memory copy from the non-volatile cache medium based on the memory offset pointer corresponding to the key value. The system uses the acquired data stream to reconstruct the complete structure of the read operation in real time and submits it upwards. It monitors the remaining capacity of the logical periodic polling buffer. When the preset physical full load warning line is reached, the system switches to throwing a standard equipment busy code to the industrial control application, triggering application layer process source rate limiting protection. The system detects that the I / O wait time has fallen back to the minimum threshold. After the safe zone is within the specified range, the virtualization driver component activates the asynchronous persistent thread, performs content consistency comparison on the data in the kernel circular buffer based on the content hash fingerprint temporarily stored in memory, and controls the storage controller to merge the incremental data marked as changed and write it to the physical medium. This method reduces the complexity of cross-node file comparison to sequential reading of memory state tags, solves the technical problem of backup tasks competing for real-time instruction bandwidth in distributed industrial control architecture, and ensures that the success rate of writing core control instructions to disk remains above 99.9% even under overload conditions where the bus load instantly reaches 110%, thus guaranteeing the logical integrity of industrial data flow in the distributed disaster recovery architecture.

[0040] Example 5: In this embodiment, in a newly commissioned flexible automobile assembly workshop, the industrial control nodes need to adapt to the changing data storage topology. Before entering the production state, the system initiates a business correlation scanning program. The system traverses all path nodes in the industrial production business topology mapping table and initiates a simulated I / O trigger test lasting 60 seconds for each production business logic line, recording the average data packet length generated under each path. and write frequency The system is based on the formula Calculate the initial weight value for each business node, where, These are the initial weight values. This is the average data packet length. To determine the write frequency, the system maps the calculation results to a range of 0.05 to 0.95, filling the initial matrix of importance weights.

[0041] The system acquires hardware capability descriptors for each industrial control node in real time within a distributed deployment environment. These descriptors include the average seek cycle of the underlying disk. and bus transmission bandwidth When physical hardware performance differences between different nodes are detected, When waiting times are uneven, the filter-driven component activates a weight correction model based on physical capability feedback, according to the formula... The importance weights are dynamically fine-tuned, whereby... The corrected weights, The initial weights established for the preceding steps are all dimensionless constants. This represents the currently measured effective bandwidth. To improve bus transmission bandwidth, the filtering driver component synchronously triggers the write backpressure action of nodes with different physical performance, maintaining the logical consistency and operational stability of the industrial control cluster when facing business concurrency surges. This dynamic fine-tuning mechanism actually constructs a penalty boundary that maps business priorities to physical health status. When a specific node experiences physical hardware frequency reduction, causing a sharp shrinkage in the measured effective bandwidth, the computational weight of all data blocks within it collapses proportionally. This does not erase the logical criticality of the data content itself, but rather uses the result of dimensionality reduction to implicitly raise the scheduling response threshold of congested nodes from the cluster's perspective, thereby enabling the external scheduler to automatically transfer incremental synchronization load to healthy nodes with sufficient bandwidth.

[0042] On electronic component production lines that continuously generate massive amounts of measurement and control results, industrial control nodes dynamically adjust the processing concurrency of incremental backup task queues based on the changing trends of real-time load parameters. When the fluctuation range of the central processing unit utilization is lower than the preset slope and the available memory space is within the preset safe range, the system uses a calculation formula... Increase the number of concurrent threads participating in the backup task, among which, The adjusted number of concurrent threads. Based on the thread constant, According to The margin adjustment coefficient, calculated from the waiting time, prevents backup data from accumulating again in the kernel buffer by increasing the throughput of data processing.

