Cross-regional data collaborative computing method and system based on multi-level cache mechanism

Abstract

The invention discloses a cross-regional data cooperative computing method and system based on a multi-level cache mechanism, and relates to the technical field of cloud computing, and the method comprises the steps: obtaining a historical operation state sequence of each cloud cache platform and task description data of a to-be-scheduled target computing task; inputting the task description data and the running state data of each cloud cache platform into a platform scheduling model to output a matched cross-regional data collaboration path; packaging the target computing task and the dependent environment thereof to generate a corresponding task container mirror image, and sending the task container mirror image to each potential cloud cache platform; and controlling the first potential cloud cache platform to run the corresponding task container mirror image, monitoring the real-time running state of the first potential cloud cache platform, and switching and controlling the second potential cloud cache platform to run the corresponding task container mirror image under the condition that the real-time running state is monitored to meet the task migration condition. Therefore, task migration scheduling and resource management based on a multi-level cache mechanism are realized.

Description

Technical Field

[0001] This application relates to the field of cloud computing technology, and particularly to a cross-regional data collaborative computing method and system based on a multi-level cache mechanism. Background Art

[0002] "Eastern Data Calculated in the Western Regions" is a national strategy aimed at promoting the optimal coordination of data centers in the eastern region and computing capabilities in the western region. With the development of cloud computing and big data, collaborative computing between data centers can effectively improve resource utilization and reduce latency.

[0003] Currently, the resource scheduling and data storage management of data centers play a crucial role in cross-regional data collaborative computing. In a large-scale distributed computing environment, the storage and access efficiency of data is crucial for computing performance. Although existing distributed computing and storage solutions have alleviated the pressure of cross-regional data processing to a certain extent, the multi-level cache system often fails to form effective coordination with the task scheduling mechanism in actual deployment.

[0004] In addition, the existing resource scheduling strategies of data centers are mainly implemented based on static or rule-based scheduling algorithms and cannot adapt to the dynamic changes of cloud platform resource status in real time. Especially in the cross-regional collaborative computing scenario, there are large fluctuations in the utilization of computing resources and cache resources. Traditional scheduling strategies are difficult to optimize resource allocation in a timely manner in the face of resource load changes, sudden task demands, etc., resulting in prominent problems such as task delay and resource waste. Summary of the Invention

[0005] This application provides a cross-regional data collaborative computing method, system, storage medium, computer program product, and electronic device based on a multi-level cache mechanism, so as to at least solve the problems of network transmission bottlenecks, insufficient cache strategies, static scheduling decisions, and limited overall coordination ability of traditional multiple cloud cache platform resources in cross-regional data collaboration in the current related technologies.

[0006] In a first aspect, an embodiment of the present application provides a cross-regional data collaborative computing method based on a multi-level caching mechanism, including: obtaining the historical operation status sequence of each cloud caching platform and the task description data of the target computing task to be collaboratively computed; the historical operation status sequence includes multiple historical time windows and corresponding historical operation status data, and the parameter types of the historical operation status data include cache hit rate, cache latency, cache capacity utilization rate, computing resource utilization rate, and network transmission latency; the task description data includes task data size, task data caching requirements, task computing resource requirements, task type, and task processing time limit; inputting the task description data and the historical operation status sequence of each cloud caching platform into a platform scheduling model to output a matching cross-regional data collaborative path; the cross-regional data collaborative path defines multiple potential cloud caching platforms arranged in descending order of priority; the platform scheduling model is a deep learning model; packaging the target computing task and its dependent environment to generate a corresponding task container image, and sending the task container image to each of the potential cloud caching platforms; controlling a first potential cloud caching platform to run the corresponding task container image, monitoring the real-time operation status of the first potential cloud caching platform, and switching to control a second potential cloud caching platform to run the corresponding task container image when the monitored real-time operation status meets the task migration condition; the priority of the first potential cloud caching platform is higher than that of the second potential cloud caching platform.

[0007] Second aspect, an embodiment of the present application provides a cross-region data collaborative computing system based on a multi-level cache mechanism, including: a data acquisition unit, configured to acquire the historical operation status sequences of each cloud cache platform and the task description data of the target computing task to be collaboratively computed; the historical operation status sequence includes a plurality of historical time windows and corresponding historical operation status data, and the parameter types of the historical operation status data include cache hit rate, cache latency, cache capacity utilization rate, computing resource utilization rate, and network transmission latency; the task description data includes task data size, task data cache requirement, task computing resource requirement, task type, and task processing time limit; a collaborative path matching unit, configured to input the task description data and the historical operation status sequences of each cloud cache platform into a platform scheduling model to output a matching cross-region data collaborative path; the cross-region data collaborative path defines a plurality of potential cloud cache platforms arranged in descending order of priority; the platform scheduling model is a deep learning model; a task image synchronization unit, configured to package the target computing task and its dependent environment to generate a corresponding task container image, and send the task container image to each of the potential cloud cache platforms; a platform monitoring and switching unit, configured to control the first potential cache cloud cache platform to run the corresponding task container image, monitor the real-time operation status of the first potential cloud cache platform, and switch to control the second potential cloud cache platform to run the corresponding task container image when the monitored real-time operation status meets the task migration condition; the priority of the first potential cloud cache platform is higher than that of the second potential cache cloud cache platform.

[0008] Third aspect, there is provided an electronic device, including: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the steps of the cross-region data collaborative computing method based on a multi-level cache mechanism according to any embodiment of the present application.

[0009] Fourth aspect, an embodiment of the present application provides a storage medium, on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the cross-region data collaborative computing method based on a multi-level cache mechanism according to any embodiment of the present application are implemented.

[0010] Fifth aspect, an embodiment of the present application provides a computer program product, including computer programs / instructions, and when the computer programs / instructions are executed by a processor, the steps of the cross-region data collaborative computing method based on a multi-level cache mechanism according to any embodiment of the present application are implemented.

[0011] Through a cross-region data collaborative computing method and system based on a multi-level cache mechanism provided by the present application, at least the following technical effects can be achieved:

[0012] (1) Analyze the historical operating status and target task data of multiple cloud caching platforms through a deep learning model, which has dynamic learning and adaptive capabilities. It can output the optimal cross-regional data collaboration path based on the historical and real-time data of task types, computing resource requirements, and platform bandwidth. Based on the intelligent matching of cloud caching platform resources and data computing tasks, it realizes the automatic coordination of resources and the dynamic migration of tasks. By real-time monitoring the operating status of the cloud caching platform, during the task execution, once it is found that the resource usage of the current cloud caching platform exceeds the threshold or no longer meets the task requirements, it can trigger the invocation of other backup cloud caching platforms with lower priorities to continue executing the task according to the cross-regional data collaboration path. Thus, based on the migration mechanism of real-time status monitoring, the successful response rate and reliability of the task are guaranteed.

