Data processing method in edge-cloud collaborative system and edge-cloud collaborative system

Abstract

The application provides a data processing method in an edge cloud cooperative system and the edge cloud cooperative system. The method comprises the following steps: configuring a dynamic resource influence interface and a static resource influence interface on a node in the edge cloud cooperative system; the node in the edge cloud cooperative system uses the dynamic resource influence interface to realize real-time sensing of application load and resource usage, and adjusts computing resources, storage resources and network resources according to the sensed application load and resource usage; the node in the edge cloud cooperative system uses the static resource influence interface to configure and manage infrastructure resources of the system; and the node in the edge cloud cooperative system obtains memory usage of an application process, and recycles and releases the memory according to the memory usage. The application can better utilize resources of the edge cloud cooperative system.

Description

TECHNICAL FIELD

[0001] One or more embodiments of the present application relate to network communication technology, and in particular to a data processing method in an edge-cloud collaborative system and the edge-cloud collaborative system. BACKGROUND

[0002] Edge-cloud collaboration, as a new computing mode, has become a new research hotspot in the academic and industrial fields. Edge computing is a new computing mode that performs computing at the network edge. The edge-cloud collaborative system is a system that adopts the edge computing mode. The downlink data in the edge-cloud collaborative system represents cloud service data, and the uplink data represents Internet of Everything service data. The edge of edge computing refers to any computing and network resources between the data source and the cloud service center.

[0003] In the edge-cloud collaborative system, user data no longer needs to be uploaded to the cloud service center in its entirety, but is quickly processed by the edge nodes deployed at the network edge, thereby greatly reducing the pressure on the network bandwidth, the pressure on the cloud service center, and the energy consumption of the intelligent devices at the network edge. The edge-cloud collaborative system has become an indispensable key link of the industrial Internet and a core supporting technology for realizing the digital transformation of the industry.

[0004] The resources and tasks in the cloud-edge collaborative system are dynamically changing and have a large scale. How to maximize the utilization of the resources of the cloud-edge collaborative system is a problem to be solved. SUMMARY

[0005] One or more embodiments of the present application describe a data processing method in an edge-cloud collaborative system and the edge-cloud collaborative system, which can better utilize the resources of the cloud-edge collaborative system.

[0006] According to a first aspect, a data processing method in an edge-cloud collaborative system is provided, the method comprising:

[0007] configuring a dynamic resource impact interface and a static resource impact interface on a node in the edge-cloud collaborative system;

[0008] The node in the edge-cloud collaborative system utilizes the dynamic resource impact interface of itself to perceive the application load and resource usage in real time, and adjusts the computing resources, storage resources, and network resources according to the perceived application load and resource usage;

[0009] The node in the edge-cloud collaborative system utilizes the static resource impact interface of itself to configure and manage the infrastructure resources of the system;

[0010] The node in the edge-cloud collaborative system obtains the memory usage of an application process, and recycles and releases the memory according to the memory usage.

[0011] The node in the edge cloud coordination system utilizes its static resource to affect the infrastructure resource of the interface configuration and management system, including:

[0012] In the edge cloud coordination system, the storage resources of multiple nodes are integrated to form a unified storage resource pool. When a reasoning service runs on any one of the multiple nodes, the static resource affecting interface on the node accesses data in the entire storage resource pool to realize network storage sharing.

[0013] Or,

[0014] In the edge cloud coordination system, when the resources on a node are insufficient, the static resource affecting interface is used to borrow resources from other nodes to meet its own needs; when the node has excess resources, the static resource affecting interface is used to contribute the excess resources to other nodes for use.

[0015] Or,

[0016] A unique SHA-256 hash value is assigned to the resource in the edge cloud coordination system as an identifier of the resource. After receiving an update package for the resource through the static resource affecting interface, the SHA-256 hash value of the update package of the resource is calculated, and the calculated SHA-256 hash value of the update package of the resource is compared with the SHA-256 hash value assigned to the resource. If they are the same, the resource is not updated, and if they are different, the resource is updated.

[0017] The method further includes:

[0018] The node dynamically increases or decreases the number of service instances according to real-time load conditions. When the system load increases, a new service instance is started to handle additional reasoning tasks; when the load decreases, part of the service instances are closed to release resources, so as to realize the service-free reasoning resource architecture.

