Power distribution network service hot migration method and device, electronic equipment, storage medium and computer program product

Abstract

The present disclosure relates to a power distribution network service hot migration method and device, electronic equipment, storage medium and computer program product. The method comprises: obtaining resource information of each edge node in a plurality of edge nodes contained in a power distribution network and load information of services on each edge node; inputting the resource information of each edge node and the load information of the services on each edge node into a residual network contained in a trained service migration prediction model to obtain a feature representation; inputting the feature representation into a migration strategy decision module contained in the service migration prediction model to obtain a service migration instruction, wherein the service migration instruction is used to indicate a to-be-migrated service and a migration path and a migration destination edge node corresponding to the to-be-migrated service; and issuing the service migration instruction to an edge node containing the to-be-migrated service in the plurality of edge nodes, so that the edge node containing the to-be-migrated service performs hot migration on the to-be-migrated service based on the service migration instruction.

Description

Technical Field

[0001] This disclosure relates to the field of business migration technology, and more specifically, to methods, apparatus, electronic devices, storage media, and computer program products for the thermal migration of distribution network services. Background Technology

[0002] With the continuous development of technology, the complexity and data processing requirements of business execution systems are constantly increasing. These systems include numerous edge computing nodes responsible for real-time processing and analysis of massive amounts of data from various sensors and devices to support critical business operations such as voltage regulation, load balancing, fault diagnosis, and prevention. In this context, hot migration of services across edge nodes has become an important means of improving the resilience and reliability of business execution systems.

[0003] Hot migration refers to moving a running virtual machine or application from one physical node to another without interrupting service. Through hot migration, services can achieve load balancing across different edge nodes, thereby improving overall performance and efficiency. Hot migration across edge nodes requires comprehensive consideration of factors such as network communication, computing resources, and data storage. This necessitates that the service execution system possess robust automated management and control mechanisms to ensure that normal operation and service quality are not affected during hot migration.

[0004] Currently, hot migration of services in business execution systems mainly relies on manually formulated migration plans and typically employs relatively simple scheduling algorithms, such as First-In-First-Out (FIFO) and fixed-priority scheduling. These algorithms can only meet basic migration requirements, lacking intelligence and flexibility, and often cannot adapt to complex and ever-changing network environments. Furthermore, the efficiency and accuracy of manually executing hot migration methods are limited, which may adversely affect business continuity and system security. Summary of the Invention

[0005] This disclosure provides a method, apparatus, electronic device, storage medium, and computer program product for hot migration of distribution network services, in order to at least solve the problem that the execution efficiency and accuracy of manual hot migration of services are limited, which may adversely affect business continuity and system security.

[0006] According to a first aspect of the present disclosure, a method for hot migration of distribution network services is provided, comprising: acquiring resource information of each edge node among a plurality of edge nodes included in the distribution network and load information of services on each edge node, wherein the resource information of each edge node is used to indicate information on resources that the services on the edge node can use, and the load information of the services on each edge node is used to indicate information on resources consumed by the services on the edge node; inputting the resource information of each edge node and the load information of the services on each edge node into a residual network included in a trained service migration prediction model to obtain a feature representation, wherein the feature representation is a representation vector related to service hot migration; inputting the feature representation into a migration strategy decision module included in the service migration prediction model to obtain a service migration indication, wherein the service migration indication is used to indicate the service to be migrated and the migration path and migration destination edge node corresponding to the service to be migrated; and distributing the service migration indication to the edge nodes among the plurality of edge nodes that contain the service to be migrated, so that the edge nodes containing the service to be migrated perform hot migration of the service to be migrated based on the service migration indication.

[0007] According to an exemplary embodiment of this disclosure, the migration strategy decision module includes a policy network and a value network; the step of inputting the feature representation into the migration strategy decision module included in the service migration prediction model to obtain a service migration indication includes: inputting the feature representation into the policy network to obtain a predicted migration strategy, wherein the predicted migration strategy includes at least one candidate service to be migrated and a migration path and a migration destination edge node corresponding to each of the at least one candidate service to be migrated; inputting the predicted migration strategy into the value network to obtain an estimated hot migration success rate of the at least one candidate service to be migrated; the step of obtaining a service migration indication further includes: selecting a candidate service to be migrated from the at least one candidate service to be migrated whose hot migration success rate is greater than or equal to a preset probability threshold as the service to be migrated.

[0008] According to an exemplary embodiment of this disclosure, before inputting the resource information of each edge node and the load information of the services on each edge node into the residual network included in the trained service migration prediction model to obtain a feature representation, the method further includes: obtaining the dependency relationship between services on each of the plurality of edge nodes; generating a directed acyclic graph based on the dependency relationship, wherein the nodes of the directed acyclic graph represent the services, and the edges of the directed acyclic graph represent the dependency relationship; the step of inputting the resource information of each edge node and the load information of the services on each edge node into the residual network included in the trained service migration prediction model to obtain a feature representation includes: inputting the resource information of each edge node, the load information of the services on each edge node, and the directed acyclic graph into the residual network included in the trained service migration prediction model to obtain the feature representation.

