CN117880145A - Multi-level 6G computing power service chain abnormality detection mechanism deployment method, device and equipment - Google Patents

Abstract

The invention relates to the technical field of resource management and discloses a method, a device and equipment for deploying an abnormality detection mechanism of a multi-level 6G computing power service chain. Therefore, by implementing the invention, different abnormal detection mechanism deployment schemes under different network states can be realized, and the problem of increasing the end-to-end delay of the C-SFC stream can be solved.

Description

Multi-level 6G computing power service chain abnormality detection mechanism deployment method, device and equipment

Technical Field

The invention relates to the technical field of resource management, in particular to a deployment method, device and equipment of a multi-level 6G computing power service chain abnormality detection mechanism.

Background

The 6G mobile communication technology will promote the human society to step into the intelligent age of everything interconnection and virtual symbiosis, wherein the augmented reality (XR) cloud service, the touch feedback and the holographic display are all possible to be mainstream applications. The service application value of the 6G network carried in the future is greatly improved from everything interconnection to intelligent interconnection, which provides great challenges for network service functions, and is mainly embodied in highly personalized service, flexibility, expandability and efficient resource utilization. Based on the idea of 'network as service', a power computing service chain supported by multi-dimensional power computing resources is focused, delay sensitive service in a 6G network scene is enabled, and the stable operation of the full life cycle of the power computing service function is ensured. Accordingly, various new network architectures are continuously proposed by related researchers, wherein the new network of SDNFV is a network architecture system where SDN and NFV are integrated, and this system makes network development, management and orchestration more efficient and flexible. SFC and NFV can help achieve highly personalized services, enabling operators to quickly configure and reconfigure network functions according to user and application requirements, thereby providing customized services. The network resource is effectively utilized by operators, and the resource waste is reduced.

The 6G business will show intelligent and immersive development trend under the support of key technologies such as flexible resource management. However, holographic applications are a potential mainstream application for future 6G, which includes a large number of computing tasks such as three-dimensional reconstruction, model decompression, codec, three-dimensional rendering, etc., which require the network to provide powerful computing and forwarding functions at the same time to achieve the goal of "network as a service" (Network as a Service, naaS). The new business has higher time delay requirement and calculation force requirement, and by introducing calculation force resources, the system needs to be considered from a brand new dimension.

At present, a programmable switch deployment anomaly detection model is utilized to provide C-SFC (Computing based Service Function Chain, force service chain) anomaly detection with strong real-time performance, low time delay and low communication overhead. However, different deployment mechanisms of the anomaly detection model may generate different increases in processing delay of the programmable switch, thereby increasing the end-to-end delay of the C-SFC stream.

Disclosure of Invention

In view of the above, the invention provides a deployment method, a device and equipment for a multi-level 6G computing power service chain abnormality detection mechanism, so as to solve the problem that different deployment mechanisms of an abnormality detection model can generate different increases to the processing delay of a programmable switch, thereby increasing the end-to-end delay of a C-SFC stream.

In a first aspect, the present invention provides a deployment method of a multi-level 6G computing power service chain anomaly detection mechanism, for a centralized controller; the method comprises the following steps:

acquiring computing resource information, link resource information, computing resource residual information and DPU residual resource information of each computing power equipment node in a multi-level 6G computing power service chain to be detected; constructing a target decision model by using the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information; and solving the target decision model by using a reinforcement learning algorithm based on a preset DQN framework to obtain an anomaly detection mechanism deployment scheme of the multi-level 6G computing service chain to be detected.

According to the deployment method of the multi-level 6G power computing service chain abnormality detection mechanism, on the basis of considering time delay, a target decision model is constructed through the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information of each power computing equipment node in the multi-level 6G power computing service chain to be detected, the computing resource and the network resource can be converted, and further, the optimal deployment scheme of the abnormality detection mechanism of the multi-level 6G power computing service chain to be detected can be obtained through solving the model through an algorithm. Therefore, by implementing the invention, different abnormal detection mechanism deployment schemes under different network states can be realized, and the problem of increasing the end-to-end delay of the C-SFC stream can be solved.

In an alternative embodiment, constructing the target decision model using the computing resource information, the link resource information, the computing resource remaining information, and the DPU remaining resource information based on the preset delay includes:

constructing a connected graph based on the computing resource information and the link resource information; and constructing a target decision model by using the connected graph, the computing resource remaining information and the DPU remaining resource information based on the preset time delay.

In an alternative embodiment, constructing the target decision model based on the preset time delay by using the connectivity graph, the computing resource remaining information and the DPU remaining resource information includes:

determining a preset decision planning requirement through a preset time slot processing method based on the connected graph and the preset time delay; constructing a reward function based on the computing resource remaining information and the DPU remaining resource information; and constructing a target decision model based on the preset decision planning requirements, the rewarding function and the connectivity graph.

In an alternative embodiment, constructing the target decision model based on the preset decision planning requirements, the reward function and the connectivity graph includes:

respectively constructing an environment state space, an action space and state transition based on the connected graph and a preset decision planning requirement; a target decision model is constructed based on the environmental state space, the action space, the state transition, and the reward function.

In an alternative embodiment, the method further comprises:

the method comprises the steps that a plurality of preset anomaly detection models are arranged in a multi-level 6G computing power service chain to be detected based on an anomaly detection mechanism deployment scheme;

and carrying out abnormality detection on the multi-level 6G power calculation service chain to be detected by utilizing a plurality of preset abnormality detection models to obtain an abnormality detection result.

According to the method and the device for detecting the abnormal condition, the obtained abnormal condition detection mechanism deployment scheme can correspondingly deploy different preset abnormal condition detection models and can complete corresponding abnormal condition detection.