[0019] And / or,

[0020] The application and its dependencies are packaged into independent containers to realize the isolation, portability and manageability of the application.

[0021] And / or,

[0022] A complex system is divided into a series of small, independent services, each of which completes a specific function.

[0023] Through real-time monitoring and analysis, the CPU utilization, memory usage, disk space utilization and network connection state of the node are collected, and the future resource demand is predicted by using deep learning algorithm and the collected CPU utilization, memory usage, disk space utilization and network connection state of the node.

[0024] The node in the edge cloud cooperative system acquires memory usage of an application process, and recycles and releases memory according to the memory usage, including:

[0025] The node in the edge cloud cooperative system performs, for each memory block: collecting access frequency to the memory block, last access time to the memory block, and life cycle of the memory block;

[0026] According to the collected access frequency, last access time, and life cycle of each memory block, an importance score is calculated for each memory block;

[0027] Predicting future memory demand;

[0028] According to the predicted memory demand and the importance score of each memory block, starting from the memory block with a low importance score, recycling is performed on the memory block based on a dynamically adjusted memory recycling strategy.

[0029] The importance score is calculated for each memory block, including:

[0030] The calculation is performed using the following calculation formula:

[0031] S(b)=w_a\cdot A(b)+w_t\cdot T(b)+w_l\cdot L(b)]

[0032] Wherein: S(b) represents the importance score of the memory block b; A(b) represents the access frequency to the memory block b; T(b) represents the last access time to the memory block b; L(b) represents the life cycle of the memory block b, which is the length of time from the creation of the memory block b to the current time; w_a, w_t, and w_l are the weight of the access frequency, the weight of the last access time, and the weight of the life cycle, respectively, and w_a+w_t+w_l=1.

[0033] The recycling of the memory block is performed based on the dynamically adjusted memory recycling strategy, including:

[0034] The recycling threshold of the memory is dynamically calculated;

[0035] When the current usage rate of the memory is greater than the newly calculated recycling threshold of the memory, recycling of the memory block is started;

[0036] Wherein, the recycling threshold of the memory is dynamically calculated using the following calculation formula:

[0037] T_{new}=T_{base}+k\cdot(U-U_{target})

[0038] Wherein: T new represents the latest recovery threshold; T base represents the pre-set base recovery threshold; U represents the current memory usage; U target represents the pre-set target memory usage; k is a pre-set adjustment coefficient, k reflects the response ability to memory pressure.

[0039] The predicted future memory requirement comprises:

[0040] The future memory requirement is predicted using a linear regression model:

[0041] D(t) = β 0 + β 1 • D(t-1) + β 2 • A(t-1) + ε

[0042] Wherein:

[0043] D(t) represents the memory requirement at time (t);

[0044] D(t-1) represents the memory requirement at the previous time;

[0045] A(t-1) represents the memory access amount at the previous time;

[0046] β 0, β 1, β 2 are parameters of the linear regression model;

[0047] ε is a set error term.

[0048] According to the second aspect, a side cloud collaborative system is provided, comprising: at least two edge nodes and a cloud service center node,

[0049] At least one of the at least two edge nodes and the cloud service center node is configured with a dynamic resource influencing interface and a static resource influencing interface;

[0050] The node uses its own dynamic resource influencing interface to perceive application load and resource usage in real time, and adjusts computing resources, storage resources and network resources according to the perceived application load and resource usage;

[0051] The node uses its own static resource influencing interface to configure and manage the infrastructure resources of the system;

[0052] At least one of the at least two edge nodes and the cloud service center node obtains the memory usage of the application process, and recovers and releases the memory according to the memory usage.

[0053] According to a third aspect, a computing device is provided, comprising a memory and a processor, the memory having stored therein executable code which, when executed by the processor, implements the method of any of the embodiments of the application.

[0054] As can be seen, the various embodiments of the application have at least the following beneficial effects:

[0055] (1) By real-time sensing of application load and resource usage, automatic adjustment of resource allocation strategy, and unified management and efficient utilization of infrastructure resources, efficient operation and high-quality service of the system are ensured.

[0056] (2) The complexity and operation and maintenance cost of the system are reduced, and the utilization rate of resources and the overall performance of the system are improved.

[0057] (3) The resource response system based on device physical principle processing is an efficient and reliable system resource management scheme. It not only can realize the accurate control and optimization of system resources, but also can reduce the system failure rate and maintenance cost, and provide strong support for the digital transformation of enterprises.