[0009] According to an exemplary embodiment of this disclosure, the service migration prediction model is trained by the following method: obtaining training samples, wherein the training samples include resource information of each simulated edge node among multiple simulated edge nodes included in the simulated distribution network and load information of services on each simulated edge node, the training samples are labeled, and the labels are used to indicate the hot migration status information of services on the simulated edge nodes when the simulated distribution network is in the state indicated by the corresponding sample; inputting the training samples into the residual network to obtain feature training representation; inputting the feature training representation into the migration strategy decision module to obtain service migration prediction results; calculating loss based on the service migration prediction results and the labels corresponding to the training samples; and adjusting the parameters of the service migration prediction model based on the loss for training.

[0010] According to an exemplary embodiment of this disclosure, the following operations are further performed during the training of the business migration prediction model: obtaining important training samples, wherein the important training samples are training samples used in the previous training process of the current training, and the difference between the business migration prediction result output by the business migration prediction model based on the important training samples and the label corresponding to the important training samples is greater than or equal to a preset difference threshold; and re-inputting the important training samples into the business migration prediction model for training.

[0011] According to an exemplary embodiment of this disclosure, the resource information of the edge node includes at least one of the following: the CPU resources of the edge node, the memory resources of the edge node, the hard disk capacity of the edge node, the hard disk input / output (IO) resources of the edge node, and the network bandwidth of the edge node; the load information of the services on the edge node includes at least one of the following: the CPU resources consumed by the service, the memory resources consumed by the service, the hard disk resources consumed by the service, the hard disk IO resources consumed by the service, and the network bandwidth consumed by the service.

[0012] According to a second aspect of the present disclosure, a hot migration device for distribution network services is provided, comprising: an information acquisition module configured to acquire resource information of each edge node among a plurality of edge nodes included in the distribution network and load information of services on each edge node, wherein the resource information of each edge node is used to indicate information about the resources that the services on the edge node can use, and the load information of the services on each edge node is used to indicate information about the resources consumed by the services on the edge node; and a feature representation acquisition module configured to input the resource information of each edge node and the load information of the services on each edge node into a trained service migration prediction model. The residual network includes a feature representation, wherein the feature representation is a representation vector related to service hot migration; a migration indication acquisition module is configured to input the feature representation into the migration strategy decision module included in the service migration prediction model to obtain a service migration indication, wherein the service migration indication is used to indicate the service to be migrated and the migration path and migration destination edge node corresponding to the service to be migrated; a migration indication distribution module is configured to distribute the service migration indication to the edge nodes containing the service to be migrated among the plurality of edge nodes, so that the edge nodes containing the service to be migrated can perform hot migration of the service to be migrated based on the service migration indication.

[0013] According to an exemplary embodiment of this disclosure, the migration strategy decision module includes a policy network and a value network; the migration indication acquisition module is configured to: input the feature representation into the policy network to obtain a predicted migration strategy, wherein the predicted migration strategy includes at least one candidate service to be migrated and a migration path and a migration destination edge node corresponding to each of the at least one candidate service to be migrated; input the predicted migration strategy into the value network to obtain an estimated hot migration success rate of the at least one candidate service to be migrated; and select candidate services to be migrated from the at least one candidate service to be migrated whose hot migration success rate is greater than or equal to a preset probability threshold as the service to be migrated.

[0014] According to an exemplary embodiment of this disclosure, the service hot migration apparatus further includes: a dependency acquisition module configured to: acquire the dependency relationships between services on each of the plurality of edge nodes; a directed acyclic graph (DAG) generation module configured to: generate a DAG based on the dependency relationships, wherein the nodes of the DAG represent the services, and the edges of the DAG represent the dependency relationships; and a feature representation acquisition module configured to: input the resource information of each edge node, the load information of the services on each edge node, and the DAG into the residual network contained in the trained service migration prediction model to obtain the feature representation.

[0015] According to an exemplary embodiment of this disclosure, the service migration prediction model is trained by the following method: obtaining training samples, wherein the training samples include resource information of each simulated edge node among multiple simulated edge nodes included in the simulated distribution network and load information of services on each simulated edge node, the training samples are labeled, and the labels are used to indicate the hot migration status information of services on the simulated edge nodes when the simulated distribution network is in the state indicated by the corresponding sample; inputting the training samples into the residual network to obtain feature training representation; inputting the feature training representation into the migration strategy decision module to obtain service migration prediction results; calculating loss based on the service migration prediction results and the labels corresponding to the training samples; and adjusting the parameters of the service migration prediction model based on the loss for training.

[0016] According to an exemplary embodiment of this disclosure, the following operations are further performed during the training of the business migration prediction model: obtaining important training samples, wherein the important training samples are training samples used in the previous training process of the current training, and the difference between the business migration prediction result output by the business migration prediction model based on the important training samples and the label corresponding to the important training samples is greater than or equal to a preset difference threshold; and re-inputting the important training samples into the business migration prediction model for training.

[0017] According to an exemplary embodiment of this disclosure, the resource information of the edge node includes at least one of the following: the CPU resources of the edge node, the memory resources of the edge node, the hard disk capacity of the edge node, the hard disk input / output (IO) resources of the edge node, and the network bandwidth of the edge node; the load information of the services on the edge node includes at least one of the following: the CPU resources consumed by the service, the memory resources consumed by the service, the hard disk resources consumed by the service, the hard disk IO resources consumed by the service, and the network bandwidth consumed by the service.

[0018] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement a hot migration method for distribution network services according to the present disclosure.

[0019] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform a hot migration method for distribution network services according to the present disclosure.

[0020] According to a fifth aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a hot migration method for distribution network services according to the present disclosure.

