A Container Image Pre-distribution Method and System Based on Semantic Layering and Cooperative Scheduling

Abstract

This invention discloses a container image pre-distribution method and system based on semantic layering and collaborative scheduling, relating to the field of edge computing technology. The method performs semantic parsing of container images, constructing a layered structure of basic, commonly used, and dynamic layers based on component dependencies, change frequency, and business function modules. It collects edge node computing resource usage, storage usage, network connectivity, and image layer hit history to construct a node state feature set and generates a comprehensive node score based on a weighted scoring model. Based on the comprehensive node score and image layer category, a suitable pre-distribution path is selected from local reuse, regional caching, hierarchical index nodes, and edge node point-to-point transmission paths. A caching strategy of long-term retention, usage frequency management, or on-demand replacement is implemented according to the image layer category, achieving differentiated storage and reuse of image layers. This invention improves image pre-distribution efficiency and loading stability, and is applicable to various edge computing deployment scenarios.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, specifically to a container image pre-distribution method and system based on semantic layering and collaborative scheduling. Background Technology

[0002] With the rapid development of edge computing in scenarios such as industrial internet, video processing, and intelligent control, container technology is gradually becoming the main way to deploy and run applications on edge nodes. Container images are the basic unit for application loading and task scheduling, and their acquisition speed and distribution efficiency directly affect the node's startup performance and resource utilization. Existing image distribution usually relies on a central repository or regional cache nodes, and reduces redundant transmissions through image layer reuse to support multi-node deployment requirements.

[0003] Existing image layering methods are mostly based on fixed strategies of image building tools, making it difficult to reflect the dependencies, change frequency, and differences in business functions among components within an image. The layering results often fail to correspond to the actual needs of edge nodes. The resource status, network conditions, and access behavior of edge nodes are highly dynamic, but node evaluation often relies on static or single metrics, making it difficult to accurately reflect the node's capacity to handle image layers under different operating conditions. Image distribution paths are typically selected based on fixed priorities or simple reachability rules, making it difficult to comprehensively assess path quality by considering node status changes, differences in image layer types, and network conditions. Edge node cache management generally employs a uniform replacement strategy, failing to differentiate between image layer stability, access frequency, and change characteristics, easily leading to insufficient cache space utilization or premature cleanup of important image layers.

[0004] Therefore, there is an urgent need for an image pre-distribution scheme that can form a collaborative processing capability in image layering, node status analysis, path selection and cache management, so as to improve the overall distribution performance of images in edge environments. Summary of the Invention

[0005] The purpose of this invention is to provide a container image pre-distribution method and system based on semantic layering and collaborative scheduling to solve the limitations of existing image layering methods, such as fixed methods, inaccurate node evaluation, lack of dynamic path selection, and single cache management. This invention achieves a higher degree of matching between image layering results and node requirements, a more reasonable selection of image transmission paths, and more efficient utilization of edge node cache resources, thereby improving the distribution efficiency and loading stability of container images in edge environments.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] On the one hand, this invention provides a container image pre-distribution method based on semantic layering and cooperative scheduling, the method comprising:

[0008] Step 1: Perform semantic parsing on the container image and build an image layering structure based on the component dependencies, change frequency, and business function modules of the container image. The image layering structure includes three image layer categories: base layer, common layer, and dynamic layer.

[0009] Step 2: Collect the resource status and mirror layer hit history of edge nodes, and generate a comprehensive node score based on a weighted scoring model. The comprehensive node score is used to characterize the degree of adaptation of edge nodes to receive mirror layers.

[0010] Step 3: Based on the node's comprehensive score and the mirror layer's category, select one or more paths from a predefined set of pre-distributed paths for transmitting the mirror layer. The set of pre-distributed paths includes local multiplexing paths, regional cache node acquisition paths, hierarchical index node acquisition paths, and point-to-point transmission paths between edge nodes.

[0011] Step 4: Transmit the image layer to the target edge node according to the selected pre-distribution path, and perform corresponding long-term retention management, usage frequency management or on-demand replacement management on the target edge node according to the image layer category to which the image layer belongs.

[0012] As a preferred embodiment of the present invention, the parsing process for constructing the mirrored hierarchical structure includes:

[0013] Perform structured analysis on the component collection in the container image and identify component dependencies based on the call relationship graph between components;

[0014] Determine the frequency of component changes based on the change history of the container image in historical versions;

[0015] Determine the correlation between components and business function modules based on the business function module division rules of the application to which the container image belongs;

[0016] A set of component attributes is constructed based on component dependencies, change frequency, and correlation between business function modules. Component stability indicators and component business coupling indicators are then calculated based on the set of component attributes.

[0017] A hierarchical determination result is generated based on the component stability index and the component business coupling index. The hierarchical determination result is used to divide the components into a basic layer component set, a commonly used layer component set, and a dynamic layer component set.