[0058] (4) Effectively solves the memory leakage problem that may occur in intelligent reasoning service, improves the reliability and stability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0060] Figure 1 is a schematic diagram of the system architecture to which an embodiment of the application is applied.

[0061] Figure 2 is a flowchart of the data processing method in the edge cloud collaborative system in an embodiment of the application.

[0062] Figure 3 is a schematic diagram of classifying and obtaining static resources and dynamic resources based on static resource influence interfaces and dynamic resource influence interfaces respectively in an embodiment of the application.

[0063] Figure 4 is a flowchart of the service-free processing process based on the service-free reasoning resource architecture in an embodiment of the application. DETAILED DESCRIPTION

[0064] The schemes provided by the application will be described below with reference to the drawings.

[0065] First of all, it should be noted that the terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0066] It should be understood that the term "and / or" used herein is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents that the front and rear associated objects are a "or" relationship.

[0067] In order to facilitate the understanding of the method provided by the present application, first, the system architecture related to and applicable to the present application is described. As shown in Figure 1 The system includes at least two edge nodes and a cloud service center node.

[0068] It should be understood that Figure 1 The number of edge nodes and cloud service center nodes in

[0069] Figure 2 is a flowchart of a data processing method in an edge-cloud collaborative system in an embodiment of the present application. Referring to Figure 1 and Figure 2 , the method comprises:

[0070] Step 201: configuring a dynamic resource influence interface and a static resource influence interface on the nodes in the edge-cloud collaborative system;

[0071] Step 203: the nodes in the edge-cloud collaborative system use their own dynamic resource influence interfaces to perceive application load and resource usage in real time, and adjust computing resources, storage resources and network resources according to the perceived application load and resource usage;

[0072] Step 205: the nodes in the edge-cloud collaborative system use their own static resource influence interfaces to configure and manage the infrastructure resources of the system;

[0073] Step 207: the nodes in the edge-cloud collaborative system obtain the memory usage of the application process, and recycle and release the memory according to the memory usage.

[0074] According to the above Figure 2As can be seen from the flow, in the embodiment of the application, a dynamic resource influence interface and a static resource influence interface are configured on the node of the edge cloud cooperative system, the two interfaces jointly constitute the basis of system resource management, and ensure that the system can be adaptively adjusted according to application requirements and environmental changes to achieve the best running efficiency and performance. Moreover, memory recovery and release are performed based on memory usage, and the memory usage efficiency is improved.

[0075] In an embodiment of the application, in order to solve the problem of resource use and fatigue management that exists universally in the computing power cloud platform, the following four aspects of technology can be used:

[0076] 1. Constructing a dynamic and static resource influence interface;

[0077] 2. Constructing a reasoning resource architecture without service;

[0078] 3. Constructing a resource response system based on core physical principle processing;

[0079] 4. A memory self-healing management system based on memory leakage.

[0080] Firstly, the first point, constructing a dynamic and static resource influence interface, is explained.

[0081] In the prior art, the interface types of nodes are not distinguished, and all interfaces have unified functions, for example, the information transmitted by each interface often needs to be encrypted or token provided, which increases the processing complexity.