[0021] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0022] This disclosure proposes an intelligent scheduling strategy for online hot migration of services based on deep reinforcement learning. Deep neural networks can be reinforced through simulation of decision-making scenarios during hot migration, and then the trained deep neural network can be used to guide hot migration in a real power distribution network environment. This enables effective scheduling and allocation of services, improving the intelligence and flexibility of hot migration. Compared to manual hot migration, using deep neural networks to guide hot migration avoids introducing human error, thereby improving the efficiency, success rate, and security of hot migration.

[0023] Furthermore, residual networks possess powerful feature extraction capabilities. By incorporating residual networks into the service migration prediction model, it is ensured that features related to service hot migration can be accurately extracted from the distribution network's operational data. This enhances the model's learning ability and prevents gradient vanishing or exploding phenomena, thereby improving the model's stability during deep training. The trained model can then adapt to various complex and dynamic edge computing environments, effectively addressing diverse service migration scenarios.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0026] Figure 1This is a flowchart illustrating a method for thermal migration of distribution network services according to exemplary embodiments of the present disclosure;

[0027] Figure 2 This is a block diagram illustrating a thermal migration apparatus for distribution network services according to exemplary embodiments of the present disclosure;

[0028] Figure 3 This is a block diagram illustrating an electronic device according to exemplary embodiments of the present disclosure. Detailed Implementation

[0029] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0030] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following examples do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0031] It should be noted that the phrase "at least one of several items" in this disclosure refers to three parallel cases: "any one of the several items", "a combination of any number of the several items", and "all of the several items". For example, "including at least one of A and B" includes the following three parallel cases: (1) including A; (2) including B; (3) including A and B. Another example is "performing at least one of step one and step two", which means the following three parallel cases: (1) performing step one; (2) performing step two; (3) performing both step one and step two.

[0032] Figure 1 This is a flowchart illustrating a method for thermal migration of distribution network services according to exemplary embodiments of the present disclosure.

[0033] Reference Figure 1In step 101, resource information of each edge node and load information of services on each edge node can be obtained from the multiple edge nodes included in the distribution network. The resource information of each edge node can be used to indicate the resources available to the services on that edge node; the load information of the services on each edge node can be used to indicate the resources consumed by the services on that edge node. It should be noted that the hot migration method for services provided in this disclosure can be applied not only to the distribution network field but also to other service execution systems. The aforementioned services may include, but are not limited to, virtual machines, applications, containers, or tasks.

[0034] According to an exemplary embodiment of this disclosure, the resource information of an edge node may include at least one of the following: the central processing unit (CPU) resources of the edge node, the memory resources of the edge node, the hard disk capacity of the edge node, the hard disk input / output (IO) resources of the edge node, and the network bandwidth of the edge node.

[0035] The load information of services on edge nodes may include at least one of the following: CPU resources consumed by the service, memory resources consumed by the service, hard disk resources consumed by the service, hard disk I / O resources consumed by the service, and network bandwidth consumed by the service.

[0036] It should be noted that the state of a distribution network can be characterized by the resource information of each edge node within the network and the load information of services on each edge node. Furthermore, since services need to be hot-migrated between edge nodes, and the network bandwidth determines the migration speed, network bandwidth is an important factor among the many factors considered in determining the state of the distribution network. Additionally, "Actions" and "Rewards" can be defined. "Actions" instruct the hot-migrating of one or more services from their current edge node to their destination edge node; "Rewards" represent positive feedback from the hot migration, and may include, but are not limited to, hot migration time and success rate.

[0037] In this disclosure, a simulation environment can be built to simulate a real power distribution network. For example, the simulated power distribution network can contain multiple simulated edge nodes, each of which can run various simulated services. Simulated service hot migration data in the simulated power distribution network can be captured as training samples for training a service migration prediction model. Alternatively, in this disclosure, service hot migration data can also be captured from the operational data of a real power distribution network without affecting the real production environment.

[0038] In addition, after obtaining the training data, preprocessing can be performed on the raw training data to improve its quality, thereby ensuring better model training results. For example, the raw training data can be cleaned to handle missing and outlier values; it can also be normalized and standardized, and so on.

[0039] According to an exemplary embodiment of this disclosure, the business migration prediction model can be trained using the following methods:

[0040] Training samples are obtained, which may contain resource information for each simulated edge node among multiple simulated edge nodes in the simulated distribution network, and load information for services on each simulated edge node. The resource information of each simulated edge node can be used to indicate the resources available to the services on that simulated edge node; the load information of the services on each simulated edge node can be used to indicate the resources consumed by the services on that simulated edge node. Furthermore, the training samples may be labeled, which can be used to indicate the hot migration status of services on the simulated edge nodes when the simulated distribution network is in the state indicated by the corresponding sample.

[0041] Then, the training samples can be input into a residual network (ResNet) to obtain a feature training representation, which is a representation vector related to the simulated hot migration of services. Next, the feature training representation can be input into a migration strategy decision module to obtain a service migration prediction result. For example, the service migration prediction result may include at least one candidate simulated service to be migrated, and the migration path and migration destination edge node corresponding to each candidate simulated service.

[0042] Then, the loss can be calculated based on the business migration prediction results and the labels corresponding to the training samples. Next, the parameters of the business migration prediction model can be adjusted based on the calculated loss for training.

[0043] For example, the business migration prediction model provided in this disclosure may include, but is not limited to, three parts: a residual network, an A3C algorithm, and a priority experience replay algorithm. The working principle and functions of each of these three parts will be described below.