[0018] Generate corresponding base layer, common layer, and dynamic layer based on the base layer component set, the common layer component set, and the dynamic layer component set.

[0019] As a preferred embodiment of the present invention, the process of collecting the resource status of edge nodes and the hit history of mirror layers includes:

[0020] Obtain information on edge node computing resource usage, storage resource usage, network connectivity, and image layer hit history;

[0021] A set of node state features is constructed based on the resource fluctuation characteristics of edge nodes, the time series characteristics of mirror layer access, the changing trend of network connectivity, and the resource status of adjacent nodes.

[0022] The operating mode of edge nodes is identified based on the set of node status characteristics, and the future resource availability indicators of the nodes are determined based on the operating mode.

[0023] As a preferred embodiment of the present invention, the process of generating a node comprehensive score based on a weighted scoring model includes:

[0024] Based on the node state feature set, a scoring factor is determined to reflect the degree of resource bottleneck, network connectivity stability and mirror layer reuse potential, and a sensitive weight related to the mirror layer category is set for the scoring factor.

[0025] The sensitive weights and basic weights of the scoring factors are dynamically adjusted according to the node operation mode and business load status, and a nonlinear combination is established among the scoring factors to characterize the synergistic influence relationship.

[0026] The node state feature set is input into the weighted scoring model to generate a node comprehensive score, which is used to indicate the bearer adaptability of the target edge node when performing pre-distribution of different mirror layers.

[0027] As a preferred embodiment of the present invention, the process of selecting one or more paths for transmitting the image layer from a predefined set of pre-distributed paths based on the node's comprehensive score and the image layer category to which the image layer belongs includes:

[0028] Construct node state feature sequences and path feature sequences of pre-distributed paths;

[0029] Calculate the sequence deviation based on the node state feature sequence and the path feature sequence of the pre-distributed path. The sequence deviation Calculated by the following formula:

[0030] ;

[0031] In the formula: Represents the sequence of node state features; Indicates the pre-distribution path Path feature sequence; Represents the feature sequence of node states With pre-distribution path The sequence difference operator after indexing and aligning the path feature sequences; Represents the sequence deviation metric operator; An index representing the pre-distribution path;

[0032] Based on sequence deviation Calculate path difference energy The path difference energy Calculated by the following formula:

[0033] ;

[0034] In the formula: Indicates the category of the image layer to which the image layer belongs; This represents the category mapping index corresponding to the category of the mirror layer; Indicates the pre-distribution path Path difference energy;

[0035] Energy based on path differences Calculate the feasibility value of the path The path feasibility value Calculated by the following formula:

[0036] ;

[0037] In the formula: Indicates the pre-distribution path The path feasibility value; This represents the natural exponential function.

[0038] As a preferred embodiment of the present invention, the process of selecting a pre-distribution path based on the path feasibility value includes:

[0039] The set of pre-distribution paths is selected based on the path feasibility value of each pre-distribution path according to the following formula:

[0040] ;

[0041] In the formula: This represents the final set of pre-distributed paths selected. This represents the maximum value among all path feasibility values; Logical OR operator; Indicates the feasibility threshold; This represents the index used to enumerate all pre-distributed paths.

[0042] As a preferred embodiment of the present invention, before performing the corresponding long-term retention management, usage frequency management, or on-demand replacement management at the target edge node, the following steps are included:

[0043] The cache management mode corresponding to the image layer is determined based on the image layer category to which the image layer belongs;

[0044] Select the cache region in the target edge node based on the cache management mode corresponding to the mirror layer.

[0045] As a preferred embodiment of the present invention, the process of performing corresponding long-term retention management, usage frequency management, or on-demand replacement management at the target edge node includes:

[0046] When the image layer belongs to the base layer category, the image layer is stored in the long-term storage area of ​​the target edge node, and a set of stable versions of the base layer image is maintained in the long-term storage area.

[0047] When the mirror layer belongs to the common layer category, the set of common layer mirrors in the common access cache area of ​​the target edge node is adjusted according to the changes in the access frequency of the mirror layer in the target edge node.

[0048] When the mirror layer belongs to the dynamic layer category, the dynamic layer mirror set in the dynamic storage area of ​​the target edge node is updated according to the most recent access information of the mirror layer in the target edge node.

[0049] On the other hand, the present invention also provides a container image pre-distribution system based on semantic layering and cooperative scheduling, applied to the container image pre-distribution method based on semantic layering and cooperative scheduling as described above, the system comprising:

[0050] The semantic parsing module is used to perform semantic parsing on container images and build an image layering structure based on the component dependencies, change frequency, and business function modules of the container image. The image layering structure includes three image layer categories: basic layer, common layer, and dynamic layer.