[0013] (2) Package the target task and its dependent environment as a container image, making the migration of tasks between different cloud caching platforms fast and seamless. Containerization technology not only ensures the portability of data computing tasks but also makes the switching between platforms have almost no impact on task execution. With the support of the standardized packaging method of container images, whether it is an edge computing platform or a public cloud caching platform, it can execute tasks in a standardized manner, ensuring the continuity of cross-platform collaboration.

[0014] Through this technical solution, by integrating a deep learning model, containerization technology, and a cross-platform resource dynamic migration mechanism, and based on the task migration scheduling and resource management of a multi-level caching mechanism, intelligent scheduling and resource management in multi-cloud and hybrid cloud environments are realized, which can adaptively respond to changes in complex dynamic environments, improving resource utilization and the continuity of task execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 Shows a flowchart of an example of a cross-regional data collaboration calculation method based on a multi-level caching mechanism according to an embodiment of the present application;

[0017] Figure 2 Shows according to Figure 1 An example of an operation flowchart of step S130 in

[0018] Figure 3 Shows according to Figure 2An operation flowchart of an example of step S230 in

[0019] Figure 4 Shows a schematic structural connection diagram of an example of a platform scheduling model according to an embodiment of the present application;

[0020] Figure 5 Shows a schematic block diagram of an example of a cross - region data collaborative computing system based on a multi - level cache mechanism according to an embodiment of the present application;

[0021] Figure 6 Is a schematic structural diagram of an embodiment of an electronic device of the present application. Detailed implementation manners

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0023] In the technical solutions of the present application, for the processing of collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved, etc., all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0024] Figure 1 Shows a flowchart of an example of a cross - region data collaborative computing method based on a multi - level cache mechanism according to an embodiment of the present application.

[0025] Regarding the execution subject of the method in the embodiments of the present application, it can be any controller or processor with computing or processing capabilities. Specifically, it can be implemented by a hybrid cloud cache platform scheduling platform or a scheduling server. Through the automated scheduling and task migration mechanism of task allocation and migration driven by a deep learning model, the complexity of manual management is reduced, the system operation cost is lowered, and the needs of enterprises and scientific research projects for an efficient cloud computing platform are met.

[0026] In some examples, it can be integrated and configured in an electronic device or a terminal in a software, hardware, or software - and - hardware combination manner, and the types of the terminal or the electronic device can be diverse, such as mobile phones, tablet computers, or desktop computers, etc.

[0027] As Figure 1 shown, in step S110, the historical operation state sequences of each cloud cache platform and the task description data of the target computing task to be collaboratively computed are obtained.

[0028] Here, the historical operation status sequence includes multiple historical time windows and corresponding historical operation status data. The parameter types of the historical operation status data include cache hit rate, cache latency, cache capacity utilization rate, computing resource utilization rate, and network transmission latency. In some embodiments, the API or log analysis tool (such as Prometheus or CloudWatch, etc.) is used to obtain the historical monitoring data of each cloud cache platform, and the data is divided according to different time windows (such as every hour, daily), and archived into parameters such as computing resource utilization rate (such as CPU, GPU, memory occupancy rate), network latency, and bandwidth utilization rate (upload / download speed and bandwidth usage). The cache hit rate refers to the proportion of the number of requests that directly obtain data from the cache in the cache requests, reflecting the effectiveness of the cache. It can be automatically recorded and updated by the cache management agent deployed on the cloud cache platform. The cache latency refers to the time required to successfully obtain data from the cache after the request is initiated. The lower the latency, the higher the data access efficiency. For example, the response time of each request is recorded through the $request_time variable in Nginx. The cache capacity utilization rate represents the proportion of the storage space already used by the cache platform in the total cache capacity, reflecting the usage of the cache resources. For example, it can be viewed through Redis commands.

[0029] The task description data includes task data size, task computing resource requirements, task type, and task processing time limit. In some embodiments, after the access request arrives at the hybrid cloud cache platform scheduling platform, the task data is extracted from the access request or the scheduling task queue, specifically including the task data size (MB / GB) used to describe the volume of the transmitted data, the task computing resource requirements (number of CPUs / GPUs, memory requirements, etc.), the task data cache requirements, the task type (real-time task, batch processing task, etc.), and the time constraint for the task to be completed. Here, the task data cache requirement represents the demand for the cache space by the target computing task, which depends on the data scale, data access frequency, and data sharing degree of the task, and can be distinguished by the task description analysis module, for example, divided into cache requirement levels of "low demand", "medium demand", and "high demand".

[0030] In step S120, the task description data and the historical operation status sequences of each cloud cache platform are input into the platform scheduling model to output a matching cross-region data collaboration path.

[0031] Here, the platform scheduling model is a deep learning model. Specifically, the model selection of the platform scheduling model can be various non-restrictive time series models, such as LSTM (Long Short-Term Memory), Transformer, etc., so as to predict the operation status of each platform in the future period of time, and then output the cross-region data collaboration path.

[0032] The cross-regional data collaboration path defines multiple potential cloud caching platforms arranged in descending order of priority. In some embodiments, the path is a list of cloud caching platforms arranged by priority, and the priority of each cloud caching platform is calculated based on comprehensive factors such as its resource utilization, latency, and bandwidth. Exemplarily, for the first potential cloud caching platform with the highest priority, it is expected to have the running state most in line with the task requirements, while the other second potential cloud caching platforms serve as alternative nodes and switch to running when the first potential cloud caching platform has insufficient resources or anomalies.

[0033] In step S130, the target computing task and its dependent environment are packaged to generate a corresponding task container image, and the task container image is sent to each potential cloud caching platform.

[0034] It should be noted that in traditional container management technologies, generally, the container images of application services are created. For example, a long-running application service (such as a Web server, a database, an API service, etc.) is packaged into a container image for cross-platform deployment and management. In contrast, in the embodiments of the present application, a specific computing task (such as data analysis, batch processing tasks, etc.) is packaged as a container image for execution across cloud caching platforms, and the resources are released after the task is completed. Different from the continuous operation of application services, tasks are usually short-term computations and stop after completion (such as a period of data analysis, a round of model training).