In an alternative embodiment, a deployment scheme based on an anomaly detection mechanism deploys a plurality of preset anomaly detection models within a multi-level 6G computing power service chain to be detected, including:

acquiring a preset abnormality detection model; determining a plurality of target computing equipment nodes in a multi-level 6G computing service chain to be detected based on an anomaly detection mechanism deployment scheme; and deploying a preset abnormality detection model on each target computing equipment node.

In an optional implementation manner, performing anomaly detection on a multi-level 6G computing power service chain to be detected by using a plurality of preset anomaly detection models to obtain an anomaly detection result, including:

Performing abnormality detection on each target power equipment node by using a preset abnormality detection model to obtain an abnormality detection result of each target power equipment node; and determining an abnormality detection result of the multi-level 6G computing power service chain to be detected based on the abnormality detection result of each target computing power equipment node.

In a second aspect, the invention provides a deployment device of a multi-level 6G computing power service chain abnormality detection mechanism, which is used for a centralized controller; the device comprises:

the acquisition module is used for acquiring computing resource information, link resource information, computing resource residual information and DPU residual resource information of each computing equipment node in the multi-level 6G computing service chain to be detected; the construction module is used for constructing a target decision model by utilizing the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information; the solving module is used for solving the target decision model by utilizing a reinforcement learning algorithm based on the DQN framework to obtain an anomaly detection mechanism deployment scheme of the multi-level 6G computing power service chain to be detected.

In a third aspect, the present invention provides a computer device comprising: the system comprises a memory and a processor, wherein the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, so that the method for deploying the multi-level 6G computing power service chain abnormality detection mechanism according to the first aspect or any implementation mode corresponding to the first aspect is executed.

In a fourth aspect, the present invention provides a computer readable storage medium having stored thereon computer instructions for causing a computer to execute the method for deploying a multi-level 6G power service chain anomaly detection mechanism according to the first aspect or any one of its corresponding embodiments.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.

FIG. 1 is a network topology according to an embodiment of the present invention;

FIG. 2 is a node resource information diagram according to an embodiment of the invention;

FIG. 3 is a flow diagram of a multi-tier 6G computing power service chain anomaly detection mechanism deployment method in accordance with an embodiment of the present invention;

FIG. 4 is a flow diagram of another multi-tier 6G power chain anomaly detection mechanism deployment method in accordance with an embodiment of the present invention;

FIG. 5 is a schematic diagram of an execution flow of a preset DQN framework-based reinforcement learning algorithm according to an embodiment of the present invention;

FIG. 6 is a flow diagram of yet another multi-tier 6G power chain anomaly detection mechanism deployment method in accordance with an embodiment of the present invention;

FIG. 7 is a block diagram of a multi-tier 6G computing power service chain anomaly detection mechanism deployment device according to an embodiment of the present invention;

fig. 8 is a schematic diagram of a hardware structure of a computer device according to an embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The current abnormality detection model can be deployed in a controller, a switch and a DPU, the abnormality detection model is deployed in a central controller, the time required for abnormality detection is relatively large, the hardware acceleration time required for deployment in the DPU is minimum, and the time required for deployment in the switch is between the two. Therefore, for some DSCSFCs (Delay-Sensitive Computing based Service Function Chain, delay-sensitive C-SFCs), flexible adjustment of the deployment policy is required, and in order to ensure that the whole system can deploy more computing power service function chains, reasonable utilization of resources is required, so that reasonable deployment is required according to the deployment requirement of each service function chain in combination with the network state use policy.

The existing deployment method has the following defects:

(1) The state of the resources in the network is dynamically changed, the deployment scheme of the service function chain cannot be dynamically adjusted according to the state of the resources, the calculation time is not outstanding when the deployment scheme is determined, overall optimization of the occupation of the resources is lacked, and rapid analysis and deployment of the high real-time problem cannot be guaranteed.

(2) The time delay sensitive multi-source multicast resource problem in the existing NFV network is optimized, the limited deployment quantity of each functional module in the network is considered, so that a plurality of identical requests are uniformly processed within the range allowed by resource constraint through a multi-source scheme, the time delay problem of SFCs is guaranteed, the uniqueness of each SFC is reduced, and the method is not applicable to an application scene that one SFC is provided with a single fault detection model. The method does not consider the consumption of the computing resources of the hardware in the network, more considers the problem of the link state, and lacks the consideration of hardware acceleration.

(3) In order to maximize profits while guaranteeing a strict service life, it is necessary to efficiently decide whether to place a new VNF instance or reuse a deployed VNF instance for NS scheduling. The prior art considers not only the placement and scheduling of VNFs, but also traffic routing. However, the problem of acceleration using a hardware DPU is relatively not considered, and there is still a disadvantage in terms of processing speed. Only a single deployment condition exists for function deployment, and multi-dimensional expansion of a deployment environment is lacked.

In accordance with an embodiment of the present invention, a multi-tiered 6G power chain anomaly detection mechanism deployment method embodiment is provided, it being noted that the steps illustrated in the flow diagrams of the figures may be performed in a computer system, such as a set of computer-executable instructions, and that although a logical order is illustrated in the flow diagrams, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.

The embodiment provides a deployment method of a multi-level 6G computing power service chain abnormality detection mechanism, which can be used for a centralized controller.

Wherein, the centralized controllers are arranged in the corresponding network architecture.