[0044] The A3C algorithm primarily consists of two parts: a policy network (Actor) and a value network (Critic). The policy network is mainly responsible for generating the hot migration strategy, determining which services need to be hot migrated and the corresponding migration path and destination edge node for each service. The value network is mainly responsible for estimating the value function, which is used to evaluate the feasibility of the hot migration strategy generated by the policy network. For example, the output of the value network may include, but is not limited to: the migration success rate for each service to be migrated, the estimated migration time, and the concurrency of all services to be migrated. Furthermore, the policy network and the value network can be trained simultaneously.

[0045] It's important to note that the Asynchronous Advantage Actor-Critic (A3C) algorithm possesses multi-threaded parallel update capabilities. Therefore, by incorporating the A3C algorithm into the business migration prediction model, a multi-threaded approach can be used to train the policy network, enabling rapid generation and adjustment of hot migration policies. This accelerates model convergence and improves training efficiency. Furthermore, the trained model can quickly adjust the hot migration order and resource allocation of services within the Directed Acyclic Graph (DAG), which consists of individual services and their dependencies, thereby maximizing the concurrency and success rate of hot migrations.

[0046] According to an exemplary embodiment of this disclosure, the following operations may also be performed during the training of the business migration prediction model:

[0047] Obtain important training samples, which can be training samples used in previous training processes during the current training iteration. The difference between the business migration prediction result output by the business migration prediction model based on these important training samples and the label corresponding to these important training samples can be greater than or equal to a preset difference threshold. It is worth noting that samples with larger errors usually contain more policy improvement information; therefore, samples with larger errors can be prioritized as important training samples. These important training samples can then be re-input into the business migration prediction model for training. In this way, by replaying historical important samples, catastrophic forgetting can be mitigated, that is, it can help the model retain its memory of old tasks, thereby improving model training efficiency.

[0048] It should be noted that the priority experience replay algorithm is mainly used for experience reuse in reinforcement learning, that is, it accelerates the convergence speed of the algorithm by prioritizing the storage of important experiences. Therefore, in this disclosure, the priority experience replay algorithm can be used to improve the training efficiency of the model. For example, the states, actions, rewards, and subsequent states of historical hot transfers with high reference value during model training can be recorded to establish a priority experience replay library. That is, information such as states, actions, and rewards involved in important hot transfers during historical hot transfers can be recorded in the experience pool. In this way, these important experiences can be used to predict and adjust for possible situations during subsequent training. For example, important experiences can be weighted and sampled by priority (e.g., network bandwidth, resource utilization, etc.) to improve the efficiency of policy optimization.

[0049] In this way, by adopting a priority experience replay mechanism, we can make full use of historical hot transfer experience, that is, we can relearn important experience, thereby accelerating the policy convergence speed and improving the model learning efficiency.

[0050] For ResNet, since it has a strong feature extraction capability, in this disclosure, the ResNet can be used to extract feature training representations related to business hot migration from the operation data of the simulated distribution network. The feature training representations can be high-dimensional feature representations for subsequent policy training, thereby achieving cross-layer information retention.

[0051] In this disclosure, by including a residual network in the service migration prediction model, the feature extraction capability of the residual network can be fully utilized to extract features related to service hot migration from the distribution network operation data as much as possible. This enhances the learning ability of the service migration prediction model, thereby preventing gradient vanishing or gradient exploding phenomena, ensuring the stability and effectiveness of the model during deep training, and enabling the trained model to adapt to various complex and changing edge computing environments.

[0052] It should be noted that the convergence speed of the algorithm can be judged by monitoring the cumulative reward. For example, after multiple experiments and training sessions, the cumulative reward curve can be used to observe that the algorithm gradually converges and stabilizes at a relatively high cumulative reward value. At this point, the resource scheduling efficiency, task concurrency, and transfer success rate of the transfer strategy formulated by the model are all high, while the transfer time is relatively short, indicating that the model has reached a good training level.

[0053] Thus, in this disclosure, by comprehensively utilizing priority experience replay, residual networks, and the A3C algorithm, an intelligent scheduler for service hot migration can be formed. This intelligent scheduler is primarily responsible for generating hot migration strategies based on the current state of the distribution network. For example, the intelligent scheduler can calculate the optimal migration path and destination edge node based on information such as network bandwidth, CPU resources, memory resources, and hard disk resources, and can make real-time adjustments throughout the entire service hot migration scheduling process to maximize the concurrency, efficiency, and success rate of service hot migration.

[0054] In step 102, the resource information of each edge node and the load information of the services on each edge node can be input into the residual network contained in the trained service migration prediction model to obtain a feature representation, wherein the feature representation can be a representation vector related to service hot migration.

[0055] According to an exemplary embodiment of this disclosure, the dependencies between services on each of a plurality of edge nodes can be obtained. Then, a directed acyclic graph (DAG) can be generated based on these dependencies, where nodes in the DAG represent services and edges represent dependencies between services. Next, the resource information of each edge node, the load information of the services on each edge node, and the DAG can be input into the residual network of a trained service migration prediction model to obtain feature representations.