[0082] Reference is made to Figure 3In a cloud-edge collaborative system, such as when building an intelligent software inference service system, a core task is to design and implement dynamic resource impact interfaces and static resource impact interfaces that can flexibly adjust resource allocation. These two interfaces together form the foundation of system resource management, ensuring that the system can adapt to application requirements and environmental changes to achieve optimal running efficiency and performance. First, for the dynamic resource impact interface. The main responsibility of this interface is to perceive application load and resource usage in real time and automatically adjust computing resources, storage resources, and network resources accordingly. In modern cloud computing environments, application load often exhibits high dynamicity and uncertainty, which requires the system to have the ability to respond quickly and adaptively. The dynamic resource impact interface integrates advanced monitoring technology and intelligent scheduling algorithms to collect and analyze system performance indicators such as CPU usage, memory usage, and network bandwidth in real time, and dynamically adjusts resource allocation strategies based on changes in these indicators. For example, when an application load suddenly increases, the interface can quickly identify this change and automatically increase the computing resources allocated to the application to ensure that it can run normally and meet user needs. This automated resource adjustment mechanism not only greatly improves the system's response speed and running efficiency, but also significantly reduces operational costs. Next, for the static resource impact interface. Unlike the dynamic resource impact interface, the static resource impact interface is mainly responsible for configuring and managing system infrastructure resources such as servers, storage devices, and network devices. These infrastructure resources are the foundation of system operation, and their performance and stability directly affect the efficiency and quality of inference services. The static resource impact interface uses network storage sharing and P2P resource sharing technology to achieve unified management and efficient use of infrastructure resources. In terms of network storage sharing, the system can integrate the storage resources of each node to form a unified storage resource pool. This way, no matter where the inference service is running, it can easily access data in the entire storage resource pool, greatly improving data availability and sharing. In terms of P2P resource sharing, the system further improves resource utilization and system scalability by enabling resource assistance and sharing between nodes. Specifically, when a node's resources are insufficient, it can borrow resources from other nodes to meet its own needs; when it has excess resources, it can contribute them to other nodes for use. This P2P-based resource sharing model not only reduces the overall cost of the system but also improves resource utilization and overall system performance. In the static resource impact interface, resource matching and management is an important link. To ensure resource accuracy and efficiency, the system uses a resource matching mechanism based on the SHA-256 algorithm. The SHA-256 algorithm is a secure hash algorithm that can map data of any length to a fixed-length hash value.In the system, each resource (such as server, storage device, network device, etc.) is assigned a unique SHA-256 hash value as an identifier. When the system needs to find or access a certain resource, it only needs to find according to the hash value of the resource. This kind of resource matching method based on hash value can greatly improve the efficiency and accuracy of the system, so as to ensure the quick response and effective use of resources. In summary, dynamic and static resource impact interface is an important part of building intelligent software reasoning service system based on cloud edge collaboration. They can automatically adjust resource allocation strategy by real-time sensing of application load and resource usage, and realize unified management and efficient use of infrastructure resources, ensuring efficient operation and high-quality service of the system.

[0083] As can be seen, in the embodiment of the application, in the step 205, the node in the edge cloud collaborative system configures and manages the infrastructure resources of the system by using its static resource impact interface, including:

[0084] In the edge cloud collaborative system, the storage resources of multiple nodes are integrated to form a unified storage resource pool. When a reasoning service runs on any one of the multiple nodes, the data in the entire storage resource pool is accessed through the static resource impact interface on the node to realize network storage sharing.

[0085] Or,

[0086] In the edge cloud collaborative system, when the resources on a node are insufficient, the static resource impact interface is used to borrow resources from other nodes to meet its own needs; when the node has excess resources, the static resource impact interface is used to contribute the excess resources to other nodes for use.

[0087] Or,

[0088] A unique SHA-256 hash value is assigned to the resource in the edge cloud collaborative system as an identifier of the resource. After receiving the update package for the resource through the static resource impact interface, the SHA-256 hash value of the update package of the resource is calculated, and the calculated SHA-256 hash value of the update package of the resource is compared with the SHA-256 hash value assigned to the resource. If they are the same, the resource is not updated, and if they are different, the resource is updated.

[0089] The following explains point 2: building a service-free reasoning resource architecture.