[0035] Table 1 Comparison between traditional application service container image creation and task container image creation in this embodiment

[0036]

[0037]

[0038] As shown in Table 1, application service image creation tends to stable operation and continuous resource occupation, while task image creation pays more attention to the elastic use and on-demand scheduling of resources. Specifically, task container image creation supports the dynamic migration of tasks, selects the best platform according to the real-time running state, and improves the elasticity and efficiency of the system. In addition, the task containers are destroyed after execution, and the resources are released, avoiding long-term occupation of cloud resources.

[0039] In particular, in business application scenarios, traditional application service container image creation is suitable for running long-term and highly concurrent services, such as API gateways or databases. For the task container image creation in the embodiments of the present application, it is more suitable for temporary tasks that need to quickly complete computations, such as the training of machine learning models or batch data analysis. Therefore, through the dynamic image creation and cross-platform scheduling of tasks, the resource utilization efficiency and system flexibility are improved.

[0040] In some embodiments, the code, configuration files, and runtime environment of the task (such as Python environment, dependent libraries like TensorFlow, etc.) are encapsulated into a Docker image or other standardized container images. Subsequently, using the image transfer protocol of the distributed image repository or cloud caching platform, the generated task container image is pushed to each potential cloud caching platform. Preferably, the image size and transfer time can be optimized according to the data scale of the task or the situation of the cloud caching platform. For example, for a cloud caching platform with a large regional span, data compression or edge caching technology can be used to reduce the transfer time.

[0041] Through the embodiments of the present application, the runtime environment and dependencies of the task are encapsulated using containerization technology, ensuring that the task can be quickly deployed and run between different cloud caching platforms, reducing the startup time. In addition, through standardized container technology, the high portability and cross-platform compatibility of the task are ensured, avoiding execution errors caused by environmental differences between different platforms, and improving the robustness of the task scheduling system.

[0042] In step S140, control the first potential cloud caching platform to run the corresponding task container image, monitor the real-time running state of the first potential cloud caching platform, and switch to control the second potential cloud caching platform to run the corresponding task container image when the monitored real-time running state meets the task migration condition. The priority of the first potential cloud caching platform is higher than that of the second potential cloud caching platform.

[0043] In some embodiments, tools such as Kubernetes or Prometheus are used to monitor the resource usage during the task execution, including CPU, memory, network latency, etc., to determine whether the task migration condition is triggered. Regarding the description of the task migration condition, by way of example, it includes resource overload, increased network latency, cost optimization requirements, etc. For example, when the computing resource utilization rate continuously exceeds a set threshold (such as 90%) for a period of time, or the network transmission latency of the platform exceeds the acceptable range, or a cloud caching platform with lower cost is selected to run the task according to a preset policy, and so on. Subsequently, when it is detected that the task migration condition is triggered, directly restore the execution state of the task through the task image on the new platform, avoiding restarting the task and improving the migration efficiency.

[0044] Through the embodiments of the present application, during the task execution process, the optimal cloud caching platform is dynamically selected according to the change of the real-time running state. Based on real-time monitoring and the automated migration mechanism, the scheduling platform can cope with emergencies, ensuring seamless migration of the task between different platforms, and improving the continuity of the task and the fault tolerance of the system.

[0045] In some examples of the embodiments of the present application, during task migration, the scheduling platform will pause the execution of the current task, and transfer the task image and intermediate state to the second potential cloud caching platform for relaying the task execution, ensuring the continuity and performance stability of the task.

[0046] Specifically, during the process of running the task container image on the first potential cloud caching platform, the first potential cloud caching platform generates a task checkpoint and synchronizes the task checkpoint among all potential cloud caching platforms. The task checkpoint is used to store the intermediate execution state data of the target computing task. When it is monitored that the first running state data meets the task migration condition, the control is switched to the second potential cloud caching platform to run the corresponding task container image according to the task checkpoint.

[0047] In some embodiments, the checkpoint file is used to save the intermediate execution state data of the task, and it can be generated periodically. For example, a checkpoint is generated every 10 minutes or every 1000 data processing batches. Furthermore, the checkpoint file is stored in a distributed storage system (such as Ceph, HDFS) or cloud storage (such as AWS S3, Alibaba Cloud OSS, etc.), ensuring that all potential cloud caching platforms can access the latest task status. Incremental synchronization is performed among multiple potential cloud caching platforms, and only the data that has changed since the last checkpoint is transmitted, reducing the network overhead caused by synchronization. After the migration condition is triggered, the task on the first potential cloud caching platform is paused to ensure that its execution state is consistent with the latest checkpoint. Furthermore, on the second potential cloud caching platform, the checkpoint file is loaded from the distributed storage, and the execution state of the task is restored to ensure that the task can continue to run seamlessly from the state before migration.

[0048] Through the embodiments of the present application, the complete intermediate execution state data is saved based on the task checkpoint, enabling the task to continue execution from the last saved state after migration, avoiding task restart or repeated calculation, and significantly shortening the pause time of the task during migration, improving the task continuity of the system. Thus, when the running state changes, the system can quickly switch to other cloud caching platforms to continue task execution, enhancing the elasticity and robustness of the system. In addition, through the optimization of incremental synchronization and distributed storage, the checkpoint mechanism reduces the network transmission overhead during migration, ensuring the efficient execution of the task.

[0049] Regarding the implementation details of the above step S130, in some embodiments, before generating the container image, the system will automatically analyze the dependency environment of the target task, which includes: the code and execution script of the computing task, the required system libraries (such as specific versions of Python, Java, etc.), the data dependencies of the initial data set required for task execution (such as configuration files or model parameters), and the hardware requirements (such as GPU, CPU core count, and memory size).

[0050] Furthermore, use Docker or OCI (Open Container Initiative) - compatible container image tools to package the code, dependency packages, and environment configurations into a standardized image file. Specifically, in the image build file (such as Dockerfile), package the task environment in stages to ensure the smallest possible image size. In the first stage, select a base image (such as ubuntu:20.04, python:3.9 - slim, etc.). In the second stage, install the task dependency packages (such as numpy, tensorflow, etc.). In the third stage, add the task code and required data.

[0051] For frequently used libraries and dependencies, build a common base image layer (Base Image) to avoid building from scratch for each task, reduce the build time, and achieve incremental image updates. In this way, when the task - dependent environment changes, only the new layer of the image is updated instead of a complete reconstruction.