[0051] The node evaluation module is used to collect the resource status and mirror layer hit history of edge nodes, and generate a comprehensive node score based on a weighted scoring model. The comprehensive node score is used to characterize the degree of adaptation of the edge node to receive the mirror layer.

[0052] The path selection module is used to select one or more paths for transmitting the image layer from a predefined set of pre-distributed paths based on the node's comprehensive score and the image layer category to which the image layer belongs. The set of pre-distributed paths includes local multiplexing paths, regional cache node acquisition paths, hierarchical index node acquisition paths, and point-to-point transmission paths between edge nodes.

[0053] The image distribution and cache management module is used to transmit the image layer to the target edge node according to the selected pre-distribution path and perform long-term retention management, usage frequency management, or on-demand replacement management on the target edge node according to the image layer category to which the image layer belongs.

[0054] The present invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being implemented when executed by a processor.

[0055] The beneficial effects of this invention are as follows: Through the above technical solution, the container image distribution process can be more closely matched with the semantic characteristics of the components within the image and the operating status of the edge nodes. The image layering method based on component dependencies, change frequency, and business function modules makes the division of the basic layer, commonly used layer, and dynamic layer more consistent with the stability and usage characteristics of the image itself, which is beneficial to improving the reuse efficiency of the image layer. The node comprehensive score generated based on node evaluation can reflect the dynamic changes of nodes in terms of resource consumption, network connectivity, and access behavior, making the selection of transmission targets for the image layer more accurate. Combining the node comprehensive score and the image layer category for path selection helps to obtain a more reasonable transmission method under multi-path conditions, improving the stability of image acquisition. After the image layer reaches the edge node, a differentiated cache management strategy is adopted according to the image layer category, so that strategies such as long-term retention, access-driven adjustment, and on-demand replacement are applied to different categories of image layers, thereby improving the utilization efficiency of edge node cache resources and improving image loading performance. This invention can enhance the efficiency and robustness of image distribution in complex edge environments, improve the system's business responsiveness and resource utilization quality. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0057] Figure 1 This is a flowchart of the method of the present invention;

[0058] Figure 2 This is a schematic diagram of the modular structure of the system of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0060] like Figure 1 As shown in this embodiment, the container image pre-distribution method first performs image semantic parsing to obtain the structural relationships and attribute information of the components inside the image.

[0061] Step 1: Perform semantic parsing on the container image and build an image layering structure based on the component dependencies, change frequency, and business function modules of the container image. The image layering structure includes three image layer categories: base layer, common layer, and dynamic layer.

[0062] In practice, the semantic parsing process can be understood as: conducting a comprehensive analysis of various software components in the container image to restore the calling relationships and functional organization structure between components, and further subdividing the image into different hierarchical structures based on the component update status and business attributes.

[0063] To implement the above-described image layering construction process, the semantic parsing includes the following parts:

[0064] 1. Perform a structured analysis of the component collection in the container image, identifying component dependencies based on the call relationship graph. In practice, this can begin by traversing and parsing the image file system, constructing a call relationship graph based on the component loading order, function call information, or configuration dependencies. In the call relationship graph, boundary nodes represent components, and edges represent the direction of dependencies between components. The purpose of this step is to determine the degree of structural association between components, so as to distinguish the importance and coupling strength of components in subsequent layering.

[0065] 2. Determine the component change frequency based on the container image's change history. Specifically, this can be achieved by extracting update logs for each component from image build records, version control systems, or image repositories to track the number of updates, update intervals, and update magnitude. Change frequency reflects the component's stability characteristics; components with fewer updates typically belong to core, fundamental functionalities and are more suitable as foundational content.

[0066] 3. Determine the correlation between components and business function modules based on the application's functional module division rules. In practice, image components can be tagged or categorized according to the application's functional module definitions (e.g., network processing module, data processing module, business logic module, etc.). By analyzing the component's calls to business modules, configuration binding relationships, and runtime behavior, the correlation between components and business function modules can be quantified, thereby determining the component's functional positioning.

[0067] 4. Construct a set of component attributes based on component dependencies, change frequency, and the relevance of business function modules. Calculate component stability and business coupling metrics based on these attribute sets. In practice, a set of attributes containing multi-dimensional characteristics can be constructed for each component, such as dependency depth, update frequency, and functional coupling. Based on these attributes, component stability and business coupling metrics can be further derived.

[0068] 5. Generate a hierarchical determination result based on component stability indicators and component business coupling indicators. This result is used to divide components into a basic layer set of components, a frequently used layer set of components, and a dynamic layer set of components. In this embodiment, components can be clustered or partitioned in a two-dimensional space based on their stability and business coupling degree. For example, components with high stability and high business coupling are classified into the basic layer; components with high business access frequency but some changes are classified into the frequently used layer; and components with high update frequency, relatively frequent changes, or low coupling with core business are classified into the dynamic layer. The hierarchical determination result describes the category attribute of each component.