[0090] Referring to Figure 4In building a high-performance and resource-efficient intelligent software inference service system, the service-free inference resource architecture becomes a key solution. This architecture aims to reduce system complexity and improve resource utilization to meet growing computing demands. In one embodiment of the invention, the service-free inference resource architecture can implement on-demand allocation, dynamic, non-fixed, high-concurrency processing threads and instances. By integrating containerization technology and microservice architecture, it realizes the rapid deployment, flexible expansion and seamless update of inference services, providing strong support for intelligent applications. First, the core of the service-free inference resource architecture lies in the application of containerization technology. Containerization technology packages application programs and their dependencies into independent containers, achieving application isolation, portability and manageability. This means that each inference service can run independently in a container without affecting other services, while also facilitating the update and upgrade of individual services. In addition, containerization technology makes it easier to dynamically allocate and recycle resources, thereby improving resource utilization. Second, the introduction of microservice architecture further enhances the flexibility and scalability of the service-free inference resource architecture. Microservice architecture divides complex systems into a series of small, independent services, each focusing on completing a specific function. This architecture pattern makes the system easier to understand and maintain, while also improving system scalability and fault tolerance. In inference services, microservice architecture allows us to quickly increase or decrease the number of service instances according to actual needs to meet different computing demands. To further improve the overall performance and response speed of the system, the service-free inference resource architecture also uses intelligent scheduling algorithms. This algorithm can real-time perceive the system load and dynamically adjust the allocation strategy of inference tasks according to the load. Through load balancing technology, the system can evenly distribute inference tasks to each service instance, avoiding the situation where some service instances are overloaded while others are idle. At the same time, the intelligent scheduling algorithm can dynamically adjust the resource allocation strategy based on historical data and prediction models to respond to future load changes. In the service-free inference resource architecture, the dynamic scaling resource mechanism of the service-free interface is the key to achieving high concurrency and memory optimization. This mechanism allows the system to dynamically increase or decrease the number of service instances according to real-time load. When the system load increases, the service-free interface can quickly start new service instances to handle additional inference tasks; when the load decreases, it can close some service instances to release resources. This dynamic scaling mechanism not only ensures that the system can handle high concurrency demands, but also avoids wasting and leaking memory resources. In addition, the service-free inference resource architecture does not need to save process data during inference, thereby maximizing resource utilization. In traditional inference services, the system usually needs to save a large amount of intermediate data and state information to support subsequent inference processes.However, in the service-free inference resource architecture, each service instance can independently process inference tasks without relying on other service instances or saving intermediate data due to the adoption of stateless design ideas, which makes the system more efficient in utilizing computing resources, and improves the scalability and fault tolerance of the system. In summary, the service-free inference resource architecture integrates containerization technology, micro-service architecture, and intelligent scheduling algorithms to achieve fast deployment, flexible expansion, and efficient operation of inference services. This architecture not only reduces the complexity and operation and maintenance costs of the system, but also improves the utilization of resources and the overall performance of the system.

[0091] Therefore, in the embodiments of the present application, the data processing method based on the edge cloud cooperative system can further include:

[0092] The number of service instances is dynamically increased or decreased according to the real-time load condition, when the system load increases, a new service instance is started to process additional inference tasks, and when the load decreases, part of the service instances are closed to release resources, to realize the service-free inference resource architecture;

[0093] And / or,

[0094] Packaging the application and its dependencies into independent containers to achieve the isolation, portability, and manageability of the application;

[0095] And / or,

[0096] Splitting a complex system into a series of small, independent services, each service completing a specific function.

[0097] The following explains point 3: building a resource response system based on core physical principle processing.

[0098] Building a resource response system based on device physical principle processing is an important way to achieve accurate control and optimization of system resources. This system is based on real-time monitoring and analysis of system operation status, combined with physical principles and mathematical models, to predict, schedule and configure system resources, ensuring efficient and stable operation of the system. First, the system collects various operating data of the system, such as CPU utilization, memory usage, disk space utilization, network connection status, etc. through real-time monitoring and analysis technology. These data are the key to understanding the current status of the system and the basis for subsequent prediction and scheduling. Second, the system processes and analyzes the collected data in combination with physical principles and mathematical models. Through the fusion of deep learning algorithms and hardware physical structure, the system can better understand the working principle and failure mode of the system, thus predicting future resource demand. This prediction is not a simple trend analysis, but a deep insight based on the physical characteristics of the system, which can more accurately reflect the real demand of the system. Based on the prediction results, the system can perform resource scheduling and configuration in advance. For example, when predicting that a service will face high concurrent requests, the system can automatically increase the corresponding computing resources to ensure the stable operation of the service. Similarly, when the load of a node is too low, the system can transfer the resources on it to other places that need them, achieving global optimization of resources. This resource response system based on device physical principle processing not only improves the utilization and stability of system resources, but also reduces system failure rate and maintenance cost. Through real-time prediction and scheduling, the system can timely discover and solve potential problems, avoiding the occurrence of faults. At the same time, since the system can better understand the physical characteristics of the system, it can quickly locate the problem and take appropriate repair measures when a fault occurs. In addition, the fusion of deep learning algorithms and hardware physical structure enables the system to better solve fault problems. Deep learning algorithms can learn the failure mode of the system from a large amount of data and predict future failure trends. While the hardware physical structure provides the physical basis for solving the fault. By combining the two, the system can achieve accurate prediction and rapid repair of faults, further improving the reliability and availability of the system. In summary, the resource response system based on device physical principle processing is an efficient and reliable system resource management solution. It not only achieves accurate control and optimization of system resources, but also reduces system failure rate and maintenance cost, providing strong support for the digital transformation of enterprises.

