A mobility-aware multi-objective service function chain deployment and migration method

Abstract

The present invention discloses a mobile-aware multi-target business function chain deployment and migration method, which is used to solve the decision-making problem of deployment and migration of business function chains of mobile terminals with multi-target business requests. The method mainly constructs a network topology structure, and uses the available resource information and mobile terminal business request information in the physical network topology as the input of a neural network based on deep reinforcement learning. The neural network of the control center calculates the most optimized strategy to achieve the optimal deployment and migration strategy of the business function chain. The present invention comprehensively considers the decision-making influence between the physical network topology, mobile terminals and the control center, realizes efficient business function chain deployment and migration strategy, and obtains the optimal deployment and migration position under different circumstances, so as to provide services for more business function chains and optimize resource usage costs and migration costs.

Description

Technical Field

[0001] The technical field involved in the present invention is computer network, and in particular, relates to a mobile-aware multi-target service function chain deployment and migration method. Background Art

[0002] Compared with using traditional dedicated hardware to provide user request services, as network traffic increases, more dedicated equipment is needed. The emergence of NFV allows the business function chain to be placed on a ready-made general-purpose server through virtualization technology or container technology, which improves the flexibility and scalability of business function chain deployment, while also reducing the expensive cost brought by dedicated equipment. However, deploying the business function chain on a general-purpose server will bring disadvantages to the user's service quality (end-to-end latency, cost, etc.), and the unreasonable deployment of the business function chain will also reduce the acceptance rate of the business function chain. Therefore, many people in the industry are committed to the research of business function chain orchestration technology.

[0003] In the existing technology, for example, ILP and MILP models are constructed to solve the problem of business function chain deployment, but because of the long solution time, they are not suitable for scenarios with large physical network scales; some heuristic algorithms are also used to implement business function chain deployment, but they are prone to fall into local optimality; in order to implement online business function chain deployment, some algorithms based on deep reinforcement learning have been proposed, but they have not yet considered the emergence of new extreme scenarios and services such as industrial Internet, AR / VR, holographic communication, and full-area coverage, which have high requirements for latency and reliability. For mobile terminals, it is also necessary to consider issues such as changes in network traffic and end-to-end latency due to mobility, which may lead to service interruptions. In order to ensure the continuity of business during the mobile terminal's movement, business function chain migration is a good solution.

[0004] However, most existing methods for implementing the migration of business function chains do not take into account the mobility of mobile terminals. Some methods take into account the mobility of mobile terminals, for example, by adjusting the position of business functions in the business function chain or by horizontal or vertical expansion technology to achieve migration, but they do not take latency as the optimization goal, but only consider the optimization of the placement cost and migration cost of business functions; or only consider the migration problem of the business function chain, and have not yet considered the deployment problem of the business function chain. However, in reality, the deployment problem of the business function chain is often a multi-objective problem, which requires considering the different requirements of mobile terminals for the service at the same time. The deployment strategy of the business function chain will also affect its migration strategy, and it is necessary to consider the deployment and migration of the business function chain at the same time. Summary of the invention

[0005] Purpose of the invention: In order to optimize the acceptance rate, end-to-end communication delay and cost of the service function chain of mobile terminals, achieve an overall improvement in service efficiency, and achieve optimal deployment and migration, the present invention provides a mobile-aware multi-objective service function chain deployment and migration method.

[0006] Technical solution: A mobile-aware multi-objective service function chain deployment and migration method, which is based on the mobile terminal submitting service request information to the control center, and the control center neural network calculating the deployment and migration location provided by the physical network topology, so as to improve the service function link yield, reduce the end-to-end delay of the service function chain, the resource usage cost and the migration cost during the deployment process, and provide the corresponding path, including the following steps:

[0007] (1) Constructing a relational network model, including establishing the physical network topology, the network topology structure of the interactive relationship between the mobile terminals and the control center;

[0008] (2) determining an overall optimization target based on physical network topology information and mobile terminal service request information, wherein the physical network topology information includes location information of server nodes, routing nodes, and access nodes, computing resource capacity information, storage resource capacity information, and related link bandwidth capacity information; the mobile terminal service request information includes the type of service request (deployment or migration), the location information of the mobile terminal, the demand for computing resources, storage resources, and bandwidth resources by the mobile terminal service request, and the restriction requirements on end-to-end delay; the overall optimization target is determined by the reception rate of the service function chain, the end-to-end delay, the resource usage cost during the deployment process, and the migration cost;

[0009] (3) Construct and train a neural network model to simulate the deployment and migration of service function chains in T consecutive time slots. In each time slot t, the physical network topology information and service request information of the current time slot are collected as input. According to the overall optimization goal, the control center neural network is used to calculate and output the deployment or migration location of the service function chain and allocate deployment or migration resources. The service function chain request in the current time slot can only enter the next time slot after it is processed.