[0069] 6. Generate the corresponding basic layer, common layer, and dynamic layer based on the basic layer component set, the common layer component set, and the dynamic layer component set.

[0070] In the final stage, corresponding image layer content is constructed based on each type of component set. During the image layer construction process, component files, metadata, and their runtime dependencies belonging to the same category can be merged to form logically independent image layers. The generated three types of image layer structures provide the basis for subsequent image distribution, scheduling, and caching strategies.

[0071] Through the above semantic parsing and attribute analysis process, container images are divided into three types of image layer structures: basic layer, common layer, and dynamic layer. Different image layers have fundamental differences in terms of stability, access frequency, and business relevance, which is conducive to achieving fine-grained scheduling and differentiated distribution by layer in subsequent steps.

[0072] In this embodiment, to achieve reasonable scheduling of edge nodes during the image pre-distribution process, it is necessary to quantitatively evaluate the current and recent operating status of the nodes to determine whether the nodes are suitable for receiving content from different image layers. Therefore, this embodiment performs two stages before image distribution: node status collection and node comprehensive score generation.

[0073] Step 2: Collect the resource status and mirror layer hit history of edge nodes, and generate a comprehensive node score based on a weighted scoring model. The comprehensive node score is used to characterize the degree of adaptation of edge nodes to receive mirror layers.

[0074] In practice, the node status collection process can be understood as: collecting real-time information on the node's computing power, storage space, network connection quality, and mirror access behavior, and organizing this information into structured feature data to provide input for the subsequent scoring model.

[0075] The node status collection may include the following:

[0076] 1. Obtain information on edge node computing resource usage, storage resource usage, network connectivity, and image layer hit history. During implementation, metrics such as node CPU utilization, memory usage, storage space usage, network round-trip latency, and packet loss rate can be obtained in real time from sources including the container runtime environment, system monitoring metrics, and edge node agents. Image layer hit history can be obtained from the node's local image repository or cache module, reflecting the node's recent access behavior to different image layers.

[0077] 2. A node status feature set is constructed based on the resource fluctuation characteristics of edge nodes, the time series characteristics of mirror layer access, the changing trend of network connectivity, and the resource status of adjacent nodes. In this embodiment, the original resource indicators are further structured, including: calculating the fluctuation amplitude and trend of node resource occupancy, statistically analyzing the time series patterns of mirror layer access frequency and access interval, and analyzing the periodic changing trend of network connectivity. Simultaneously, based on the topological relationship between nodes, the resource status of adjacent nodes can be synchronously obtained to reflect the collaborative load situation of nodes in a local area.

[0078] 3. Identify the operating mode of edge nodes based on the node status feature set, and determine the future resource availability indicators of the nodes based on the operating mode.

[0079] This embodiment can identify different operating modes of nodes, such as stable operation, high load fluctuation, and periodic jitter, by statistically analyzing the resource characteristics of nodes at different time periods. There is a strong correlation between the operating mode and the future resource availability of a node; therefore, the node's ability to process new image layers in the coming period can be inferred from this.

[0080] After constructing the node state feature set, these features need to be input into the scoring model to quantify the degree to which a node adapts to different image layers. The process is as follows:

[0081] 1. Based on the node status feature set, determine the scoring factors to reflect the degree of resource bottleneck, network connectivity stability, and mirror layer reuse potential, and assign sensitive weights related to the mirror layer category to the scoring factors. In this embodiment, the scoring factors can be derived from multiple features such as node CPU peak usage trend, changes in available memory space, network latency stability, and mirror layer hit time series patterns. Different mirror layer categories (basic layer, commonly used layer, dynamic layer) have different sensitivities to node resources and access patterns. Therefore, it is necessary to introduce sensitive weights related to the layer category to the scoring factors to reflect the differentiated needs of each mirror layer in the pre-distribution process.

[0082] 2. The sensitive and basic weights of the scoring factors are dynamically adjusted based on the node's operating mode and service load status, and a nonlinear combination is established among the scoring factors to characterize the synergistic influence relationship. This embodiment can dynamically adjust the contribution of the scoring factors based on the node's operating mode, such as stable mode or load fluctuation mode, on the basis of the basic sensitive weights, enabling the model to more accurately reflect the node's capacity to support the mirror layer under different states. There may be synergistic relationships between the scoring factors; for example, increased CPU fluctuations are often accompanied by network jitter. Therefore, this embodiment uses a nonlinear combination method to enhance the overall expressive power of the scoring model.