[0099] Therefore, based on the above third point, it can be concluded that in the embodiment of the present application, the CPU utilization, memory usage, disk space utilization, and network connection status of the node can be collected through real-time monitoring and analysis, and the future resource demand can be predicted by using the deep learning algorithm and the collected CPU utilization, memory usage, disk space utilization, and network connection status of the node.

[0100] The following explains point 4: memory self-healing management system based on memory leakage.

[0101] To solve the problem of memory leakage that may occur in intelligent reasoning services, the present application proposes an innovative memory self-healing management system based on memory leakage. This system uses a service interface to count resource errors, thereby achieving real-time monitoring and automatic processing of memory leakage. In intelligent reasoning services, memory leakage is a common problem that can cause system performance degradation, response time extension, and even service crashes. Traditional memory management methods often require human intervention, which is not only inefficient but also difficult to find potential memory leakage problems. To overcome these limitations, the memory self-healing management system in this paper has emerged. The system first counts resource errors through a service interface to achieve real-time monitoring of application process memory usage. This service interface design allows the system to obtain application process memory usage data in real time and accurately analyze it. Once the system detects memory usage anomalies, such as high memory occupancy or insufficient memory release, it will immediately trigger the memory leakage warning mechanism. After discovering potential memory leakage problems, the memory self-healing management system will automatically recycle and release memory. Through intelligent algorithms and strategies, the system can accurately determine which memory blocks can be recycled and which need to be retained. Then, the system will automatically release those no longer needed memory blocks to release system resources and prevent further deterioration of memory leakage. In addition to automatically recycling and releasing memory, the system also provides memory leakage analysis and diagnosis tools. These tools can help developers quickly locate and solve memory leakage problems. By deeply analyzing memory usage data and code logic, developers can identify the root cause of memory leakage and take appropriate remedial measures. This ability to quickly locate and solve problems greatly improves the reliability and stability of the system. In addition, the memory self-healing management system is scalable and customizable. It can be flexibly configured and adjusted according to different application scenarios and requirements. For example, developers can set memory usage thresholds, recycling strategies, and other parameters according to actual conditions to meet the special needs of the system. This flexibility and customizability make the memory self-healing management system better adapt to various intelligent reasoning service scenarios. In summary, the memory self-healing management system based on memory leakage is an efficient, reliable, and flexible system resource management solution. It effectively solves the problem of memory leakage that may occur in intelligent reasoning services by monitoring application process memory usage in real time, automatically recycling and releasing memory, and providing memory leakage analysis and diagnosis tools, thereby improving the reliability and stability of the system.

[0102] Based on the above point 4, it can be concluded that in one embodiment of the present application, in step 207, the nodes in the edge-cloud collaborative system obtain the memory usage of the application process and recycle and release the memory according to the memory usage, including:

[0103] Step 2071: The node in the edge-cloud cooperative system performs, for each memory block: collecting the access frequency of the memory block, the last access time of the memory block, and the life cycle of the memory block;

[0104] Step 2073: According to the collected access frequency, last access time, and life cycle of each memory block, an importance score is calculated for each memory block;

[0105] Step 2075: Predicting future memory demand;

[0106] Step 2077: According to the predicted memory demand and the importance score of each memory block, starting from the memory block with a low importance score, the memory block is recycled based on a dynamically adjusted memory recycling strategy.

[0107] In an embodiment of the present application, in step 2073, the importance score is calculated for each memory block, including:

[0108] The calculation is performed using the following calculation formula:

[0109] S(b) = w_a * A(b) + w_t * T(b) + w_l * L(b)

[0110] Wherein: S(b) represents the importance score of the memory block b; A(b) represents the access frequency of the memory block b; T(b) represents the last access time of the memory block b; L(b) represents the life cycle of the memory block b, which is the length of time from the creation of the memory block b to the current time; w_a, w_t, and w_l are the weights of the access frequency, the last access time, and the life cycle, respectively, and w_a + w_t + w_l = 1.