[0043] Example 6: In this embodiment, within an industrial cloud storage cluster processing large-scale high-frequency sensor measurement point data streams, industrial control nodes respond to cross-node incremental state synchronization requests. When a physical storage node experiences address jitter due to hardware failure and the I / O request packet backlog rate exceeds the system emptying rate, the system executes a pre-deployment calibration procedure to determine the replacement benchmark for the memory temporary queue. By simulating the highest concurrent write requests in the initial environment, the system statistically analyzes the allocation success rate and reclamation latency of the kernel's non-paged memory, and determines the overflow threshold of the kernel's circular buffer. The value is set to 0.85, determined based on the inflection point of memory page swapping performance degradation. To determine the occupancy rate, the system simultaneously establishes a hash mapping table for indexing content hash fingerprints; the system detects when the current buffer occupancy rate exceeds the overflow threshold. At that time, through the calculation formula Get the non-access duration of each fingerprint entry in the hash map table, where, Non-access duration This is the current system master clock tick value. This is the beat value at the last time the corresponding fingerprint entry was hit.

[0044] The system retrieves the importance weights of the corresponding business paths from the preset industrial production business topology mapping table. For paths with an importance weight lower than 0.20 and... For expired entries exceeding 1000ms, the filtering driver component performs an in-situ memory overwrite operation, writing the newly arrived high-weight data block information to the corresponding memory address. Under simulated stress testing, this method maintains the queuing latency of production core data with an importance weight of 0.95 at less than 50μs in the hash comparison stage. After the bus load decreases, the virtualization driver component controls the storage controller to complete the incremental data merging and writing to the physical medium. This process achieves coordinated triggering of write backpressure actions on nodes with different physical performance. When the generation frequency of incremental backup task queues causes the storage resource consumption rate to exceed the preset storage warning value, the system initiates a cross-node deduplication procedure. The virtualization driver component extracts the content hash fingerprint of the data block to be processed and performs a search and comparison in the global feature library of the distributed cluster. If it is confirmed that the same fingerprint entry already exists in the global feature library, the virtualization driver component intercepts the physical disk writing action of the data block and establishes a logical mapping pointing to the existing fingerprint at the target address. This method reduces the dynamic load of the storage bus and improves the disaster recovery efficiency of the storage system by suppressing redundant writes at the physical layer.

Claims

1. An enterprise-level incremental backup method based on content awareness and intelligent scheduling, characterized in that, Includes the following steps: Step 101: Load the filter driver component into the operating system kernel input / output stack of the industrial control node. When the industrial control application initiates a write request for a data block, the filter driver component synchronously intercepts the data stream before the data block enters the storage controller queue. Step 102: The filter driver component calculates the content hash fingerprint of the intercepted data stream and counts the frequency of change of the data blocks within a preset window time. Step 103: The filter driver component matches the preset industrial production business topology mapping table with the file path index carried in the write request to establish the business relevance identifier and importance weight of the data block. Step 104: Monitor the CPU utilization, available memory space, and I / O wait time of the industrial control node in real time as real-time load parameters, and combine them with business relevance identifiers and importance weights to construct an incremental backup task queue containing multiple incremental backup tasks. Step 105: When the real-time load parameter exceeds the preset load threshold, the write back pressure logic is started, the filter drive component intercepts the physical disk write action for the storage controller, and sends a write success confirmation signal to the industrial control application. Step 106: The control virtualization driver component compares data between nodes based on the priority of each incremental backup task in the incremental backup task queue and the content hash fingerprint, and the target storage node completes the merging and disking of incremental data.

2. The enterprise-level incremental backup method based on content awareness and intelligent scheduling according to claim 1, characterized in that, Step 103 includes the following sub-steps: Step 1031, parse the directory level and file name characteristics corresponding to the file path index; Step 1032, match the parsing results with the process nodes in the industrial production business topology mapping table to determine the production link to which the data block belongs. Step 1033: Query the preset coefficients corresponding to the production links in the industrial production business topology mapping table, and map the preset coefficients to business relevance identifiers.