[0052] In some preferred embodiments, use image signing and encryption to ensure the security of the image during transmission between different cloud caching platforms. Specifically, sign the image through a tool (such as Docker Content Trust) to ensure the integrity of the image. Furthermore, encrypt the sensitive data layer before image transmission to prevent data leakage.

[0053] Figure 2 Shows an operation flowchart of an example according to Figure 1 step S130 in

[0054] As Figure 2 shown, in step S210, parse the task code, dependency packages, and environment configurations corresponding to the target computing task.

[0055] In some embodiments, use a static code analysis tool (such as PyLint) to automatically parse the task code and identify dependencies. Extract the library and module references in the task code to generate a dependency list. Exemplarily, the dependency management tool can adopt "npm" or "Maven" to capture libraries and versions.

[0056] In addition, record the system environment information, such as the operating system version, required runtime libraries, Python version, etc. Furthermore, generate a configuration file (such as "requirements.txt" or "environment.yml") to describe the required runtime environment of the task. Use tools such as "conda" or "pip" to automatically generate the dependency file.

[0057] In step S220, call the container image tool to package the parsed task code, dependency packages, and environment configurations to generate the corresponding initial image file.

[0058] Exemplarily, write a "Dockerfile" to specify the base image and build steps. Specifically, select a suitable base image and add instructions in the Dockerfile to install the resolved dependent packages. The relevant code snippet is as follows:

[0059] ```Dockerfile

[0060] FROM python:3.9-slim

[0061] COPY requirements.txt.

[0062] RUN pip install -r requirements.txt

[0063] COPY. / app

[0064] WORKDIR / app

[0065] CMD ["python", "task.py"]

[0066] ```

[0067] Then, use the multi-stage build technique to reduce the volume of the final image. Specifically, use tools (such as docker-slim) to analyze and delete unnecessary files and layers to optimize the image. Thereby, reduce the image volume, shorten the pulling time, and lower the storage cost.

[0068] In step S230, encrypt the initial image file to obtain the corresponding task container image, and send the task container image to each potential cloud caching platform.

[0069] It should be noted that the encryption method can be diversified, and encryption tools (such as "OpenSSL", etc.) can be used to encrypt the initial image file to ensure the security of data transmission. By adopting the container image signature mechanism, ensure the integrity of the image during transmission. In some embodiments, push the encrypted image file to the image repository of the cloud caching platform, and the encryption and signature mechanisms can be used to ensure data security, realizing the fast transmission and storage of the image between different cloud caching platforms.

[0070] Through the embodiments of the present application, automatically resolve task dependencies and perform standardized image processing to ensure seamless connection of task execution on different cloud caching platforms. Ensure the security of the image during transmission through encryption and signature.

[0071] Figure 3 shows according to Figure 2An operation flowchart example of step S230 in

[0072] As Figure 3 shown, in step S310, the base layer data and application layer data in the initial image file are parsed, and a random session key and the cloud cache platform public key for each potential cloud cache platform are obtained.

[0073] Specifically, the initial image file includes the dependency environment and execution logic of the task, is organized according to a multi-layer structure, and contains base layer data and application layer data. For example, the base layer data may include an operating system, basic dependency packages (such as the Alpine Linux base image of a container), and general library files. The application layer data may include the business logic of the task, dedicated dependencies (such as deep learning models, database connection configurations), task code, and runtime environment.

[0074] In addition, the scheduling platform can generate a unique random session key for each potential cloud cache platform to ensure the independence of the data transmission and access process for each platform. The corresponding cloud cache platform public key (the public key part of the asymmetric key) is obtained from the key management system of each potential cloud cache platform (such as AWS KMS, Alibaba Cloud KMS).

[0075] Thus, the base layer data and application layer data are separated, enabling the image file to be loaded on different cloud cache platforms as needed, reducing redundant data transmission, and improving transmission efficiency. In addition, an independent random session key is generated for each cloud cache platform, and the key provided by its public key system is used to ensure data security and avoid the risk of key leakage or reuse.

[0076] In step S320, the application layer data is symmetrically encrypted according to each random session key to generate the corresponding encrypted application layer data.

[0077] In some embodiments, the application layer data is symmetrically encrypted using a random session key (such as AES-256). The AES algorithm provides efficient and powerful encryption capabilities to ensure that the application layer data is not stolen by a third party during transmission. Preferably, large-scale application layer data (such as large model files) is encrypted in chunks, and each data chunk is encrypted in parallel using multi-threading or GPU acceleration to improve the encryption speed.

[0078] In step S330, for each potential cloud cache platform, the corresponding random session key is asymmetrically encrypted using the cloud cache platform public key of the potential cloud cache platform to obtain the corresponding encrypted session key, and the base layer data, encrypted application layer data, and encrypted session key are packaged into the corresponding encrypted transmission package.

[0079] Here, various algorithms can be adopted for the asymmetric encryption algorithm, such as the RSA encryption algorithm, to asymmetrically encrypt the random session key, ensuring that the session key will not be leaked during the transmission process. Each encrypted session key corresponds to a unique cloud cache platform, and other platforms cannot use their own private keys to decrypt the key, thereby ensuring access control of the data.

[0080] Furthermore, the basic layer data (unencrypted), the encrypted application layer data, and the encrypted session key are integrated into an encrypted transmission packet. By adopting a standardized data format (such as JSON or Protobuf), the transmission packet can be smoothly parsed and used between different platforms.

[0081] In this way, the application layer data is protected by symmetric encryption, and then the session key is protected by asymmetric encryption, ensuring that the data cannot be cracked even if the transmission packet is intercepted. In addition, since the basic layer data does not need to be encrypted, it can be shared between different platforms, reducing duplicate transmissions and improving efficiency.

[0082] In step S340, each encrypted transmission packet is respectively sent to the corresponding potential cloud cache platform, so that the potential cloud cache platform decrypts the encrypted session key in the encrypted transmission packet based on its respective cloud cache platform private key to crack the application layer data, and then restores the complete task container image according to the cracked application layer data and the basic layer data in the encrypted transmission packet.

[0083] In some embodiments, a secure communication channel is established through the TLS / SSL protocol to ensure that the transmission packet will not be tampered with or stolen during the transmission process on the network. Preferably, if the transmission path is complex or there are network restrictions, the transmission packet can be transmitted in segments through different paths and recombined at the cloud cache platform end.