[0111] In an embodiment of the present application, in step 2077, the memory block is recycled based on a dynamically adjusted memory recycling strategy, including:

[0112] Dynamically calculating the memory recycling threshold;

[0113] When the current usage of the memory is greater than the newly calculated memory recycling threshold, the memory block is recycled;

[0114] Wherein, the memory recycling threshold is dynamically calculated using the following calculation formula:

[0115] T_new = T_base + k * (U - U_target)

[0116] Wherein: T new represents the latest recovery threshold; T base represents the pre-set base recovery threshold; U represents the current memory usage; U target represents the pre-set target memory usage; k is a pre-set adjustment coefficient, k reflects the response ability to memory pressure.

[0117] In an embodiment of the present application, the prediction of future memory demand in step 2075 comprises:

[0118] The linear regression model is used to predict the future memory demand:

[0119] D(t) = beta_0 + beta_1 cdot D(t-1) + beta_2 cdot A(t-1) + epsilon

[0120] Wherein:

[0121] D(t) represents the memory demand at time (t);

[0122] D(t-1) represents the memory demand at the previous time;

[0123] A(t-1) represents the memory access amount at the previous time;

[0124] beta_0, beta_1, beta_2 are parameters of the linear regression model;

[0125] epsilon is a set error term.

[0126] In the embodiment of the present application, by implementing the above calculation formula, an intelligent memory recovery system can be constructed, the memory utilization rate can be effectively improved, the memory fragmentation can be reduced, and the overall system performance and response speed can be improved.

[0127] Referring to Figure 1 , the embodiment of the present application provides a side cloud cooperative system, which comprises: at least two edge nodes and a cloud service center node,

[0128] At least one node of the at least two edge nodes and the cloud service center node is configured with a dynamic resource influence interface and a static resource influence interface;

[0129] The node uses the dynamic resource influence interface of itself to perceive the application load and resource usage in real time, and adjusts the computing resource, the storage resource and the network resource according to the perceived application load and resource usage;

[0130] The node uses the static resource influence interface of itself to configure and manage the infrastructure resource of the system;

[0131] At least two edge nodes and at least one of the cloud service center nodes acquire memory usage of the application process, and memory is recycled and released according to the memory usage.

[0132] Therefore, the various embodiments of the present application have at least the following beneficial effects:

[0133] (1) By real-time sensing of application load and resource usage, automatic adjustment of resource allocation strategy, and unified management and efficient utilization of infrastructure resources, efficient operation and high-quality service of the system are ensured.

[0134] (2) The complexity and operation and maintenance cost of the system are reduced, and the utilization rate of resources and the overall performance of the system are improved.

[0135] (3) The resource response system based on device physical principle processing is an efficient and reliable system resource management scheme. It not only can realize accurate control and optimization of system resources, but also can reduce system failure rate and maintenance cost, and provide strong support for digital transformation of enterprises.

[0136] (4) Effectively solve the memory leak problem that may occur in intelligent reasoning service, improve the reliability and stability of the system.

[0137] One embodiment of the present application provides a computer readable storage medium, which stores a computer program, and when the computer program is executed in a computer, the computer executes the method in any one of the embodiments of the specification.

[0138] One embodiment of the present application provides a computing device, comprising a memory and a processor, the memory stores executable code, and when the processor executes the executable code, the method in any one of the embodiments of the specification is implemented.

[0139] It can be understood that the structure shown in the embodiments of the present application does not constitute a specific limitation on the device of the embodiments of the present application. In other embodiments of the specification, the above device can include more or fewer components than the illustration, or combine certain components, or split certain components, or different component arrangement. The components shown can be implemented in hardware, software or a combination of software and hardware.

[0140] Each of the embodiments in the present application is described in a progressive manner, and the same and similar parts of each embodiment can be referred to each other. Each embodiment mainly explains the difference from other embodiments. Especially, since the device embodiment is basically similar to the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the method embodiment.

[0141] Those skilled in the art should understand that, in one or more examples described above, the functions described in the present application can be implemented by hardware, software, a plug-in or any combination thereof. When implemented by software, the functions can be stored in a computer readable medium or transmitted as one or more instructions or codes on a computer readable medium.

[0142] The above detailed description sets forth the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above detailed description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made on the basis of the technical solutions of the present application shall be included in the protection scope of the present application.