3. The enterprise-level incremental backup method based on content awareness and intelligent scheduling according to claim 1, characterized in that, The method for calculating the content hash fingerprint in step 102 is as follows: call the kernel-mode encryption interface to perform segmented hash operation on the data stream in the non-paged memory area; and store the generated content hash fingerprint in the resident lookup table located in the kernel space. By comparing the hash fingerprint of the current write request with the historical fingerprint in the resident lookup table, the data blocks whose content has changed are identified and marked as incremental backup candidates.

4. The enterprise-level incremental backup method based on content awareness and intelligent scheduling according to claim 1, characterized in that, Step 104 includes the following sub-steps: Step 1041, extract the amount of data to be processed corresponding to each incremental backup task in the incremental backup task queue; Step 1042, dynamically adjust the processing concurrency of the incremental backup task queue according to the changing trend of real-time load parameters.

5. The enterprise-level incremental backup method based on content awareness and intelligent scheduling according to claim 1, characterized in that, After step 105, the following sub-steps are also included: Step 1051, temporarily storing the data blocks intercepted by the write backpressure logic in a preset kernel buffer; Step 1052, monitoring the dynamic decline process of the real-time load parameters; Step 1053, when the real-time load parameters are lower than the preset load threshold, releasing the kernel buffer and asynchronously flushing the data blocks therein to the storage controller.

6. The enterprise-level incremental backup method based on content awareness and intelligent scheduling according to claim 1, characterized in that, In step 101, the filter driver component achieves in-situ residency in the following way: during the operating system boot phase, the filter driver component is loaded to the top layer of the storage driver's protocol stack; By hooking the dispatch function of the underlying storage port driver, it takes over the addressing instructions for sector-level offsets.

7. The enterprise-level incremental backup method based on content awareness and intelligent scheduling according to claim 1, characterized in that, After step 104, the following sub-steps are also included: Step 1043, calculate the consumption rate of storage resources based on the generation frequency of the incremental backup task queue; Step 1044, when the consumption rate exceeds the preset storage warning value, start the data deduplication logic and use content hash fingerprint to perform cross-node redundant block scanning.

8. The enterprise-level incremental backup method based on content awareness and intelligent scheduling according to claim 1, characterized in that, In step 105, the operation of sending a successful write confirmation signal to the industrial control application is only performed on data whose importance weight is lower than a preset safety threshold.

9. The enterprise-level incremental backup method based on content awareness and intelligent scheduling according to claim 1, characterized in that, The process of counting the frequency of changes in data blocks in step 102 includes: step 1021, establishing a scoring counter indexed by the physical address of the data block; step 1022, whenever a change in the content hash fingerprint of the same physical address is detected, accumulating the score of the scoring counter corresponding to the physical address.

10. An enterprise-level incremental backup system based on content awareness and intelligent scheduling, used to implement the enterprise-level incremental backup method based on content awareness and intelligent scheduling as described in claim 1, characterized in that, include: The data stream interception module is loaded into the operating system kernel input / output stack of the industrial control node. It is used to synchronously intercept the data stream before the data block enters the storage controller queue when the industrial control application initiates a write request for the data block. The feature parsing module is used to calculate the content hash fingerprint of the intercepted data stream and to count the frequency of changes of data blocks within a preset window time. The business association mapping module is used to match the preset industrial production business topology mapping table based on the file path index carried in the write request, and to establish the business association degree identifier and importance weight of the data block. The scheduling and management module is used to monitor the CPU utilization, available memory space, and I / O wait time of industrial control nodes in real time as real-time load parameters. Combined with business relevance identifiers and importance weights, it constructs an incremental backup task queue containing multiple incremental backup tasks. The back pressure control module is used to initiate the write back pressure logic when the real-time load parameters exceed the preset load threshold. The data flow interception module intercepts the physical disk write operation for the storage controller and sends a write success confirmation signal back to the industrial control application to maintain the data flow logic of the industrial control application under the storage bus saturation condition. The data synchronization and merging module is used to compare data between nodes based on the priority of each incremental backup task in the incremental backup task queue and the content hash fingerprint, and control the target storage node to complete the merging and disking of incremental data.

Images