[0083] 3. Input the node state feature set into the weighted scoring model to generate a node comprehensive score. The node comprehensive score is used to indicate the bearer adaptability of the target edge node when performing pre-distribution of different mirror layers.

[0084] The scoring model outputs a comprehensive node score, indicating whether a node is suitable for hosting different types of mirror layers under current and projected future conditions. A higher comprehensive node score indicates a stronger adaptability to receiving mirror layers.

[0085] Through the above steps, the node comprehensive score fully reflects the node's resource capabilities, operational trends, access characteristics, and business needs, and can be used to make decisions on subsequent image distribution paths.

[0086] After completing the comprehensive node evaluation, this embodiment further selects a suitable pre-distribution path for transmitting the corresponding mirror layer based on the comprehensive evaluation results of the nodes and the mirror layer category. To achieve scientific and adaptive path selection, this embodiment introduces a comparative analysis mechanism between node state feature sequences and path feature sequences. By calculating the degree of difference between the two, the feasibility of the path is deduced and the final decision is made.

[0087] Step 3: Based on the node's comprehensive score and the mirror layer's category, select one or more paths from a predefined set of pre-distributed paths for transmitting the mirror layer. The set of pre-distributed paths includes local multiplexing paths, regional cache node acquisition paths, hierarchical index node acquisition paths, and point-to-point transmission paths between edge nodes.

[0088] In its implementation, path selection relies not only on the overall node score but also on the mirror layer category and the path's own attributes, determining the optimal path through a series of calculation steps. This embodiment performs path suitability analysis as follows:

[0089] 1. Construct the node state feature sequence and the path feature sequence of the pre-distributed path;

[0090] The process of selecting one or more paths for transmitting the image layer from a predefined set of pre-distributed paths based on the node's comprehensive score and the image layer category to which the image layer belongs includes: constructing a node state feature sequence and a path feature sequence of the pre-distributed paths;

[0091] In implementation, node status characteristics (such as resource fluctuations, connectivity changes, access patterns, etc.) can be organized into a node status characteristic sequence in chronological order. At the same time, the connectivity, stability, reachability and other attributes of each path can be organized into a path characteristic sequence in chronological order, which facilitates sequence-level difference analysis.

[0092] 2. Calculate the deviation between the node state feature sequence and the path feature sequence.

[0093] Calculate the sequence deviation based on the node state feature sequence and the path feature sequence of the pre-distributed path. The sequence deviation Calculated by the following formula:

[0094] ;

[0095] In the formula: Represents the sequence of node state features; Indicates the pre-distribution path Path feature sequence; Represents the feature sequence of node states With pre-distribution path The sequence difference operator after indexing and aligning the path feature sequences; Represents the sequence deviation metric operator; An index representing the pre-distribution path;

[0096] In practice, the alignment operation allows path features and node states to be compared on the same time scale, thus obtaining difference results that are closer to the actual operation.

[0097] 3. Calculate the path difference energy

[0098] Based on sequence deviation Calculate path difference energy The path difference energy Calculated by the following formula:

[0099] ;

[0100] In the formula: Indicates the category of the image layer to which the image layer belongs; This represents the category mapping index corresponding to the category of the mirror layer; Indicates the pre-distribution path The path difference energy is calculated based on the different image layer categories. In practice, these categories exhibit varying sensitivities to path distribution. For instance, the base layer is more sensitive to transmission stability, while the dynamic layer is more sensitive to transmission latency. Through exponential mapping, the influence of image layer categories can be directly injected into the path difference energy, making path evaluation more aligned with the actual transmission needs of each image layer category.

[0101] 4. Calculate the feasibility value of the path

[0102] Energy based on path differences Calculate the feasibility value of the path The path feasibility value Calculated by the following formula:

[0103] ;

[0104] In the formula: Indicates the pre-distribution path The path feasibility value; This represents the natural exponential function. In this embodiment, the smaller the difference energy, the better the match between the path and the node state, and the higher the feasibility value obtained. The closer a value is to 1, the greater the difference energy and the smaller the feasibility value. This function achieves a natural mapping from difference to feasibility.

[0105] After calculating the feasibility values ​​of all paths, this embodiment selects one or more paths for the transmission of the mirror layer according to a preset strategy.

[0106] The process of selecting a pre-distribution path based on the path feasibility value includes:

[0107] The set of pre-distribution paths is selected based on the path feasibility value of each pre-distribution path according to the following formula:

[0108] ;

[0109] In the formula: This represents the final set of pre-distributed paths selected. This represents the maximum value among all path feasibility values; Logical OR operator; Indicates the feasibility threshold; This represents an index that enumerates all pre-distributed paths. In practical applications, if the feasibility value of a certain path reaches the highest value, then that path can be directly selected as the priority transmission path; if multiple paths meet the feasibility threshold, then multiple paths can be selected for parallel distribution according to business needs to improve the efficiency and reliability of mirror transmission.