[0056] In this disclosure, services on edge nodes within a distribution network can be abstracted and analyzed to generate a Directed Acyclic Graph (DAG). Specifically, when dependencies exist between multiple services on an edge node, a DAG can be used to represent the order of these services. For example, by analyzing the service logic, the priority and dependencies of each service can be clearly defined, ultimately abstracted into a weighted directed acyclic graph. Furthermore, each node in this DAG can represent a service or task, the edges can represent dependencies between services, and the edge weights can represent the amount of data migrated or the migration time. Then, the resource status of each edge node in the distribution network, the resource usage of services on each edge node, the DAG dependencies of the services, and the execution status of the services on each DAG node can be input into a residual network to obtain feature representations related to service hot migration.

[0057] In this way, by introducing a weighted directed acyclic graph (DAG), the dependencies between different services can be clearly defined, which facilitates the generation of subsequent intelligent scheduling strategies.

[0058] In step 103, the aforementioned feature representation can be input into the migration strategy decision module included in the business migration prediction model to obtain a business migration indication. The business migration indication can be used to indicate the business to be migrated, as well as the migration path and migration destination edge node corresponding to the business to be migrated.

[0059] According to an exemplary embodiment of this disclosure, the migration strategy decision module described above may include a policy network and a value network.

[0060] The aforementioned feature representation can be input into a policy network to obtain a predicted migration policy. This predicted migration policy may include at least one candidate service to be migrated, and for each candidate service, a corresponding migration path and a destination edge node. Then, the predicted migration policy can be input into a value network to obtain an estimated hot migration success rate for at least one candidate service. Next, candidates with a hot migration success rate greater than or equal to a preset probability threshold can be selected from the at least one candidate service as the final migration service.

[0061] In this way, the policy network can predict hot migration strategies, and the value network can evaluate the feasibility and success rate of these strategies. This allows for hot migration to be performed on services with a higher probability of success, ensuring the overall success rate of hot migration. Simultaneously, it avoids the indiscriminate hot migration of all candidate services predicted by the policy network, preventing the blind abuse of hot migration resources and avoiding unnecessary waste.

[0062] It should be noted that the aforementioned policy network can output multiple migration policies in a single instance. Each migration policy can include at least one candidate service to be migrated, as well as the migration path and destination edge node for each candidate service. Furthermore, in addition to outputting the hot migration success rate for each candidate service, the aforementioned value network can also output information such as the hot migration time for each candidate service and the overall hot migration concurrency for the corresponding migration policy. The overall hot migration concurrency can refer to the total number of hot migration services executed in a single hot migration process.

[0063] At this point, when determining the final services to be migrated, in addition to considering the hot migration success rate, the hot migration time for each candidate service and the overall hot migration concurrency of the corresponding migration strategy can also be considered. For example, the higher the hot migration success rate, the shorter the hot migration time, and the greater the overall hot migration concurrency, the higher the probability that the corresponding migration strategy will be selected.

[0064] In this way, when determining the final hot migration strategy, information from multiple dimensions can be comprehensively considered, making the factors considered more comprehensive and covering a wider range, thereby further ensuring the rationality and effectiveness of the final hot migration strategy.

[0065] In step 104, the aforementioned service migration instruction can be sent to multiple edge nodes containing the services to be migrated, so that the edge nodes containing the services to be migrated can perform hot migration of the services to be migrated based on the service migration instruction.

[0066] For example, in this disclosure, the Checkpoint / Restore In Userspace (CRIU) technology can be used to perform hot migration for each service. The CRIU technology can mainly consist of two parts: a pre-check part and an online hot migration part.

[0067] The pre-check section primarily refers to checking whether the resources and network bandwidth of the destination edge node are sufficient before performing a business hot migration. If not, a new destination edge node can be selected, or the hot migration can be postponed. This further ensures the success rate of the hot migration, effectively avoiding migration failures caused by the destination edge node's inability to support the migration's overhead.

[0068] For the online hot migration part, firstly, the CRIU tool can be used to perform a checkpoint operation on the service (container) to save the container's current state to a file. Then, this file can be transferred to the destination edge node over the network. Next, a restore operation can be performed on the destination edge node to restore the container to operation. This completes a full service hot migration process.

[0069] In this disclosure, after each business hot migration is completed, the data of that hot migration can be recorded to facilitate subsequent analysis and optimization of the business migration prediction model.

[0070] It should be noted that if a single point of failure or insufficient computing power occurs at an edge node within the distribution network, a hot migration of services may be necessary. In this case, if the feature extraction capability of the service migration prediction model is poor, the migration strategy predicted by the model will deviate from the actual operating conditions of the distribution network, resulting in a low success rate for the hot migration of services.

[0071] Furthermore, residual networks possess powerful feature extraction capabilities. Therefore, to address the low success rate of service hot migration due to the poor feature extraction capabilities of the model, this disclosure includes a residual network in the service migration prediction model. Thus, during actual service hot migration, only the resource status of the edge nodes in the distribution network (e.g., CPU resources, memory resources, hard disk capacity, hard disk I / O resources, and network bandwidth) and the load status of services on the edge nodes (e.g., CPU resources consumed, memory resources consumed, hard disk resources consumed, hard disk I / O resources consumed, and network bandwidth consumed) need to be input into the residual network included in the service migration prediction model. This ensures that features related to service hot migration are accurately extracted from the distribution network's operational data. This facilitates subsequent effective prediction of service hot migration strategies based on the accurately extracted features, ensuring that the predicted migration strategy conforms to the actual operating conditions of the distribution network. Consequently, this guarantees a high degree of service hot migration concurrency, low hot migration time, and a high hot migration success rate.