Specifically, the method is divided into two main parts, wherein a data plane is below a virtual layer, and a control layer is above the virtual layer. The control layer is responsible for collecting network state information, running a deployment algorithm of the power service chain fault detection model and issuing a corresponding decision result. The data plane, i.e. various network devices, may perform specific network tasks such as forwarding, computing, etc. The control plane comprises a bottom layer network multidimensional resource sensing, a K8s orchestrator and an ONOS controller. The data plane includes network components with different computing power and network resources, and is responsible for deployment of network functions constituting a computing power service chain, forwarding of data packets, and the like. The computing power and network information of the bottom layer network component are monitored and acquired through an application program interface API and an open source tool (such as Zabbix and the like) at the data layer, and the computing power and network information specifically comprises a CPU, an available memory, an available storage, a link delay and a link bandwidth, and the information is uploaded to the control layer, so that some computing tasks can be executed, and a computing result is returned.

In one example, as shown in FIG. 1, the physical space is comprised of physical entity devices (e.g., controllers, switches, etc.) and various heterogeneous resources (CPU and GPU computing resources, DPU computing resources, storage, and networks) that exist in these physical devices. The centralized controller is used for centralized arrangement planning, the whole function is split into different functional components by different C-SFCs at the virtual level, chained arrangement is carried out, and then proper computing equipment is selected for deployment. The method is embodied in a physical layer, namely different virtualized function modules are deployed to different computing nodes under the control of a centralized orchestration controller.

Further, in the network, the centralized controller is generally operated as a server, and has high computational power, so that various complex calculations can be performed. The power computing network equipment is integrated equipment integrating calculation, storage and forwarding, has certain calculation capacity, is basically provided by a CPU and GPU combination, deploys an abnormality detection model in a power computing node, occupies partial calculation resources when abnormality detection is carried out, increases the time of package forwarding output, and has a larger influence on time delay sensitive C-SFC. The DPU is a specially designed processor unit for processing data, can be inserted on a computing node, has good acceleration capability on a neural network, improves the calculation speed in a hardware acceleration mode, and does not occupy the calculation resources of a forwarding function.

As shown in fig. 2, the control node deploys the detection model to consume CPU and GPU resources, and the storage and calculation integrated device deploys the model to consume CPU and GPU resources, so that the model can be unloaded to the DPU, and the DPU resources are consumed to perform accelerated calculation.

FIG. 3 is a flow chart of a method for deploying a multi-tier 6G computing power service chain anomaly detection mechanism, as shown in FIG. 3, comprising the steps of:

step S301, obtaining computing resource information, link resource information, computing resource remaining information and DPU remaining resource information of each computing device node in the multi-level 6G computing service chain to be detected.

Specifically, the computing resource information represents all of the computing resources of each computing device node; the link resource information represents the link resources connected by every two computing power equipment nodes; the computing resource remaining information represents remaining unoccupied computing resources of each computing device node; the DPU remaining resource information indicates unoccupied DPU devices.

Step S302, based on the preset time delay, a target decision model is constructed by using the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information.

Specifically, on the basis of considering the preset time delay, the obtained computing resource information, link resource information, computing resource residual information and DPU residual resource information are combined to construct a corresponding target decision model.

Step S303, solving the target decision model by utilizing a reinforcement learning algorithm based on a preset DQN framework to obtain an anomaly detection mechanism deployment scheme of the multi-level 6G computing power service chain to be detected.

Specifically, by presetting a reinforcement learning algorithm based on the DQN framework to solve the target decision model, an anomaly detection mechanism deployment scheme of a corresponding multi-level 6G computing power service chain to be detected can be obtained.

According to the deployment method of the multi-level 6G power service chain anomaly detection mechanism, on the basis of considering time delay, a target decision model is constructed through the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information of each power equipment node in the multi-level 6G power service chain to be detected, the computing resource and the network resource can be converted, and further, the optimal deployment scheme of the anomaly detection mechanism of the multi-level 6G power service chain to be detected can be obtained through solving the model through an algorithm. Therefore, by implementing the invention, different abnormal detection mechanism deployment schemes under different network states can be realized, and the problem of increasing the end-to-end delay of the C-SFC stream can be solved.

The embodiment provides a deployment method of a multi-level 6G computing power service chain abnormality detection mechanism, which can be used for a centralized controller. FIG. 4 is a flowchart of a method for deploying a multi-tier 6G computing power service chain anomaly detection mechanism, as shown in FIG. 4, comprising the steps of:

Step S401, obtaining computing resource information, link resource information, computing resource remaining information and DPU remaining resource information of each computing device node in the multi-level 6G computing service chain to be detected. Please refer to step S301 in the embodiment shown in fig. 3 in detail, which is not described herein.

Step S402, based on the preset time delay, a target decision model is constructed by using the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information.

Specifically, the step S402 includes:

step S4021, constructing a connected graph based on the computing resource information and the link resource information.

The link resource information is side information of the connectivity graph.

Specifically, a corresponding connectivity graph g= (N, L) may be constructed from the obtained computing resource information and link resource information. Wherein, N represents a computing power equipment node set, and the computing resource information of each computing power equipment node is stored; l represents a link set, and link resource information corresponding to each computing power equipment node is stored.

Further, the resources of the compute nodes and the centralized controller are denoted as S, where s=<S _CPU/GPU ,S _DPU >。

Step S4022, constructing a target decision model by using the connected graph, the computing resource remaining information and the DPU remaining resource information based on the preset time delay.

Specifically, on the basis of considering the preset time, the obtained connectivity graph, the computing resource remaining information and the DPU remaining resource information are combined to construct a corresponding target decision model.

In some alternative embodiments, step S4022 described above includes:

step a1, determining a preset decision planning requirement through a preset time slot processing method based on the connected graph and the preset time delay.

And a step a2, constructing a reward function based on the computing resource remaining information and the DPU remaining resource information.

And a step a3, constructing a target decision model based on the preset decision planning requirement, the rewarding function and the connection diagram.

Specifically, during operation, each small period of time is regarded as an operation time slot, a plurality of DSCSFCs are received in each operation time slot, each C-SFC needs to operate a fault detection model which is unique to itself to ensure the accuracy of the whole detection, and each C-SFC can select the deployment position of the fault detection model, so that each computing equipment node in the connectivity graph is theoretically possible to deploy the fault detection model.