[0084] In addition, on the target cloud cache platform, the private key of the platform is used to decrypt the encrypted session key in the transmission packet. Using the decrypted session key, the encrypted application layer data is decrypted to obtain the original application layer content. According to the basic layer data and the decrypted application layer data, a complete task container image is reconstructed on the cloud cache platform. Therefore, each platform can only use its own private key to decrypt the corresponding session key, ensuring that the data can only be decrypted and used on the specified platform, realizing an isolated data transmission process for different cloud cache platforms.

[0085] In some preferred embodiments of the embodiments of the present application, the cloud cache platform public key and the matching cloud cache platform private key for each potential cloud cache platform are a corresponding pair of independent elliptic curve (Elliptic Curve Cryptography, ECC) public and private key pairs.

[0086] Exemplarily, each cloud caching platform (such as A, B, and C) generates its own ECC public key and private key, and shares the public key with the scheduling server in advance. The scheduling server maintains a key library that stores the ECC public keys of each potential cloud caching platform for encrypting session keys.

[0087] It should be noted that compared with the traditional RSA algorithm, ECC has the advantages of shorter key length and higher computing efficiency, and is particularly suitable for the resource-limited and efficient scheduling requirements in the multi-cloud caching platform environment. Specifically, under the same security level, the key length required by ECC is much shorter than that of the traditional RSA. For example, the security of ECC-256 bits is equivalent to that of RSA-3072 bits. Therefore, a shorter key means faster encryption speed and less memory occupancy, which is very suitable for the multi-cloud caching platform environment with frequent task migration and dynamic scheduling. In addition, ECC has high computing efficiency and small resource overhead, which can accelerate the encryption and decryption speed and reduce the task startup time.

[0088] Figure 4 The structural connection diagram of an example of the platform scheduling model according to an embodiment of the present application is shown.

[0089] As Figure 4 shown, the platform scheduling model 400 includes an LSTM layer 410, a feature splicing layer 420, and a simulated annealing matching layer 430.

[0090] The LSTM layer 410 is used to respectively infer the predicted operating state data of the corresponding cloud caching platform in the next time window according to each historical operating state data.

[0091] Here, the LSTM can capture the time dependence of the historical operating state sequence and predict the predicted operating state data such as resource utilization rate and latency of each cloud caching platform in the next time window.

[0092] Regarding the training set of the LSTM layer 410, historical data can be used, and different time period training samples can be generated through the sliding window technique. It can use the mean square error (MSE) as the loss function to measure the gap between the predicted operating state and the true future operating state.

[0093] The feature splicing layer 420 is used to splice the task description data and each of the predicted operating state data to obtain the corresponding spliced feature vectors of each cloud caching platform, so as to construct an estimated operating state matrix.

[0094] In some embodiments, the predicted operating state data (from the LSTM layer) of each cloud caching platform is concatenated with the task description data of the task to be executed to form a concatenated feature vector. Further, the concatenated feature vectors of all cloud caching platforms are combined into an estimated operating state matrix. For example, each row of the matrix corresponds to the feature vector of a cloud caching platform.

[0095] The simulated annealing matching layer 430 is used to solve the estimated operating state matrix through the simulated annealing algorithm to generate a cross-region data collaboration path.

[0096] The simulated annealing algorithm (SA) is a heuristic algorithm that is good at solving complex combinatorial optimization problems. It avoids getting stuck in local optimal solutions by simulating the heating and cooling in the metal annealing process, thereby finding the global optimal solution. Accordingly, a group of cloud caching platforms is selected from the estimated operating state matrix, and their priorities are arranged to generate an optimal cross-region data collaboration path. The collaboration path provides a basis for the dynamic migration of tasks, ensuring that when a performance bottleneck occurs in the first cloud caching platform, the system can quickly switch to an alternative cloud caching platform.

[0097] In some examples of the embodiments of the present application, the LSTM layer 410 is an LSTM layer based on the attention mechanism:

[0098] h i,t ,c i,t = LSTM(x i,t ,h i,t-1 ,c i,t-1 ), Equation (1)

[0099] Here, the LSTM has the ability to capture the long-term and short-term dependencies of time series data. In a cloud computing environment, the resource usage in different time periods may have correlations and periodicities. For example, the fluctuations in CPU and memory usage during peak and off-peak hours, the changes in network latency and bandwidth at different times, and so on. Through the time series analysis of the LSTM layer, the resource usage of each cloud caching platform can be predicted, making task scheduling more forward-looking.

[0100] α i,t = Softmax(W a h i,t ), Equation (2)

[0101] Here, the attention mechanism can reduce the impact of noise or outliers on the model by weighted summation of the historical time series states. Even in the face of a complex cloud environment (such as random fluctuations in the operating state), the model can provide reliable predictions.

[0102]

[0103] It should be noted that the operating states at different time points are not equally important. For example, the resource usage status at certain moments is more valuable for future prediction. The attention mechanism will assign different weights α to different time steps i,t , ensuring that the model pays more attention to the operating states of critical time windows and improving the prediction accuracy.

[0104]

[0105] Here, the prediction model can output the operating state of each platform at future time t+1, providing data support for task scheduling decisions. For example, predicting the CPU usage and latency changes of the AWS platform to help the system determine whether task migration needs to be carried out in advance.

[0106] In the above formula, i represents the i-th cloud caching platform, t represents the t-th time window, and x i,t represents the input vector of the operating state of the i-th cloud caching platform at the t-th time window, and h i,t represents the hidden state of the i-th cloud caching platform at the t-th time window, and c i,t represents the LSTM cell state of the i-th cloud caching platform at the t-th time window; α i,t is the attention weight of the i-th cloud caching platform at the t-th time window, indicating the importance of this moment; W a is the weight matrix of the attention mechanism; N represents the total number of time windows corresponding to the historical operating state sequence; is the weighted hidden state of the i-th cloud caching platform, representing the feature representation after the attention mechanism; represents the predicted operating state of the i-th cloud caching platform at the (t + 1)-th time window.

[0107] Through the embodiments of the present application, by adopting the LSTM layer based on the attention mechanism, the scheduling platform can capture the long-term and short-term change trends of the operating state, and use the attention mechanism to make the model pay more attention to the time points that have a greater impact on the prediction, reducing the interference of noise and abnormal data, and improving the robustness of the prediction result. Thus, providing accurate prediction data for task migration and allocation, reducing migration losses, and ensuring that the resource utilization rate and task response efficiency of the cloud caching platform reach the optimal.