[0072] Figure 2 This is a block diagram illustrating a thermal migration apparatus 200 for distribution network services according to an exemplary embodiment of the present disclosure.

[0073] Reference Figure 2 The thermal migration device 200 for the power distribution network service may include an information acquisition module 201, a feature representation acquisition module 202, a migration instruction acquisition module 203, and a migration instruction issuing module 204.

[0074] The information acquisition module 201 can acquire resource information of each edge node among multiple edge nodes in the distribution network, as well as load information of services on each edge node. The resource information of each edge node can be used to indicate the resources that the services on that edge node can use; the load information of the services on each edge node can be used to indicate the resources consumed by the services on that edge node.

[0075] According to an exemplary embodiment of this disclosure, the resource information of an edge node may include at least one of the following:

[0076] CPU resources of edge nodes, memory resources of edge nodes, hard disk capacity of edge nodes, hard disk I / O resources of edge nodes, and network bandwidth of edge nodes.

[0077] The load information of services on edge nodes may include at least one of the following:

[0078] CPU resources consumed by the service, memory resources consumed by the service, hard disk resources consumed by the service, hard disk I / O resources consumed by the service, and network bandwidth consumed by the service.

[0079] According to an exemplary embodiment of this disclosure, the business migration prediction model can be trained using the following methods:

[0080] Training samples are obtained, which may contain resource information for each simulated edge node among multiple simulated edge nodes in the simulated distribution network, and load information for services on each simulated edge node. The resource information of each simulated edge node can be used to indicate the resources available to the services on that simulated edge node; the load information of the services on each simulated edge node can be used to indicate the resources consumed by the services on that simulated edge node. Furthermore, the training samples may be labeled, which can be used to indicate the hot migration status of services on the simulated edge nodes when the simulated distribution network is in the state indicated by the corresponding sample.

[0081] Then, the training samples can be input into the residual network to obtain a feature training representation, which is a representation vector related to the simulated hot migration of services. Next, the feature training representation can be input into the migration strategy decision module to obtain the service migration prediction result. For example, the service migration prediction result may include at least one candidate simulated service to be migrated, and the migration path and migration destination edge node corresponding to each candidate simulated service.

[0082] Then, the loss can be calculated based on the business migration prediction results and the labels corresponding to the training samples. Next, the parameters of the business migration prediction model can be adjusted based on the calculated loss for training.

[0083] According to an exemplary embodiment of this disclosure, the following operations may also be performed during the training of the business migration prediction model:

[0084] Obtain important training samples, which can be training samples used in previous training processes during the current training iteration. The difference between the business migration prediction result output by the business migration prediction model based on these important training samples and the label corresponding to these important training samples can be greater than or equal to a preset difference threshold. It is worth noting that samples with larger errors usually contain more policy improvement information; therefore, samples with larger errors can be prioritized as important training samples. Then, these important training samples can be re-input into the business migration prediction model for training.

[0085] The feature representation acquisition module 202 can input the resource information of each edge node and the load information of the services on each edge node into the residual network contained in the trained service migration prediction model to obtain the feature representation, wherein the feature representation can be a representation vector related to service hot migration.

[0086] According to an exemplary embodiment of this disclosure, the above-described heat migration device 200 for power distribution network services may further include a dependency acquisition module and a directed acyclic graph generation module.

[0087] The dependency acquisition module can acquire the dependencies between services on each of multiple edge nodes. Then, the directed acyclic graph (DAG) generation module can generate a DAG based on these dependencies, where nodes represent services and edges represent dependencies between services. Next, the feature representation acquisition module 202 can obtain feature representations by inputting the resource information of each edge node, the load information of the services on each edge node, and the DAG into the residual network of the trained service migration prediction model.

[0088] The migration indication acquisition module 203 can input the aforementioned feature representation into the migration strategy decision module included in the business migration prediction model to obtain a business migration indication. The business migration indication can be used to indicate the business to be migrated, as well as the migration path and migration destination edge node corresponding to the business to be migrated.

[0089] According to an exemplary embodiment of this disclosure, the migration strategy decision module described above may include a policy network and a value network.

[0090] The migration indication acquisition module 203 can input the aforementioned feature representation into the policy network to obtain a predicted migration policy. This predicted migration policy may include at least one candidate service to be migrated, and the migration path and destination edge node corresponding to each candidate service. Then, the migration indication acquisition module 203 can input the predicted migration policy into the value network to obtain an estimated hot migration success rate for at least one candidate service to be migrated. Next, the migration indication acquisition module 203 can select candidate services with a hot migration success rate greater than or equal to a preset probability threshold from the at least one candidate service to be migrated as the final service to be migrated.

[0091] The migration instruction distribution module 204 can distribute the aforementioned service migration instruction to multiple edge nodes containing the service to be migrated, so that the edge nodes containing the service to be migrated can perform hot migration of the service to be migrated based on the service migration instruction.

[0092] Figure 3 This is a block diagram illustrating an electronic device 300 according to an exemplary embodiment of the present disclosure.

[0093] Reference Figure 3The electronic device 300 includes at least one memory 301 and at least one processor 302. The at least one memory 301 stores instructions that, when executed by the at least one processor 302, perform a hot migration method for distribution network services according to an exemplary embodiment of the present disclosure.