[0010] (4) The control center manages the physical network topology to provide resource services for business requests and calculates the remaining resource information of the physical network topology. The entire deployment and migration process is completed only when all business function chains in T consecutive time slots are processed.

[0011] Furthermore, in the relational network model described in step (1), two-way information communication is performed between the physical network topology, the mobile terminal and the control center, including the sending and receiving of instructions. The mobile terminal sends service request information to the control center, and the control center issues instructions to provide services for the service request based on the resource information and service request information of the physical network topology, and the physical network topology provides services to the mobile terminal.

[0012] Furthermore, each mobile terminal provides the control center with its own location information, computing resources, storage resources, bandwidth resources and maximum tolerable end-to-end delay required for the service request. The control center calculates the resources required for the mobile terminal service request based on this information and the information of the physical network topology. Therefore, there is no direct connection between the mobile terminal and the physical network topology. The relationship between the service provider and the service provider is realized through the control center.

[0013] In the assumed environment, each mobile terminal generates only one service request, that is, the service request of each mobile terminal can be represented by a service function chain (SFC), which simulates the deployment and migration of the service function chain during the movement of the mobile terminal in T time slots. The environment also includes a physical network topology including server nodes, routing nodes, access nodes and related links, as well as a control center. The control center decides how to arrange the deployment and migration of the service function chain based on the corresponding information.

[0014] In the above method, in step (2), Ω={SFC 1 ,SFC 2 ,...,SFC |U|} represents the service request information of |U| mobile terminals, each They represent the service function chain request information of the u-th mobile terminal, including all the service functions F of the service function chain. u ={f u1 ,f u2 ,...,f uh ,...,f u|Fu|}, each business function f uh There is a fixed demand for resources. represents f uh The demand for computing and storage resources, represents the maximum tolerable end-to-end communication delay of the service function chain, b u represents the required bandwidth of the service function chain, L u A represents the logical link of the service function chain. The ability of the terminal to which the service function chain is attached to transmit data during wireless access is A. u , the life cycle of the business function chain is TTL u , indicating that the occupied resources will be released after its life cycle ends. Indicates the position of the mobile terminal to which the service function chain is attached in the current time slot and the position in the next time slot. Except for the position feature which changes with the movement of the mobile terminal, other attributes are the initial given values ​​in the whole process.

[0015] The information of the physical network topology is expressed as G = (N∪E), where N represents the nodes of the physical network topology and E represents the links of the physical network topology. The nodes include server nodes, routing nodes, and access nodes, and each node has its own pre-planned location. In addition, the server nodes have available computing resources that can provide services. and available storage resources There is bandwidth available on each link As the physical network topology provides services to mobile terminals, the available resources on the server nodes also change dynamically.

[0016] Considering the mobility of mobile terminals and the expectation to improve the service function link rate, reduce the end-to-end delay of the service function chain, and the resource usage cost and migration cost during the deployment process, the overall optimization goals are given below, including the successful reception rate of the service function chain, the end-to-end delay of the service function chain, the resource usage cost and the migration cost. The delay includes three parts: the wireless access delay of the mobile terminal accessing the physical network, the processing delay of the server node for the service function, and the delay of the link transmitting data, which are respectively used and The cost consists of two parts: resource usage cost and migration cost, which are respectively and The calculation formulas are as follows:

[0017]

[0018] Where W is the channel bandwidth, ρ 2 is the signal-to-noise ratio;

[0019]

[0020] in, is a binary variable whose value is 1, indicating that in time slot t, f uh Placed on server node n, In time slot t, the service function chain F u The rate at which data arrives at server node n, is the server node n for business function f uh The rate of service provided;

[0021]

[0022]

[0023] in, is a binary variable whose value is 1, indicating that SFC u Logical link l pq Placed on the physical link ij superior, In time slot t, the service function chain F u Data arrives at the physical link e ij The rate on Is the physical link ij F u The service rate provided, D(t) is the end-to-end delay of all service function chains that are successfully deployed or migrated in time slot t;

[0024]

[0025] in, represents the unit computation cost on server node n in time slot t, represents the unit storage cost on server node n;

[0026]

[0027]

[0028] Among them, m fuh (t) is a binary variable, which determines whether the service function f uh Does it need to be migrated? Yes uh The placement position in time slot t-1, Yes uh The placement position in time slot t, is the shortest path between these two locations, using the business function f uh exist and The link transmission delay between uh C(t) is the migration cost of all business function chains that are successfully deployed or migrated in time slot t.