[0108] In some examples of the embodiments of the present application, the feature splicing layer 420 is a non-linear activation layer:

[0109]

[0110] Here, the operating state predicted by the LSTM is spliced with the description information of the task (data size D, computing resource requirement R, task type L, and processing time limit S) into a high-dimensional feature vector Fi Therefore, the requirement information of each task is dynamically combined with the operation status information of the cloud caching platform to achieve personalized scheduling for different tasks.

[0111]

[0112] Here, the ReLU activation function and the square term are used to capture the complex non-linear relationship between task requirements and platform status, enhancing the expressive power of the model. Through the weight matrix W F The influence of different features can be adaptively adjusted, enabling the model to make full use of key feature information during the scheduling process. In addition, the ReLU activation function can effectively avoid the vanishing gradient problem, ensuring efficient optimization of the model during training.

[0113]

[0114] In this way, the scheduling server can analyze the operation status of all cloud caching platforms simultaneously based on matrix M, and formulate the optimal task scheduling strategy from a global perspective.

[0115] In the above formula, F i represents the concatenated feature vector of the i-th cloud caching platform, D represents the task data size, R represents the task computing resource requirement, L represents the task type, and S represents the task processing time limit; W F and b F represent the weight matrix and bias term of the non-linear activation layer respectively, M represents the estimated operation status matrix, represents the feature vector corresponding to the i-th cloud caching platform in M, and n represents the number of cloud caching platforms.

[0116] Through the embodiments of the present application, based on the information of the operation status matrix M, the scheduling server can dynamically allocate tasks to ensure that tasks always run on the optimal cloud caching platform. If the operation status of a certain cloud caching platform changes (such as a sharp increase in the predicted CPU or memory usage rate), the operation status of other platforms can be quickly evaluated through this matrix, and the task can be migrated to a more suitable cloud caching platform with the least migration loss.

[0117] In some examples of the embodiments of the present application, the simulated annealing matching layer 430 is used to perform the following operations:

[0118] Randomly generate an initial path P0 according to the estimated operation status matrix, and perform annealing iteration operations based on the initial path P0 and the initial temperature T max

[0119] In each iteration, calculate the objective function value of the current path P. The expression of the objective function is:

[0120] ​

[0121] In the embodiments of the present application, the simulated annealing algorithm is used to find the optimal platform path for current and future task executions, balancing execution efficiency and migration loss. The first term in the energy function of Equation (8) represents the resource consumption of the current task execution, while the second term in Equation (8) represents the loss of future task migration. Specifically, through the weighted summation of the feature matrix, the operating state of each platform (such as CPU, memory, and network latency) can be dynamically sensed, the most suitable tasks can be allocated, resource waste can be avoided, and according to the platform characteristics and cost information, the cloud caching platform with lower cost can be preferentially selected to reduce the operating cost.

[0122]

[0123] Here, by using the reference variable to normalize different metrics, it is ensured that the metrics of migration loss are compared within the same scale, avoiding errors introduced due to different data volumes, network costs, and latency units. In addition, by introducing the square term for latency difference, the influence degree of latency on task performance is amplified, ensuring that the system will not easily migrate tasks to platforms with high latency, thus improving the user experience.

[0124] In the above formula, E(P) represents the energy function value of the current path; x i is a binary variable indicating whether to allocate the task to platform i; λ is the weight coefficient of migration loss, and loss ij represents the loss of migrating the task from cloud caching platform i to cloud caching platform j; W1 and W2 represent the weight matrices used to adjust the relationship between platform characteristics; when the task migrates from cloud caching platform i to cloud caching platform j, the indicator function otherwise d represents the task data size, d BM represents the reference data volume, c represents the network traffic cost of cloud caching platform j after migration, c BM represents the reference cost, (l j -l i ) represents the latency change amount from cloud caching platform i to cloud caching platform j, l BM represents the reference latency; β1, β2, and β3 are weight parameters used to balance the influence of different loss terms.

[0125] Randomly select a new path P′ from the neighborhood of the current path, and calculate the acceptance probability of the new path P′:

[0126]

[0127] In the formula, P(accept) represents the acceptance probability; T g represents the annealing temperature at the g-th iteration, which is used to control the probability of accepting a worse solution.

[0128] Here, in each iteration process, the simulated annealing algorithm randomly accepts sub-optimal solutions according to the current temperature. Thus, it ensures that the scheduling system has a certain degree of flexibility and fault tolerance. When facing sudden changes in the operating state, it can explore more scheduling schemes and reduce system bottlenecks caused by resource tension.

[0129] Perform a temperature reduction operation after each iteration:

[0130] T g+1 =γ·T g , Equation (11)

[0131] In the formula, γ represents the temperature reduction coefficient.

[0132] Here, through the above temperature reduction operation, it can ensure that the algorithm fully explores in the early stage and gradually converges to the optimal solution in the later stage, improving the stability of the scheduling system, and can reduce invalid migrations and task blockages, thus responding to user requests faster.

[0133] When the annealing temperature drops to T min , terminate the continued iteration and output the optimal cross-region data collaboration path:

[0134] P terget =argmin P {E(P)}, Equation (12)

[0135] In the formula, P terget represents the cross-region data collaboration path, and argmin P {E(P)} represents the path that minimizes the energy function value in multiple iterations.

[0136] In this way, through multiple iterations and neighborhood search, the simulated annealing algorithm avoids the system falling into a local optimum, ensuring that the selected path P target is the global optimum solution, realizing the dynamic selection of the optimal task allocation scheme among multiple cloud caching platforms, and ensuring that the resource utilization rate and task response time reach the best.

[0137] It should be noted that the weights λ and β1, β2, β3 can be flexibly adjusted according to different task types and business requirements. For example, giving priority to platforms with low latency or preferring to select platforms with sufficient computing resources, etc., to achieve flexible scheduling, which can ensure that the system can provide optimized task scheduling schemes in different scenarios.

[0138] Through the simulated annealing algorithm provided by the embodiments of the present application, task scheduling can be performed globally in advance according to the prediction results, reducing the number and loss of task migrations, avoiding service interruptions caused by frequent migrations, and improving the reliability of multi-platform intelligent scheduling results. In addition, based on the design of the above objective function or energy function, comprehensively weighing the resource consumption of the current task execution and the potential loss of future task migrations, and based on the design of the loss function, the best balance can be found between performance and cost, ensuring the efficient operation of the system in a complex multi-platform environment.

[0139] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of actions combined. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application. In the above embodiments, the descriptions of each embodiment have their own focuses. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0140] Figure 5 The structural block diagram of an example of a cross-regional data collaborative computing system based on a multi-level cache mechanism according to an embodiment of the present application is shown.