[0110] Through the above process, step 3 effectively associates the node comprehensive score, mirror layer category, path attributes with the actual operating environment, thereby realizing dynamic and refined control of path selection.

[0111] After selecting the pre-distribution path, the image layer needs to be transmitted to the target edge node, where a caching management strategy appropriate to the image layer category is executed. This embodiment ensures the long-term availability, reusability, and dynamic adaptability of the image layer on the edge node through a differentiated caching management mode.

[0112] Step 4: Transmit the image layer to the target edge node according to the selected pre-distribution path, and perform corresponding long-term retention management, usage frequency management or on-demand replacement management on the target edge node according to the image layer category to which the image layer belongs.

[0113] In practice, the system performs mirror layer transfer operations based on the path selected in step 3. For example, when the path set includes a local multiplexed path, the mirror layer can be read directly from the local cache; if a regional cache node path or an index node path is selected, the corresponding layer data is pulled from the corresponding node via a network connection. For multi-path scenarios, parallel prefetching or priority path strategies can be used to reduce transmission latency.

[0114] To ensure that the mirror layer can be properly managed according to its characteristics after reaching the target edge node, this embodiment first determines the cache management mode that the mirror layer should adopt.

[0115] Before performing the corresponding long-term retention management, usage frequency management, or on-demand replacement management on the target edge node, the following steps are included:

[0116] The cache management mode corresponding to the image layer is determined based on the image layer category to which the image layer belongs;

[0117] Select the cache region in the target edge node based on the cache management mode corresponding to the mirror layer.

[0118] In actual implementation, the system determines the cache management mode to be adopted based on the category of the image layer (basic layer, frequently used layer, or dynamic layer): basic layer → long-term retention mode; frequently used layer → usage frequency management mode; dynamic layer → on-demand replacement management mode. Based on the mode difference, the system selects different cache areas in the local storage resources of the target node, such as long-term storage area, frequently used cache area, or dynamic cache area, to ensure that subsequent management strategies can be effectively executed.

[0119] After the image layer is stored in the corresponding cache area, the system executes the corresponding management strategy according to the image layer category to ensure that the availability and update rhythm of the image layer meet business needs.

[0120] The process of performing corresponding long-term retention management, usage frequency management, or on-demand replacement management at the target edge node includes:

[0121] When the image layer belongs to the base layer category, it is stored in the long-term storage area of ​​the target edge node, and a stable set of base layer image versions is maintained in the long-term storage area. In implementation, the base layer usually contains key components with high internal stability and low update frequency, so they are written to the long-term storage area to avoid frequent replacements. At the same time, by maintaining a stable set of base layer image versions, it is ensured that the system always maintains a runnable basic environment, so that even if the image is upgraded or the node is restarted, the basic operating capabilities are not affected.

[0122] When a mirrored layer belongs to the frequently accessed layer category, the set of frequently accessed layer mirrors in the frequently accessed cache area of ​​the target edge node is adjusted according to the changes in the access frequency of the mirrored layer at the target edge node. In implementation, the system periodically collects access records of frequently accessed layer mirrors and adjusts the cache space according to the access frequency. For example, frequently accessed layer mirrors with high frequency of access are retained first; frequently accessed layer mirrors with decreasing access frequency can be gradually migrated or replaced. This access-driven management approach makes the cache area more aligned with actual business needs and improves the reuse efficiency of the mirrored layer.

[0123] When a mirror layer belongs to the dynamic layer category, the dynamic layer mirror set in the dynamic storage area of ​​the target edge node is updated based on the most recent access information of the mirror layer in the target edge node. Dynamic layers contain components with high update frequency or low coupling with business logic, and their usage characteristics include short lifecycles and high variability. Therefore, in the implementation, by tracking the most recent access information of dynamic layer mirrors, the system can replace mirror layers that have not been accessed for a long time or have been accessed infrequently, maintaining the agility and availability of the dynamic cache area.

[0124] This strategy can prevent dynamic layer images from accumulating indefinitely and occupying cache space, while ensuring that the latest version of the dynamic layer image can be loaded quickly when the node actually needs it.

[0125] Through the aforementioned differentiated caching management strategy, target edge nodes can adopt targeted management methods based on image layer categories, ensuring stable availability of the base layer, efficient reuse of frequently used layers, and flexible updates of dynamic layers, thereby significantly improving image loading efficiency and node resource utilization. Simultaneously, this strategy guarantees the rapid startup and resource scheduling of subsequent business tasks.

[0126] like Figure 2 As shown, this embodiment provides a container image pre-distribution system applied to the aforementioned method embodiments. This system is based on a modular design, dividing functions such as image semantic parsing, node evaluation, path selection, and cache management into independent modules to facilitate deployment, expansion, and maintenance.