Let T be the explicit requirement for failure detection time by each DSCSFC _req The DSCSFC occurring in the last time slot T-1 needs to be arranged in the time slot T, firstly according to T _req And sequencing is carried out, and DSCSFC with higher time demand is subjected to priority planning, so that the task with high time demand in the same time period can be ensured to be subjected to priority planning.

By the method for introducing the time slots, different C-SFCs can be divided according to the generated time slots, the ordering concept according to time requirements is introduced into the time slots, the ordering is completed according to the time requirements, and then the ordering is sequentially processed according to the ordering result, so that the preferential processing of the DSCSFC is ensured, and the dynamic scene can be effectively supported.

Further, the failure detection deployment of the DSCSFC is an adaptive adaptation, i.e. deploying as many SFCs as possible on the basis of meeting latency requirements. Therefore, time factors are prioritized in the deployment process, and the reward function R is modeled as a polynomial function, as shown in the following relation (1):

R＝aNode _CPU +bSW _CPU/GPU +cSW _DPU (1)

wherein: node _CPU Representing core node CPU computing resourcesThe remaining amount; SW (switch) _CPU/GPU A state matrix representing the remaining amount of node resources in the overall network topology; a state matrix representing the remaining amount of node DPU resources in the overall network topology. a. b and c respectively represent the state transition weights after successful deployment.

Furthermore, the rewarding function takes time as a main optimization target, and can provide better service for the C-SFC by using the DPU acceleration function more deeply, so that the scheduling time is saved, and the service quality is guaranteed.

Further, when three data of a, b and c are taken as integers, values of a= -1, b=1 and c=3 can be used for deploying more abnormality detection models on the basis of ensuring normal operation of the DSCSFC.

Further, in some optional embodiments, the step a3 includes:

step a31, respectively constructing an environment state space, an action space and a state transition based on the connection diagram and a preset decision planning requirement.

Step a32, constructing a target decision model based on the environment state space, the action space, the state transition and the rewarding function.

Specifically, within one slot τ e T, the environmental state includes physical network state information G _τ And computing resource information F of core controller and computing node in network node _τ Denoted as s (G) _τ ,F _τ ) E S. The physical network state information comprises network nodes, wherein the dimension of the network nodes is N, N represents the number of the network nodes, and the available link bandwidth resourcesWherein the dimension is |n|×|n|. Computing resource information F of core controller and computing node _τ . Wherein the Node information is Node _CPU 、SW _CPU/GPU 、SW _DPU . The ambient state space S can thus be expressed as +.>Its dimension size is N (2n+1).

Further, the action space a is composed of all operational actions of the learning agent, e.g., indexes for marking x=1, 2, k, |n| note each network node, and a e a actions represent actions of node resource changes.

Further, the motion space a= { a _node ,A _SW }. Wherein A is _node Representing the resource condition of the centralized controller; a is that _SW Representing the status of CPU/GPU computing resources and DPU resources in a compute node, the change in A is a dynamic case, so there are (2N+1) actions selectable for SFC.

Further, state transition P may be marked as (s _τ ,a _τ ,r _τ ,s _τ+1 ). Wherein s is _τ Representing a current network state; a, a _τ Representing orchestration actions taken in the current network state; r is (r) _τ Representing a reward for performing the orchestration action; s is(s) _τ+1 Representing execution of orchestration action a _τ The next network state entered later.

Further, for each network state s _τ E S, state transition probabilityRepresenting network state s after performing an action _τ From transition to s _τ+1 Is a probability of (2).

Further, a corresponding four-tuple MDP model, namely, a target decision model < S, a, P, R >, may be obtained, representing the environmental state space, the action space, the state transition, and the reward function, respectively.

Further, in each decision stage, the learning agent makes a corresponding decision a (τ) e a by observing the network state S (τ) e S, and after executing the decision a (τ), obtains a corresponding reward r (τ) and the network state S (τ+1) at the next moment. In turn, the awards obtained may be used to evaluate the effectiveness of the action.

Step S403, solving the target decision model by utilizing a reinforcement learning algorithm based on a preset DQN framework to obtain an anomaly detection mechanism deployment scheme of the multi-level 6G computing power service chain to be detected.

The execution logic of the reinforcement learning algorithm based on the DQN framework is preset as shown in the following table 1:

further, the execution flow of the reinforcement learning algorithm based on the DQN frame is shown in fig. 5, and includes:

1. initializing a neural network, loading network information and training data to be performed into the network, and then observing the current system state through an intelligent agent in a set mode to estimate the Q value of different actions to be taken in each state;

2. calculating the Q value of each action through a neural network by taking node resource state information in the current state as input;

3. selecting an action (e.g., using a greedy strategy, selecting an action randomly with a certain probability, otherwise selecting an action with the highest Q value)

4. Performing an action and observing the next state and the rewards obtained, transitioning from state to state;

5. storing the status, action, next status, and rewards in an experience playback buffer;

6. when a certain amount of state experiences are stored in the experience playback pool, randomly selecting a batch of samples from the experience playback buffer area, calculating a target Q value, and updating the neural network parameters to enable the neural network parameters to approach the target Q value;

7. Repeating the steps 2-6 until a preset number of training times is reached or a stable performance level is reached.

Further, through the execution logic and execution flow of the reinforcement learning algorithm based on the DQN framework, the corresponding action when the output Q value is maximum is selected as the selected unloading scheme and the deployment node, and the final abnormal detection mechanism deployment scheme is obtained.