[0141] As Figure 5 shown, the cross-regional data collaborative computing system 500 based on the multi-level cache mechanism includes a data acquisition unit 510, a collaborative path matching unit 520, a task image synchronization unit 530, and a platform monitoring and switching unit 540.

[0142] The data acquisition unit 510 is used to acquire the historical operation status sequence of each cloud cache platform and the task description data of the target computing task to be collaboratively calculated; the historical operation status sequence includes multiple historical time windows and corresponding historical operation status data, and the parameter types of the historical operation status data include cache hit rate, cache latency, cache capacity utilization rate, computing resource utilization rate, and network transmission latency; the task description data includes task data size, task data cache requirement, task computing resource requirement, task type, and task processing time limit.

[0143] The collaborative path matching unit 520 is used to input the task description data and the historical operation status sequence of each cloud cache platform into the platform scheduling model to output a matching cross-regional data collaborative path; the cross-regional data collaborative path defines multiple potential cloud cache platforms arranged in descending order of priority; the platform scheduling model is a deep learning model.

[0144] The task image synchronization unit 530 is used to package the target computing task and its dependent environment to generate a corresponding task container image, and send the task container image to each of the potential cloud caching platforms.

[0145] The platform monitoring and switching unit 540 is used to control the first potential caching cloud caching platform to run the corresponding task container image, monitor the real-time running state of the first potential cloud caching platform, and switch the control to the second potential cloud caching platform to run the corresponding task container image when the monitored real-time running state meets the task migration condition; the priority of the first potential cloud caching platform is higher than that of the second potential caching cloud caching platform.

[0146] In some embodiments, the present application provides a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to perform the steps of any one of the above cross-region data collaborative computing methods based on a multi-level caching mechanism.

[0147] In some embodiments, the present application also provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer is enabled to perform the steps of any one of the above cross-region data collaborative computing methods based on a multi-level caching mechanism.

[0148] In some embodiments, the present application also provides an electronic device, which includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the cross-region data collaborative computing method based on a multi-level caching mechanism.

[0149] Figure 6 is a schematic hardware structure diagram of an electronic device for performing the cross-region data collaborative computing method based on a multi-level caching mechanism provided by another embodiment of the present application, as Figure 6 shown, the device includes:

[0150] One or more processors 610 and a memory 620, Figure 6 Taking one processor 610 as an example.

[0151] The device for performing the cross-region data collaborative computing method based on a multi-level caching mechanism may further include: an input device 630 and an output device 640.

[0152] The processor 610, the memory 620, the input device 630, and the output device 640 can be connected via a bus or other means. Figure 6 Taking the connection via the bus as an example.

[0153] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the cross-region data collaborative computing method based on the multi-level cache mechanism in the embodiments of the present application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, that is, implements the cross-region data collaborative computing method based on the multi-level cache mechanism in the above method embodiments.

[0154] The memory 620 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the electronic device, etc. In addition, the memory 620 can include high-speed random access memory, and can also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 620 can optionally include a memory remotely set relative to the processor 610, and these remote memories can be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0155] The input device 630 can receive input digital or character information, and generate signals related to the user settings and function control of the electronic device. The output device 640 can include display devices such as a display screen.

[0156] The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, execute the cross-region data collaborative computing method based on the multi-level cache mechanism in any of the above method embodiments.

[0157] The above product can execute the method provided by the embodiments of the present application, and has the corresponding functional modules and beneficial effects of the executed method. For technical details not described in detail in this embodiment, reference can be made to the method provided by the embodiments of the present application.

[0158] The electronic device in the embodiments of the present application exists in various forms, including but not limited to:

[0159] (1) Mobile communication devices: The characteristic of such devices is that they have mobile communication functions and mainly aim to provide voice and data communication. Such terminals include: smart phones, multimedia phones, functional phones, and low-end phones, etc.

[0160] (2) Ultra-mobile personal computer devices: These devices fall within the category of personal computers, have computing and processing capabilities, and generally also have the characteristic of mobile Internet access. Such terminals include: PDA, MID, and UMPC devices, etc.

[0161] (3) Portable entertainment devices: These devices can display and play multimedia content. Such devices include: audio and video players, handheld game consoles, e-books, as well as smart toys and portable in-vehicle navigation devices.

[0162] (4) Other on-board electronic devices with data interaction functions, such as in-vehicle device installed on vehicles.

[0163] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0164] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solutions, or the part that contributes to the relevant technologies, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0165] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. A cross - regional data collaborative computing method based on a multi - level caching mechanism, comprising: Obtaining the historical operation state sequences of each cloud caching platform and the task description data of the target computing task to be collaboratively computed; The historical operation state sequences include multiple historical time windows and corresponding historical operation state data, and the parameter types of the historical operation state data include cache hit rate, cache latency, cache capacity utilization rate, computing resource utilization rate, and network transmission latency; the task description data includes task data size, task data caching requirements, task computing resource requirements, task type, and task processing time limit; Inputting the task description data and the historical operation state sequences of each cloud caching platform into a platform scheduling model to output a matching cross - regional data collaborative path; The cross - regional data collaborative path defines multiple potential cloud caching platforms arranged in descending order of priority; The platform scheduling model is a deep learning model; Packing the target computing task and its dependent environment to generate a corresponding task container image, and sending the task container image to each of the potential cloud caching platforms; Controlling the first potential cloud caching platform to run the corresponding task container image, monitoring the real - time operation state of the first potential cloud caching platform, and switching to control the second potential cloud caching platform to run the corresponding task container image when the monitored real - time operation state meets the task migration condition; The priority of the first potential cloud caching platform is higher than that of the second potential cloud caching platform.

2. The method according to claim 1, wherein During the process of the first potential cloud caching platform running the task container image, the first potential cloud caching platform generates a task checkpoint and synchronizes the task checkpoint among each of the potential cloud caching platforms; the task checkpoint is used to store the intermediate execution state data of the target computing task; Among them, when it is monitored that the first operation state data meets the task migration condition, switch to control the second potential cloud caching platform to run the corresponding task container image according to the task checkpoint.

3. The method according to claim 1, wherein, The deep learning model includes an LSTM layer, a feature splicing layer, and a simulated annealing matching layer; The LSTM layer is used to respectively infer the predicted operation state data of the corresponding cloud caching platform in the next time window according to each of the historical operation state data; The feature splicing layer is used to splice the task description data and each of the predicted operation state data to obtain the corresponding splicing feature vectors of each cloud caching platform, so as to construct an estimated operation state matrix; The simulated annealing matching layer is used to solve the estimated operation state matrix through a simulated annealing algorithm to generate a cross - regional data collaborative path.