[0094] As an example, electronic device 300 may be a PC, tablet, personal digital assistant, smartphone, or other device capable of executing the aforementioned instructions. Here, electronic device 300 is not necessarily a single electronic device, but may be a collection of any devices or circuits capable of executing the aforementioned instructions (or instruction sets) individually or in combination. Electronic device 300 may also be part of an integrated control system or system manager, or may be configured to interface with a portable electronic device locally or remotely (e.g., via wireless transmission).

[0095] In electronic device 300, processor 302 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller, or a microprocessor. By way of example and not limitation, processor may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, etc.

[0096] The processor 302 can execute instructions or code stored in the memory 301, which can also store data. Instructions and data can also be sent and received over a network via a network interface device, which can employ any known transmission protocol.

[0097] The memory 301 may be integrated with the processor 302, for example, by placing RAM or flash memory within an integrated circuit microprocessor. Alternatively, the memory 301 may include a separate device, such as an external disk drive, a storage array, or other storage device usable by any database system. The memory 301 and the processor 302 may be operatively coupled, or may communicate with each other, for example, via I / O ports, network connections, etc., enabling the processor 302 to read files stored in the memory.

[0098] In addition, the electronic device 300 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). All components of the electronic device 300 can be interconnected via a bus and / or network.

[0099] According to exemplary embodiments of this disclosure, a computer-readable storage medium may also be provided, which, when executed by a processor of an electronic device, enables the electronic device to perform the aforementioned hot-migrating method for power distribution network services. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R, BD-R The computer program in the aforementioned computer-readable storage medium can run in an environment deployed in computer devices such as clients, hosts, agents, and servers. Furthermore, in one example, the computer program and any associated data, data files, and data structures are distributed across a networked computer system, allowing the computer program and any associated data, data files, and data structures to be stored, accessed, and executed in a distributed manner through one or more processors or computers.

[0100] According to exemplary embodiments of the present disclosure, a computer program product may also be provided, including a computer program that, when executed by a processor, implements a hot migration method for distribution network services according to the present disclosure.

[0101] According to the hot migration method, apparatus, electronic equipment, storage medium, and computer program product for distribution network services disclosed herein, deep neural networks can be reinforced through simulation of decision-making scenarios during hot migration. The trained deep neural network can then guide hot migration in a real distribution network environment. This enables effective scheduling and allocation of services, improving the intelligence and flexibility of hot migration. Compared to manual hot migration, using a deep neural network to guide hot migration avoids introducing human error, thereby improving the efficiency, success rate, and security of hot migration.

[0102] Furthermore, residual networks possess powerful feature extraction capabilities. By incorporating residual networks into the service migration prediction model, it is ensured that features related to service hot migration can be accurately extracted from the distribution network's operational data. This enhances the model's learning ability and prevents gradient vanishing or exploding phenomena, thereby improving the model's stability during deep training. The trained model can then adapt to various complex and dynamic edge computing environments, effectively addressing diverse service migration scenarios.

[0103] According to exemplary embodiments of this disclosure, by including the A3C algorithm in the business migration prediction model, a multi-threaded approach can be used to train the policy network, enabling rapid generation and adjustment of hot migration policies. This accelerates model convergence and improves training efficiency. Furthermore, the trained model can quickly adjust the hot migration order and resource allocation of services in the directed acyclic graph (DAG) composed of individual services and their dependencies, thereby maximizing the concurrency and success rate of hot migration.

[0104] According to exemplary embodiments of this disclosure, replaying important historical samples can mitigate catastrophic forgetting, thus helping the model retain its memory of old tasks and improving model training efficiency. Specifically, by employing a priority experience replay mechanism, historical hot transfer experience can be fully utilized, enabling relearning of important experiences, thereby accelerating policy convergence and improving model learning efficiency.

[0105] According to an exemplary embodiment of this disclosure, by introducing a weighted directed acyclic graph, the dependencies between different services can be clearly defined, thereby facilitating the generation of subsequent intelligent scheduling strategies.

[0106] According to exemplary embodiments of this disclosure, a policy network can be used to predict hot migration strategies; a value network can be used to evaluate the feasibility and success rate of hot migration strategies. This allows for hot migration to be performed on services with a higher probability of success, ensuring the overall success rate of service hot migration. Simultaneously, it avoids the indiscriminate hot migration of all candidate services predicted by the policy network, preventing the blind abuse of hot migration resources and thus avoiding unnecessary waste.

[0107] According to exemplary embodiments of this disclosure, when determining the final thermal migration strategy, information from multiple dimensions can be comprehensively considered to make the factors considered more comprehensive and cover a wider range, thereby further ensuring the rationality and effectiveness of the final thermal migration strategy.

[0108] According to exemplary embodiments of this disclosure, before performing a service hot migration, it is possible to check whether the resources of the destination edge node are sufficient and whether the network bandwidth meets the requirements. If not, the destination edge node can be reselected or the hot migration can be postponed. This further ensures the success rate of the hot migration, effectively avoiding hot migration failures caused by the destination edge node's inability to support the hot migration consumption.