The deployment method of the multi-level 6G computing power service chain abnormality detection mechanism provided by the embodiment combines the dynamic optimization strategy of the position of the model deployment and the load balance of the computing resources, and introduces the dynamic transition network state and the computing resource state of the MDP model. Further, a DSCSFC anomaly detection model deployment mechanism is designed by accumulating the total gains of the long-term system performance, so that the strategies of obtaining the states of the network and the computing resources and the deployment actions can be obtained, and the influence of network disturbance on rescheduling or re-optimizing the deployment model is reduced. At the same time, the method comprises the steps of,

the embodiment provides a deployment method of a multi-level 6G computing power service chain abnormality detection mechanism, which can be used for a centralized controller. FIG. 6 is a flowchart of a method for deploying a multi-tier 6G computing power service chain anomaly detection mechanism, as shown in FIG. 6, comprising the steps of:

Step S601, obtaining computing resource information, link resource information, computing resource remaining information and DPU remaining resource information of each computing device node in the multi-level 6G computing service chain to be detected. Please refer to step S301 in the embodiment shown in fig. 3 in detail, which is not described herein.

Step S602, based on the preset time delay, a target decision model is constructed by using the computing resource information, the link resource information, the computing resource remaining information and the DPU remaining resource information. Please refer to step S402 in the embodiment shown in fig. 4 in detail, which is not described herein.

And step S603, solving the target decision model by utilizing a reinforcement learning algorithm based on a preset DQN framework to obtain an anomaly detection mechanism deployment scheme of the multi-level 6G computing power service chain to be detected. Please refer to step S403 in the embodiment shown in fig. 4 in detail, which is not described herein.

Step S604, deploying a plurality of preset anomaly detection models in the multi-level 6G computing power service chain to be detected based on the anomaly detection mechanism deployment scheme.

Specifically, the determined deployment scheme of the anomaly detection mechanism can be used for deploying a plurality of corresponding preset anomaly detection models in the multi-level 6G computing power service chain to be detected.

Specifically, the step S604 includes:

In step S6041, a preset abnormality detection model is acquired.

Step S6042, determining a plurality of target computing power device nodes in the multi-level 6G computing power service chain to be detected based on the anomaly detection mechanism deployment scheme.

Specifically, according to the obtained deployment scheme of the anomaly detection mechanism, a corresponding deployment node, namely a target computing equipment node, in the multi-level 6G computing service chain to be detected can be determined.

Step S6043, deploying a preset anomaly detection model on each target computing device node.

Specifically, a corresponding preset anomaly detection model is deployed on each determined deployment node.

Step S605, performing abnormality detection on the multi-level 6G power calculation service chain to be detected by using a plurality of preset abnormality detection models to obtain an abnormality detection result.

Specifically, the anomaly detection of the multi-level 6G computing power service chain to be detected can be realized through a plurality of deployed preset anomaly detection models.

Specifically, the step S605 includes:

step S6051, performing abnormality detection on each target power computing device node by using a preset abnormality detection model to obtain an abnormality detection result of each target power computing device node.

Specifically, anomaly detection for each target computing device node may be achieved using a preset anomaly detection model deployed on each target computing device node.

Step S6052, determining an abnormality detection result of the multi-level 6G computing power service chain to be detected based on the abnormality detection result of each target computing power device node.

Specifically, according to the obtained abnormal detection result of each target computing equipment node, an abnormal detection result of the multi-level 6G computing service chain to be detected can be determined.

According to the deployment method of the multi-level 6G power service chain anomaly detection mechanism, on the basis of considering time delay, a target decision model is constructed through the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information of each power equipment node in the multi-level 6G power service chain to be detected, the computing resource and the network resource can be converted, and further, the optimal deployment scheme of the anomaly detection mechanism of the multi-level 6G power service chain to be detected can be obtained through solving the model through an algorithm. Further, by the obtained configuration scheme of the abnormality detection mechanism, a plurality of different preset abnormality detection models can be correspondingly deployed in the configuration scheme of the abnormality detection mechanism, and corresponding abnormality detection can be completed.

In an example, a method for unloading a multi-level 6G power-computing service chain abnormality detection mechanism based on DPU acceleration is provided, and flexible and efficient power-computing service chain abnormality detection is realized through cooperation of a data plane and a control plane.

Specifically, in this example, the architecture is divided into two main parts, the virtual layer is a data plane below, and the virtual layer is a control layer above, where the control layer is responsible for collecting network state information, running a deployment algorithm of the failure detection model of the computing power service chain, and issuing a corresponding decision result. The data plane, i.e. various network devices, may perform specific network tasks such as forwarding, computing, etc. The control plane comprises a bottom layer network multidimensional resource sensing, a K8s orchestrator and an ONOS controller. The data plane includes network components with different computing power and network resources, and is responsible for deployment of network functions constituting a computing power service chain, forwarding of data packets, and the like. The computing power and network information of the bottom layer network component are monitored and acquired through an application program interface API and an open source tool (such as Zabbix and the like) at the data layer, and the computing power and network information specifically comprises a CPU, an available memory, an available storage, a link delay and a link bandwidth, and the information is uploaded to the control layer, so that some computing tasks can be executed, and a computing result is returned.

As shown in fig. 1, the physical space is composed of physical entity devices (e.g., controllers, switches, etc.) and various heterogeneous resources (CPU and GPU computing resources, DPU computing resources, storage, and networks) existing in these physical devices. The centralized controller is used for centralized arrangement planning, the whole function is split into different functional components by different C-SFCs at the virtual level, chained arrangement is carried out, and then proper computing equipment is selected for deployment. The method is embodied in a physical layer, namely different virtualized function modules are deployed to different computing nodes under the control of a centralized orchestration controller. The specific flow is as follows:

The centralized orchestrator performs demand analysis on the C-SFC of the current fault detection model to obtain the residual computing resources of the current nodes and unoccupied DPU equipment through the application program interface.