4. The method according to claim 3, wherein The LSTM layer is an LSTM layer based on an attention mechanism: h i,t ,c i,t = LSTM(x i,t ,h i,t-1 ,c i,t-1 ), α i,t = Softmax(W a h i,t ), where \(i\) represents the \(i\)-th cloud caching platform, \(t\) represents the \(t\)-th time window, and \(x\) i,t represents the input vector of the operating state of the \(i\)-th cloud caching platform in the \(t\)-th time window, and \(h\) i,t represents the hidden state of the \(i\)-th cloud caching platform in the \(t\)-th time window, and \(c\) i,t represents the LSTM cell state of the \(i\)-th cloud caching platform in the \(t\) time window; \(\alpha\) i,t is the attention weight of the \(i\)-th cloud caching platform in the \(t\)-th time window, indicating the importance at this moment; \(W\) a is the weight matrix of the attention mechanism; \(N\) represents the total number of time windows corresponding to the historical operating state sequence; is the weighted hidden state of the \(i\)-th cloud caching platform, representing the feature representation after the attention mechanism; represents the predicted operating state of the \(i\)-th cloud caching platform in the \((t + 1)\)-th time window.

5. The method according to claim 4, wherein The feature splicing layer is a non - linear activation layer: Where, F i represents the splicing feature vector of the i-th cloud caching platform, D represents the task data size, R represents the task computing resource requirement, L represents the task type, and S represents the task processing time limit; W F and b F respectively represent the weight matrix and the bias term of the non-linear activation layer, M represents the estimated operating state matrix, represents the feature vector corresponding to the i-th cloud caching platform in M, and n represents the number of cloud caching platforms.

6. The method according to claim 5, wherein, The simulated annealing matching layer is used to perform the following operations: Randomly generate an initial path P0 according to the estimated operating state matrix, and perform annealing iteration operations based on the initial path P0 and the initial temperature T max Perform annealing iteration operations; In each iteration, calculate the objective function value of the current path P; The expression of the objective function is: where \(E(P)\) represents the energy function value of the current path; \(x\) i is a binary variable indicating whether to assign the task to platform \(i\); λ is the weight coefficient of the migration loss, loss ij represents the loss of task migration from cloud caching platform i to cloud caching platform j; W1, W2 represent weight matrices for adjusting the relationship between platform features; When a task is migrated from cloud caching platform i to cloud caching platform j, the indication function Otherwise d represents the task data size, d BM represents the reference data volume, c represents the network traffic cost of cloud caching platform j after migration, c BM represents the reference cost, (l j -l i ) represents the change in latency from cloud caching platform i to cloud caching platform j, l BM represents the reference latency; β1, β2, β3 are weight parameters used to balance the impacts of different loss terms; Randomly select a new path P′ from the neighborhood of the current path and calculate the acceptance probability of the new path P′; where P(accept) represents the acceptance probability; T g represents the annealing temperature at the g-th iteration, which is used to control the probability of accepting a worse solution; Perform a cooling operation after each iteration: T g+1 = γ·T g , Wherein, γ represents the cooling coefficient; When the annealing temperature drops to T min , stop further iteration and output the optimal cross-region data collaboration path: P terget = argmin P {E(P)}, where P terget represents the cross - regional data collaboration path, and argmin P {E(P)} represents the path that minimizes the energy function value in multiple iterations.

7. The method according to claim 1, wherein The step of packaging the target computing task and its dependent environment to generate a corresponding task container image and sending the task container image to each of the potential cloud caching platforms includes: Parsing the task code, dependent packages, and environment configuration corresponding to the target computing task; Invoking a container image tool to package the parsed task code, dependent packages, and environment configuration to generate a corresponding initial image file; Encrypting the initial image file to obtain a corresponding task container image, and sending the task container image to each of the potential cloud caching platforms.

8. The method according to claim 7, wherein, The step of encrypting the initial image file to obtain a corresponding task container image and sending the task container image to each of the potential cloud caching platforms includes: Parsing the base layer data and application layer data in the initial image file, and obtaining a random session key and a cloud caching platform public key for each potential cloud caching platform; Symmetrically encrypting the application layer data according to each of the random session keys to generate corresponding encrypted application layer data; For each potential cloud caching platform, asymmetrically encrypting the corresponding random session key using the cloud caching platform public key of the potential cloud caching platform to obtain a corresponding encrypted session key, and packaging the base layer data, the encrypted application layer data, and the encrypted session key into a corresponding encrypted transmission packet; Sending each of the encrypted transmission packets to the corresponding potential cloud caching platform, so that the potential cloud caching platform decrypts the encrypted session key in the encrypted transmission packet based on its own cloud caching platform private key to crack the application layer data, and thus restores the complete task container image according to the cracked application layer data and the base layer data in the encrypted transmission packet.

9. The method according to claim 8, wherein The cloud caching platform public key and the matching cloud caching platform private key for each potential cloud caching platform are a corresponding pair of independent elliptic curve public-private key pairs.

10. A cross-region data collaborative computing system based on a multi-level caching mechanism, including: A data acquisition unit, configured to acquire the historical operation state sequence of each cloud caching platform and the task description data of the target computing task to be collaboratively computed; The historical operation state sequence includes a plurality of historical time windows and corresponding historical operation state data, and the parameter types of the historical operation state data include cache hit rate, cache latency, cache capacity utilization rate, computing resource utilization rate, and network transmission latency; the task description data includes task data size, task data caching requirement, task computing resource requirement, task type, and task processing time limit; A collaborative path matching unit, configured to input the task description data and the historical operation state sequence of each cloud caching platform into a platform scheduling model to output a matching cross-region data collaborative path; The cross-region data collaborative path defines a plurality of potential cloud caching platforms arranged in descending order of priority; The platform scheduling model is a deep learning model; The task image synchronization unit is used to package the target computing task and its dependent environment to generate a corresponding task container image, and send the task container image to each of the potential cloud caching platforms; The platform monitoring and switching unit is used to control the first potential caching cloud caching platform to run the corresponding task container image, monitor the real-time running state of the first potential cloud caching platform, and switch and control the second potential cloud caching platform to run the corresponding task container image when the monitored real-time running state meets the task migration condition; The priority of the first potential cloud caching platform is higher than that of the second potential caching cloud caching platform.

Images