[0127] This invention also provides a container image pre-distribution system based on semantic layering and cooperative scheduling, applied to the container image pre-distribution method based on semantic layering and cooperative scheduling described above. The system includes:

[0128] The semantic parsing module is used to perform semantic parsing on container images and build an image layering structure based on the component dependencies, change frequency, and business function modules of the container image. The image layering structure includes three image layer categories: basic layer, common layer, and dynamic layer.

[0129] The node evaluation module is used to collect the resource status and mirror layer hit history of edge nodes, and generate a comprehensive node score based on a weighted scoring model. The comprehensive node score is used to characterize the degree of adaptation of the edge node to receive the mirror layer.

[0130] The path selection module is used to select one or more paths for transmitting the image layer from a predefined set of pre-distributed paths based on the node's comprehensive score and the image layer category to which the image layer belongs. The set of pre-distributed paths includes local multiplexing paths, regional cache node acquisition paths, hierarchical index node acquisition paths, and point-to-point transmission paths between edge nodes.

[0131] The image distribution and cache management module is used to transmit the image layer to the target edge node according to the selected pre-distribution path and perform long-term retention management, usage frequency management, or on-demand replacement management on the target edge node according to the image layer category to which the image layer belongs.

[0132] To facilitate software-based deployment, the present invention also provides a computer-readable storage medium for implementing the above method.

[0133] A computer-readable storage medium having a computer program stored thereon, the computer program being implemented when executed by a processor. The storage medium may be a solid-state drive (SSD), a portable hard drive, flash memory, read-only memory (ROM), random access memory (RAM), optical disc (CD-ROM), or other media capable of storing computer program instructions.

[0134] When the above computer program is deployed on a computing device containing a processor and executed, the semantic layering, node evaluation, path selection and cache management operations of the container image can be performed according to steps 1 to 4, thereby realizing the container image pre-distribution method proposed in this invention.

[0135] In summary, this invention forms a complete technical chain in aspects such as image semantic layering, node evaluation, path selection, and cache management, with clear logic and close collaboration between each step. This specification has fully and clearly described the technical solutions and implementation methods of this invention, enabling those skilled in the art to implement all the contents of this invention. This invention is applicable to various edge computing deployment scenarios and has good engineering feasibility and application promotion value.

[0136] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0137] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0138] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A container image pre-distribution method based on semantic layering and cooperative scheduling, characterized in that, The method comprises: Performing semantic analysis on the container image, constructing an image layer structure according to the component dependency relationship, change frequency and business function module of the container image, the image layer structure comprising three image layer categories of a base layer, a common layer and a dynamic layer; Collecting resource state and image layer hit history of the edge node, and generating a node comprehensive score based on a weighted scoring model, the node comprehensive score being used to represent the adaptation degree of the edge node to receive the image layer; According to the node comprehensive score and the image layer category to which the image layer belongs, selecting one path or multiple paths for transmitting the image layer from a pre-defined pre-distribution path set, the pre-distribution path set comprising a local multiplexing path, a regional cache node acquisition path, a layered index node acquisition path and an edge node-to-edge node point-to-point transmission path; Transmitting the image layer to the target edge node according to the selected pre-distribution path, and performing corresponding long-term retention management, usage frequency management or on-demand replacement management at the target edge node according to the image layer category to which the image layer belongs; The process of selecting one path or multiple paths for transmitting the image layer from the pre-defined pre-distribution path set according to the node comprehensive score and the image layer category to which the image layer belongs comprises: Constructing a node state feature sequence and a path feature sequence of the pre-distribution path; A sequence deviation amount is calculated based on a node state feature sequence and a path feature sequence of a pre-dispatch path , the sequence deviation amount is calculated by the following formula: ； wherein: represents a sequence of node state features; represents a sequence of path features of a pre-distribution path ; represents a sequence difference operator on a sequence of node state features and a sequence of path features of a pre-distribution path ; represents a sequence alignment operator on a sequence of node state features and a sequence of path features of a pre-distribution path ; represents an index of a pre-distribution path According to the sequence bias amount Computing path difference energy , the path difference energy Is calculated by the following formula: ； In the formulae: represents a mirror layer category to which the mirror layer belongs; represents a category mapping index corresponding to the mirror layer category C; represents a path difference energy of the pre-distribution path ; According to path difference energy Computing path feasibility values , the path feasibility values are computed from the following formula: ； In the formulae: represents a pre-distribution path path feasibility value; represents a natural exponential function.