The control layer analyzes the data and completes the decision, the control node models the computing resource information of each node device of the physical layer as a graph model, models the link resource information between the nodes as the side information of the graph, takes the whole network state as an MDP model, calculates a proper unloading and deployment strategy through a designed reinforcement learning algorithm, and finally completes the issuing of the execution information.

Further, in the network, the centralized controller is generally operated as a server, and has high computational power, so that various complex calculations can be performed. The power computing network equipment is integrated equipment integrating calculation, storage and forwarding, has certain calculation capacity, is basically provided by a CPU and GPU combination, deploys an abnormality detection model in a power computing node, occupies partial calculation resources when abnormality detection is carried out, increases the time of package forwarding output, and has a larger influence on time delay sensitive C-SFC. The DPU is a specially designed processor unit for processing data, can be inserted on a computing node, has good acceleration capability on a neural network, improves the calculation speed in a hardware acceleration mode, and does not occupy the calculation resources of a forwarding function. As shown in fig. 2, the control node deploys the detection model to consume CPU and GPU resources, and the storage and calculation integrated device deploys the model to consume CPU and GPU resources, so that the model can be unloaded to the DPU, and the DPU resources are consumed to perform accelerated calculation. In order to meet the deployment requirement of the DSCSFC fault detection function, the system mainly faces two major core problems, one of which is: the network state and the node resource utilization rate have dynamic performance, so that the deployment mechanism of the C-SFC anomaly detection model for deterministic optimization modeling has poor expandability, and the NP-hard problem is solved based on instantaneous system benefits. And two,: the deployment mechanism of the DSCSFC anomaly detection model needs to consider the two problems of the number of the AI models deployed on the DPU and the load balance of the computing resources under the condition that the upper bound of the time delay is met, and an optimal solution cannot be found.

Further, the C-SFC system is builtModulo is a connectivity graph g= (N, L), where N represents the set of nodes and L represents the set of links. The resources of the compute nodes and the centralized controller are denoted as S, where s=<S _CPU/GPU ,S _DPU >In the running process of the system, each small period of time is regarded as an operating time slot, a plurality of DSCSFCs are received in each operating time slot, each C-SFC needs to run a fault detection model which is independent of the C-SFC to ensure the accuracy of the whole detection, and each C-SFC can select the deployment position of the fault detection model, so that each node in the network has the possibility of deploying the fault detection model theoretically. Assuming that each DSCSFC has clear requirements for fault detection time as Treq, it is arranged in a time slot t to arrange the DSCSFC in the last time slot t-1, firstly, ordering is needed according to Treq, priority planning is conducted on the DSCSFC with higher time requirements, and therefore the task with high time requirements in the same time period can be guaranteed to be planned preferentially.

Further, the use condition of the resources in different time slots in the system model can be regarded as a state space, and the next different deployment positions correspond to different state transitions, and the optimal deployment scheme can be solved by giving different rewarding schemes to different transition schemes.

In order to realize dynamic deployment of the DSCSFC in the network, the novel deployment scheme considers the time of the DSCSFC and the resource consumption problem of the fault detection model, and under the resource constraint condition, the heterogeneous computing equipment deployment anomaly detection model quantity and the computing resource load balancing multi-objective optimization strategy in the network component layer are realized through combining the resource adaptation layer and the network component layer. The time delay requirement of DSCSFC and the occupation requirement of computing resources are comprehensively considered, DPU is introduced to accelerate, the deployable range is widened, the time consumption is optimized, and the deployment of time delay sensitive tasks in a 6G application scene can be better supported.

Specifically, since the DSCSFC fault detection model is a multi-state transition process, the present invention models the deployment problem of the anomaly detection model as a four-tuple MDP model < S, a, P, R >, representing the state space, the action space, the state transition probability, and the reward function, respectively. In each decision stage, the learning agent makes a corresponding decision a (tau) epsilon A by observing the system state S (tau) epsilon S, and obtains a corresponding system reward r (tau) and the system state S (tau+1) at the next moment after executing the decision a (tau). The obtained system prize r (τ) may, in turn, be used to evaluate the effectiveness of the action. Based on this, learning agents constantly interact with the network environment and adjust their deployment strategy of anomaly detection models, aiming at maximizing long-term system revenue.

The modeling process of the MDP model, S, a, P, R > refers to the description in step S4022 above.

Further, the corresponding DQN frame-based algorithm for offloading deployment of the algorithm for detecting abnormalities of the computational power business chain in this example is described with reference to step S403 above.

Further, the network model obtained through the algorithm can carry out online arrangement on the service request, the system state is input into the model, the network state is evaluated through an agent, a possible action Q value is output, the action with the maximum Q value is selected, namely, the selected unloading scheme and the deployment node, and finally the optimal adaptation and unloading strategy is obtained.

Further, experiments prove that the algorithm can dynamically process the deployment request of the fault detection model of the DSCSFC, and deploy the fault detection model according to corresponding calculation resource consumption and time delay requirements, so that flexible service is provided for the calculation service request by fully utilizing the dynamic change trend of the network state, and fine-granularity and time delay sensitive resource adaptation is realized.

The DPU acceleration-based multi-level 6G power calculation service chain abnormality detection mechanism unloading method provided by the embodiment comprises two planes (a data plane and a control plane), wherein the control plane comprises the step of analyzing power calculation service requirements to obtain fine granularity requirements such as resources, time delay and the like of services. The network node calculation information and the network link information can be monitored and collected in real time through the core controller, and the function module is executed to unload the deployment algorithm and complete cross-layer transfer and execution of the strategy. The data plane is formed by each infrastructure node, has calculation and forwarding capabilities, and carries out calculation acceleration processing fault detection algorithm by introducing a DPU. Through mutual cooperation between the data plane and the control plane, an optional deployment platform of the centralized controller and the DPU is introduced on the basis of a traditional deployment scheme, and the fault detection time requirements of various C-SFCs can be flexibly met through the multi-stage time characteristics, so that the scheduling time is saved, and the service quality is guaranteed. The method realizes the deployment of a flexible and intelligent DSCSFC fault detection model and the service guarantee of delay sensitive service flows.