Claims

1. A data processing method in an edge-cloud collaborative system, characterized in that, The method includes: Configure dynamic resource impact interfaces and static resource impact interfaces on nodes in the edge-cloud collaborative system; In the edge-cloud collaborative system, nodes use their own dynamic resource influence interfaces to perceive application load and resource usage in real time, and adjust computing resources, storage resources and network resources according to the perceived application load and resource usage. In an edge-cloud collaborative system, nodes can use their static resources to influence interface configuration and the infrastructure resources of the management system. Nodes in the edge-cloud collaboration system obtain the memory usage of application processes and reclaim and release memory based on the memory usage. Nodes in the edge-cloud collaborative system utilize their static resources to influence interface configuration and system infrastructure resources, including: In the edge-cloud collaborative system, the storage resources of multiple nodes are integrated to form a unified storage resource pool. When an inference service runs on any of the multiple nodes, it can access the data in the entire storage resource pool through the static resource influence interface on that node to achieve network storage sharing. or, In an edge-cloud collaborative system, when a node lacks resources, it borrows resources from other nodes through a static resource influence interface to meet its own needs; when the node has surplus resources, it contributes the surplus resources to other nodes through the static resource influence interface. or, A unique SHA-256 hash value is assigned to each resource in the edge-cloud collaborative system as the identifier of the resource. After receiving the update packet for the resource through the static resource impact interface, the SHA-256 hash value of the update packet for the resource is calculated. The calculated SHA-256 hash value of the update packet for the resource is compared with the SHA-256 hash value assigned to the resource. If they are the same, the resource is not updated. If they are different, the resource is updated. The method further includes: The node dynamically increases or decreases the number of service instances based on real-time load. When the system load increases, new service instances are started to handle additional inference tasks; when the load decreases, some service instances are shut down to release resources, thus achieving a serverless inference resource architecture. And / or, Package the application and its dependencies into independent containers to achieve application isolation, portability and manageability; And / or, Break down complex systems into a series of small, independent services, each of which performs a specific function; By monitoring and analyzing in real time, the CPU utilization, memory usage, disk space utilization, and network connection status of nodes are collected. Using deep learning algorithms and the collected data on node CPU utilization, memory usage, disk space utilization, and network connection status, future resource requirements are predicted. The nodes in the edge-cloud collaborative system acquire the memory usage of application processes and perform memory reclamation and release based on the memory usage, including: In the edge-cloud collaborative system, each node performs the following for each memory block: collecting the access frequency of the memory block, the last access time of the memory block, and the lifecycle of the memory block; Based on the collected access frequency, last access time, and lifespan of each memory block, an importance score is calculated for each memory block. Predict future memory demand; Based on the predicted memory demand and the importance score of each memory block, memory reclamation starts from the memory blocks with low importance scores, and memory blocks are reclaimed based on a dynamically adjusted memory reclamation strategy. The calculation of importance scores for each memory block includes: The calculation is performed using the following formula: S(b) = w_a \cdot A(b) + w_t \cdot T(b) + w_l \cdot L(b) ] Where: S(b) represents the importance score of memory block b; A(b) represents the access frequency of memory block b; T(b) represents the last access time of memory block b; L(b) represents the lifespan of memory block b, which is the length of time from the creation of memory block b to the current time; w_a, w_t, w_l are the weights of access frequency, last access time, and lifespan, respectively, and w_a + w_t + w_l = 1; The memory block reclamation based on the dynamically adjusted memory reclamation strategy includes: Dynamically calculate the memory reclamation threshold; Memory block reclamation begins when the current memory usage exceeds the latest calculated memory reclamation threshold. The memory reclamation threshold is dynamically calculated using the following formula: T_{new} = T_{base} + k \cdot (U - U_{target}) Where: T_{new} represents the latest reclamation threshold; T_{base} represents the pre-set base reclamation threshold; U represents the current memory utilization rate; U_{target} represents the pre-set target memory utilization rate; k is a pre-set adjustment coefficient, which reflects the responsiveness to memory pressure; The predicted future memory requirements include: Predicting future memory requirements using a linear regression model: D(t) = \beta_0 + \beta_1 \cdot D(t-1) + \beta_2 \cdot A(t-1) + \epsilon in: D(t) represents the memory requirement at time (t). D(t-1) represents the memory requirement at the previous time step; A(t-1) represents the amount of memory accessed in the previous time step; \beta_0, \beta_1, \beta_2 are all parameters of the linear regression model; \epsilon is the error term that is set.

Images