2. The container image pre-distribution method based on semantic hierarchy and cooperative scheduling according to claim 1, characterized in that, The analysis process of constructing the image layer structure comprises: Performing structural analysis on the component set in the container image, and identifying the component dependency relationship according to the call relationship graph between the components; Determining the change frequency of the components according to the change record of the container image in the historical version; Determining the association degree between the components and the business function modules according to the business function module division rule to which the container image belongs; Constructing a component attribute set based on the component dependency relationship, the change frequency and the business function module association degree, and calculating a component stability index and a component business coupling index according to the component attribute set; Generating a layering determination result according to the component stability index and the component business coupling index, the layering determination result being used to divide the components into a base layer component set, a common layer component set and a dynamic layer component set; Generating the corresponding base layer, common layer and dynamic layer according to the base layer component set, common layer component set and dynamic layer component set.

3. The container image pre-distribution method based on semantic hierarchy and cooperative scheduling according to claim 1, characterized in that, The process of collecting the resource state and image layer hit history of the edge node comprises: Obtaining the computing resource occupation information, storage resource occupation information, network connectivity information and image layer hit history of the edge node; Constructing a node state feature set according to the resource fluctuation characteristics of the edge node, the time sequence characteristics of the image layer access, the change trend of the network connectivity and the resource state of the adjacent node; Identifying the running mode of the edge node according to the node state feature set, and determining the future resource availability index of the node according to the running mode.

4. The container image pre-distribution method based on semantic hierarchy and cooperative scheduling according to claim 3, characterized in that, The process of generating the node comprehensive score based on the weighted scoring model comprises: Determining the scoring factors for reflecting the resource bottleneck degree, network connectivity stability and image layer multiplexing potential according to the node state feature set, and setting the sensitive weight of the scoring factors related to the image layer category; The sensitive weight and the basic weight of the scoring factor are dynamically adjusted according to the node operation mode and the service load state, and a nonlinear combination is established between the scoring factors to represent the synergistic influence relationship. The node state feature set is input into the weighted scoring model to generate a node comprehensive score, which is used to indicate the load adaptation of the target edge node when performing different mirror layer pre-distribution.

5. The container image pre-distribution method based on semantic hierarchy and cooperative scheduling according to claim 1, characterized in that, The process of selecting a pre-distribution path according to the path feasibility value includes: The pre-distribution path set is selected according to the path feasibility value of each pre-distribution path as follows: ； In the formula: represents the final selected pre-distribution path set; represents the maximum value among all path feasibility values; a logical OR operator; represents a feasibility threshold value; represents an index enumerating all pre-distribution paths.

6. The container image pre-distribution method based on semantic hierarchy and cooperative scheduling according to claim 1, characterized in that, The process includes: The cache management mode corresponding to the mirror layer is determined according to the mirror layer category to which the mirror layer belongs; The cache area in the target edge node is selected according to the cache management mode corresponding to the mirror layer.

7. The container image pre-distribution method based on semantic hierarchy and cooperative scheduling according to claim 6, characterized in that, The process of performing the corresponding long-term retention management, usage frequency management or on-demand replacement management in the target edge node includes: When the mirror layer belongs to the basic layer category, the mirror layer is stored in the long-term storage area of the target edge node, and a set of stable versions of the basic layer mirror in the long-term storage area is maintained; When the mirror layer belongs to the frequently used layer category, the set of frequently used layer mirrors in the frequently used access cache area in the target edge node is adjusted according to the change of the access frequency of the mirror layer in the target edge node; When the mirror layer belongs to the dynamic layer category, the set of dynamic layer mirrors in the dynamic storage area in the target edge node is updated according to the recent access of the mirror layer in the target edge node.

8. The container image pre-distribution system based on semantic layering and cooperative scheduling, applied to the container image pre-distribution method based on semantic layering and cooperative scheduling according to any one of claims 1-7, characterized in that, The system includes: A semantic analysis module for performing semantic analysis on the container mirror and constructing a mirror layer structure according to the component dependency relationship, change frequency and business function module of the container mirror, the mirror layer structure including three mirror layer categories of a basic layer, a frequently used layer and a dynamic layer; A node evaluation module for collecting resource states and mirror layer hit history of edge nodes and generating a node comprehensive score based on a weighted scoring model, the node comprehensive score being used to represent the adaptation degree of edge nodes receiving mirror layers; A path selection module for selecting one or more paths for transmitting mirror layers from a pre-defined pre-distribution path set according to the node comprehensive score and the mirror layer category to which the mirror layer belongs, the pre-distribution path set including a local multiplexing path, a regional cache node acquisition path, a layered index node acquisition path and an edge node-to-edge node point-to-point transmission path; A mirror distribution and cache management module for transmitting mirror layers to target edge nodes according to the selected pre-distribution path and performing long-term retention management, usage frequency management or on-demand replacement management in the target edge node according to the mirror layer category to which the mirror layer belongs.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1-7.

Images