[0109] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0110] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for thermal migration of distribution network services, characterized in that, include: The resource information of each edge node and the load information of the services on each edge node are obtained from the multiple edge nodes included in the distribution network. The resource information of each edge node is used to indicate the resources that the services on the edge node can use, and the load information of the services on each edge node is used to indicate the resources consumed by the services on the edge node. The resource information of each edge node and the load information of the services on each edge node are input into the residual network of the trained service migration prediction model to obtain feature representations, wherein the feature representations are representation vectors related to service hot migration. The feature representation is input into the migration strategy decision module included in the business migration prediction model to obtain a business migration indication, wherein the business migration indication is used to indicate the business to be migrated and the migration path and migration destination edge node corresponding to the business to be migrated; The service migration instruction is sent to the edge nodes of the plurality of edge nodes that contain the service to be migrated, so that the edge nodes containing the service to be migrated can perform hot migration of the service to be migrated based on the service migration instruction; The following operations are also performed during the training of the business migration prediction model: Obtain important training samples, wherein the important training samples are the training samples used in the previous training process of the current training, and the difference between the business migration prediction result output by the business migration prediction model based on the important training samples and the label corresponding to the important training samples is greater than or equal to a preset difference threshold. The important training samples are re-input into the business migration prediction model for training. Before inputting the resource information of each edge node and the load information of the services on each edge node into the residual network of the trained service migration prediction model to obtain feature representations, the process further includes: Obtain the dependencies between services on each of the multiple edge nodes; Based on the dependencies, a directed acyclic graph is generated, wherein the nodes of the directed acyclic graph represent the services, and the edges of the directed acyclic graph represent the dependencies. The step of inputting the resource information of each edge node and the load information of the services on each edge node into the residual network of the trained service migration prediction model to obtain feature representations includes: The feature representation is obtained by combining the resource information of each edge node, the load information of the services on each edge node, and the residual network contained in the service migration prediction model trained by the directed acyclic graph input.

2. The thermal migration method as described in claim 1, characterized in that, The migration strategy decision-making module includes a strategy network and a value network; The step of inputting the feature representation into the migration strategy decision module included in the business migration prediction model to obtain a business migration indication includes: The feature representation is input into the policy network to obtain a predicted migration policy, wherein the predicted migration policy includes at least one candidate service to be migrated and a migration path and a migration destination edge node corresponding to each of the at least one candidate service to be migrated. The predicted migration strategy is input into the value network to obtain the estimated hot migration success rate of the at least one candidate service to be migrated; The process of obtaining the business migration instruction also includes: From the at least one candidate service to be migrated, select the candidate service whose hot migration success rate is greater than or equal to a preset probability threshold as the service to be migrated.

3. The thermal migration method as described in claim 1, characterized in that, The business migration prediction model was trained using the following method: Acquire training samples, wherein the training samples contain resource information of each simulated edge node among multiple simulated edge nodes included in the simulated distribution network and load information of services on each simulated edge node. The training samples are labeled, and the labels are used to indicate the hot migration status information of services on the simulated edge nodes when the simulated distribution network is in the state indicated by the corresponding sample. The training samples are input into the residual network to obtain the feature training representation; The trained feature representation is input into the migration strategy decision module to obtain the business migration prediction result; Based on the business migration prediction results and the labels corresponding to the training samples, the loss is calculated; The parameters of the business migration prediction model are adjusted based on the loss to train the model.

4. The thermal migration method as described in claim 1, characterized in that, The resource information of the edge node includes at least one of the following: The edge node's CPU resources, memory resources, hard disk capacity, hard disk I / O resources, and network bandwidth; The load information of services on the edge node includes at least one of the following: CPU resources consumed by the service, memory resources consumed by the service, hard disk resources consumed by the service, hard disk I / O resources consumed by the service, and network bandwidth consumed by the service.

5. A thermal migration device for power distribution network operations, characterized in that, include: The information acquisition module is configured to acquire resource information of each edge node among multiple edge nodes included in the distribution network and load information of services on each edge node, wherein the resource information of each edge node is used to indicate the resources that the services on the edge node can use, and the load information of the services on each edge node is used to indicate the resources consumed by the services on the edge node. The feature representation acquisition module is configured to input the resource information of each edge node and the load information of the services on each edge node into the residual network contained in the trained service migration prediction model to obtain a feature representation, wherein the feature representation is a representation vector related to service hot migration; The migration indication acquisition module is configured to input the feature representation into the migration strategy decision module included in the business migration prediction model to obtain a business migration indication, wherein the business migration indication is used to indicate the business to be migrated and the migration path and migration destination edge node corresponding to the business to be migrated; The migration instruction distribution module is configured to distribute the service migration instruction to the edge nodes containing the service to be migrated among the plurality of edge nodes, so that the edge nodes containing the service to be migrated can perform hot migration of the service to be migrated based on the service migration instruction. The following operations are also performed during the training of the business migration prediction model: Obtain important training samples, wherein the important training samples are the training samples used in the previous training process of the current training, and the difference between the business migration prediction result output by the business migration prediction model based on the important training samples and the label corresponding to the important training samples is greater than or equal to a preset difference threshold. The important training samples are re-input into the business migration prediction model for training. The hot migration device for the service also includes: The dependency acquisition module is configured to: acquire the dependency relationships between services on each of the multiple edge nodes; The directed acyclic graph generation module is configured to: generate a directed acyclic graph based on the dependency relationship, wherein the nodes of the directed acyclic graph represent the business, and the edges of the directed acyclic graph represent the dependency relationship; The feature representation acquisition module is configured to: obtain the feature representation by taking the resource information of each edge node, the load information of the services on each edge node, and the residual network contained in the service migration prediction model trained by the directed acyclic graph input.

6. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the hot migration method for distribution network services as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the hot migration method for distribution network services as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the hot migration method for distribution network services as described in any one of claims 1 to 4.

Images