Further, according to the embodiment, through the scheme of introducing the time slots, different C-SFCs are divided according to the generated time slots, the ordering concept according to time requirements is introduced into the time slots, the ordering is completed according to the time requirements, and then the ordering is sequentially processed according to the ordering result, so that the preferential processing of the DSSFCs is ensured, and the scheme can effectively support dynamic scenes.

Further, the embodiment combines a dynamic optimization strategy of the model deployment position and the computing resource load balance, and introduces an MDP model to dynamically convert the network state and the computing resource state. And a DSCSFC anomaly detection model deployment mechanism is designed by accumulating the total performance benefits of the long-term system by utilizing an RL algorithm, so that a strategy of a network, a computing resource state and deployment actions is obtained, and the influence of system disturbance on rescheduling or re-optimizing the deployment model is reduced.

Therefore, the multi-level 6G power-computing service chain abnormality detection mechanism unloading method based on DPU acceleration provided by the embodiment provides a flexible deployment scheme by introducing the centralized controller and the DPU, is convenient for further optimizing aiming at the demands of users, can better and rapidly carry out blocking processing based on a time slot allocation strategy, and has good adaptability to dynamically-changed resource scenes. The rewarding function takes time as a main optimization target, and can provide better service for C-SFC by using DPU acceleration function more deeply, thereby saving scheduling time and guaranteeing service quality.

The embodiment also provides a deployment device for the multi-level 6G computing power service chain abnormality detection mechanism, which is used for implementing the foregoing embodiments and preferred implementation manners, and is not described in detail. As used below, the term "module" may be a combination of software and/or hardware that implements a predetermined function. While the means described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

The embodiment provides a deployment device of a multi-level 6G computing power service chain abnormality detection mechanism, which is used for a centralized controller; as shown in fig. 7, includes:

the obtaining module 701 is configured to obtain computing resource information, link resource information, computing resource remaining information, and DPU remaining resource information of each computing power device node in the multi-level 6G computing power service chain to be detected.

A construction module 702 is configured to construct a target decision model using the computing resource information, the link resource information, the computing resource remaining information, and the DPU remaining resource information.

The solving module 703 is configured to solve the target decision model by using a reinforcement learning algorithm based on the DQN frame, so as to obtain an anomaly detection mechanism deployment scheme of the multi-level 6G computing power service chain to be detected.

In some alternative embodiments, the build module 702 includes:

and the first construction submodule is used for constructing the connectivity graph based on the computing resource information and the link resource information.

And the second construction submodule is used for constructing a target decision model by using the connected graph, the calculation resource residual information and the DPU residual resource information based on the preset time delay.

In some alternative embodiments, the second building sub-module comprises:

the determining unit is used for determining a preset decision planning requirement through a preset time slot processing method based on the connection diagram and the preset time delay.

A first construction unit for constructing a reward function based on the computing resource remaining information and the DPU remaining resource information.

And the second construction unit is used for constructing a target decision model based on the preset decision planning requirement, the rewarding function and the connected graph.

In some alternative embodiments, the second building element comprises:

the first construction subunit is used for respectively constructing an environment state space, an action space and state transition based on the connection diagram and a preset decision planning requirement.

And the second construction subunit is used for constructing a target decision model based on the environment state space, the action space, the state transition and the rewarding function.

In some alternative embodiments, the apparatus further comprises:

the deployment module is used for deploying a scheme based on an anomaly detection mechanism to deploy a plurality of preset anomaly detection models in the multi-level 6G computing power service chain to be detected.

The detection module is used for carrying out abnormality detection on the multi-level 6G power calculation service chain to be detected by utilizing a plurality of preset abnormality detection models to obtain an abnormality detection result.

In some alternative embodiments, the deployment module comprises:

and the acquisition sub-module is used for acquiring a preset abnormality detection model.

The determining submodule is used for determining a plurality of target computing power equipment nodes in the to-be-detected multi-level 6G computing power service chain based on the abnormal detection mechanism deployment scheme.

The deployment sub-module is used for deploying a preset abnormality detection model on each target computing equipment node.

In some alternative embodiments, the detection module includes:

and the detection sub-module is used for carrying out abnormality detection on each target power calculation equipment node by utilizing a preset abnormality detection model to obtain an abnormality detection result of each target power calculation equipment node.

The determining submodule is used for determining an abnormal detection result of the multi-level 6G computing power service chain to be detected based on the abnormal detection result of each target computing power equipment node.

Further functional descriptions of the above respective modules and units are the same as those of the above corresponding embodiments, and are not repeated here.

The multi-level 6G power service chain anomaly detection mechanism deployment apparatus in this embodiment is presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit ) circuits, processors and memory executing one or more software or fixed programs, and/or other devices that can provide the above-described functionality.

The embodiment of the invention also provides computer equipment, which is provided with the deployment device of the multi-level 6G computing power service chain abnormality detection mechanism shown in the figure 7.

Referring to fig. 8, fig. 8 is a schematic structural diagram of a computer device according to an alternative embodiment of the present invention, as shown in fig. 8, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are communicatively coupled to each other using different buses and may be mounted on a common motherboard or in other manners as desired. The processor may process instructions executing within the computer device, including instructions stored in or on memory to display graphical information of the GUI on an external input/output device, such as a display device coupled to the interface. In some alternative embodiments, multiple processors and/or multiple buses may be used, if desired, along with multiple memories and multiple memories. Also, multiple computer devices may be connected, each providing a portion of the necessary operations (e.g., as a server array, a set of blade servers, or a multiprocessor system). One processor 10 is illustrated in fig. 8.

The processor 10 may be a central processor, a network processor, or a combination thereof. The processor 10 may further include a hardware chip, among others. The hardware chip may be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general-purpose array logic, or any combination thereof.

Wherein the memory 20 stores instructions executable by the at least one processor 10 to cause the at least one processor 10 to perform a method for implementing the embodiments described above.

The memory 20 may include a storage program area that may store an operating system, at least one application program required for functions, and a storage data area; the storage data area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, memory 20 may optionally include memory located remotely from processor 10, which may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

Memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk, or solid state disk; the memory 20 may also comprise a combination of the above types of memories.

The computer device also includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

The embodiments of the present invention also provide a computer readable storage medium, and the method according to the embodiments of the present invention described above may be implemented in hardware, firmware, or as a computer code which may be recorded on a storage medium, or as original stored in a remote storage medium or a non-transitory machine readable storage medium downloaded through a network and to be stored in a local storage medium, so that the method described herein may be stored on such software process on a storage medium using a general purpose computer, a special purpose processor, or programmable or special purpose hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, a solid state disk or the like; further, the storage medium may also comprise a combination of memories of the kind described above. It will be appreciated that a computer, processor, microprocessor controller or programmable hardware includes a storage element that can store or receive software or computer code that, when accessed and executed by the computer, processor or hardware, implements the methods illustrated by the above embodiments.

Although embodiments of the present invention have been described in connection with the accompanying drawings, various modifications and variations may be made by those skilled in the art without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope of the invention as defined by the appended claims.

Claims (10)

1. The deployment method of the multi-level 6G computing power service chain abnormality detection mechanism is characterized by being used for a centralized controller; the method comprises the following steps:

acquiring computing resource information, link resource information, computing resource residual information and DPU residual resource information of each computing power equipment node in a multi-level 6G computing power service chain to be detected;

based on preset time delay, constructing a target decision model by utilizing the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information;

and solving the target decision model by using a reinforcement learning algorithm based on a preset DQN framework to obtain an anomaly detection mechanism deployment scheme of the multi-level 6G computing service chain to be detected.

2. The method of claim 1, wherein constructing a target decision model using the computing resource information, the link resource information, the computing resource remaining information, and the DPU remaining resource information based on a preset time delay comprises:

Constructing a connectivity graph based on the computing resource information and the link resource information;

and constructing the target decision model by utilizing the communication graph, the computing resource residual information and the DPU residual resource information based on a preset time delay.

3. The method of claim 2, wherein constructing the target decision model using the connectivity graph, the computing resource remaining information, and the DPU remaining resource information based on a preset latency, comprises:

determining a preset decision planning requirement through a preset time slot processing method based on the connected graph and the preset time delay;

constructing a reward function based on the computing resource remaining information and the DPU remaining resource information;

and constructing the target decision model based on the preset decision planning requirement, the reward function and the connected graph.

4. A method according to claim 3, wherein constructing the target decision model based on the preset decision planning requirements, the reward function and the connectivity map comprises:

respectively constructing an environment state space, an action space and state transition based on the connected graph and the preset decision planning requirement;

The target decision model is constructed based on the environmental state space, the action space, the state transitions, and the reward function.

5. The method according to claim 1, wherein the method further comprises:

a plurality of preset abnormality detection models are arranged in the multi-level 6G computing power service chain to be detected based on the abnormality detection mechanism deployment scheme;

and carrying out abnormality detection on the multi-level 6G computing power service chain to be detected by utilizing the plurality of preset abnormality detection models to obtain an abnormality detection result.

6. The method of claim 5, wherein deploying a plurality of preset anomaly detection models within the multi-level 6G computing power service chain to be detected based on the anomaly detection mechanism deployment scheme comprises:

acquiring a preset abnormality detection model;

determining a plurality of target computing equipment nodes in the to-be-detected multi-level 6G computing service chain based on the anomaly detection mechanism deployment scheme;

and deploying the preset anomaly detection model on each target computing equipment node.

7. The method according to claim 5, wherein performing anomaly detection on the multi-level 6G computing power service chain to be detected by using the plurality of preset anomaly detection models to obtain anomaly detection results, comprises:

Performing abnormality detection on each target power calculation equipment node by using the preset abnormality detection model to obtain an abnormality detection result of each target power calculation equipment node;

and determining an abnormality detection result of the multi-level 6G computing power service chain to be detected based on the abnormality detection result of each target computing power equipment node.

8. The deployment device of the multi-level 6G computing power service chain abnormality detection mechanism is characterized by being used for a centralized controller; the device comprises:

the acquisition module is used for acquiring computing resource information, link resource information, computing resource residual information and DPU residual resource information of each computing equipment node in the multi-level 6G computing service chain to be detected;

the construction module is used for constructing a target decision model by utilizing the computing resource information, the link resource information, the computing resource residual information and the DPU residual resource information;

and the solving module is used for solving the target decision model by utilizing a reinforcement learning algorithm based on a preset DQN framework to obtain an anomaly detection mechanism deployment scheme of the multi-level 6G computing power service chain to be detected.

9. A computer device, comprising:

a memory and a processor, the memory and the processor being communicatively connected to each other, the memory having stored therein computer instructions, the processor executing the computer instructions to perform the multi-tier 6G power chain anomaly detection mechanism deployment method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the multi-tier 6G power service chain anomaly detection mechanism deployment method of any one of claims 1 to 7.