[0029] In order to increase the success rate of business function chain acceptance, a fixed reward r will be given to business function chains that can be successfully deployed or migrated. base , which will increase the reception rate of the service function chain. Therefore, the overall optimization goal of the present invention is to minimize the end-to-end delay and cost of the service function chain while maximizing the reception rate of the service function chain. This goal can be expressed as follows:

[0030] max R=[R base,-D,-C]

[0031] Where num is the number of business function chains that are successfully deployed or migrated, R base =num·r base , denote the reward, end-to-end delay and cost of successful deployment of all successfully deployed or migrated service function chains in the entire time slot T, R base The larger the -D and -C are, the more business function chains are successfully received and the smaller the end-to-end delay and cost are.

[0032] Furthermore, the neural network in step (3) sets weights for end-to-end delay, deployment cost, and migration cost based on the optimization objective function. The structure and training process of the neural network are as follows:

[0033] (31) Set a set of weights, w = (w 1 ,w 2 ), respectively representing the weights of average end-to-end delay, deployment and migration costs. The objective function is expressed as:

[0034] max R base -w 1 ·Dw 2 C;

[0035] (32) The business function chain deployment and migration problem is divided into two sub-problems: business function chain deployment and migration. The two sub-problems are linked by the migration cost, and the Advantage-Actor-Critic algorithm is used to train the network.

[0036] (33) According to (32), the objective function solution of the two sub-problems is further converted into T optimization sub-problems. These optimization sub-problems make decisions through neighbor-based strategies. During the training process, the training parameters of the previous neural network will be used as the initial parameters of the next neural network. After iterative learning, more business function chains are received, resulting in lower end-to-end latency, deployment cost, and migration cost. The following conditions need to be met during the iteration process:

[0037] In the current time slot t, the service resources available to any server node should be higher than the resources used by all service function chains using the node. Similarly, the bandwidth resources available to any link should be higher than the bandwidth used by all service function chains using the link. The constraints are as follows:

[0038]

[0039]

[0040] in, represents the computing resources available to server node n in time slot t, represents the storage resources available to server node n in time slot t;

[0041]

[0042] in, represents the physical link e in time slot t ij Available bandwidth;

[0043] In the current time slot t, the delay of the service function chain that can be successfully deployed or migrated must be less than or equal to its maximum tolerable delay.

[0044]

[0045] Where D u (t) is the business function chain F u The end-to-end latency of current communications.

[0046] Furthermore, in step (4), the control center obtains the deployment or migration position of the service function chain in each time slot, and needs to determine whether the current service function chain can be fully deployed or migrated. If so, it sends instructions to the mobile terminal and the physical network topology to allow the two to interact, be served, and provide services.

[0047] Beneficial effect: The method described in the present invention is for the deployment and migration of multi-objective business function chains for mobile perception. The method comprehensively considers the information of physical network topology and mobile terminal business request information, and implements the objective functions of business function link yield, end-to-end delay, deployment cost and migration cost based on the algorithm of the control center, which has not been considered in previous work. Compared with the deployment and migration method based on the greedy idea, the method of the present invention has improved performance in terms of business function chain reception rate, end-to-end delay, deployment cost and migration cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the process of the method of the present invention;

[0049] Figure 2 A schematic diagram of the relationship between the physical network topology, mobile terminals and the control center in the present invention;

[0050] Figure 3 It is a schematic diagram for comparing the average receiving rate of a neural network trained with different numbers of service function chains and based on the greedy idea in an embodiment of the present invention;

[0051] Figure 4A schematic diagram showing a comparison of average end-to-end delays of a neural network trained with different numbers of service function chains and based on a greedy approach in an embodiment of the present invention;

[0052] Figure 5 It is a schematic diagram comparing the average resource usage cost of a neural network trained with different numbers of service function chains and based on the greedy idea in an embodiment of the present invention;

[0053] Figure 6 It is a schematic diagram comparing the average migration cost of a neural network trained with different numbers of business function chains and based on the greedy idea in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to explain the technical solution disclosed by the present invention in detail, the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0055] The present invention provides a mobile-aware multi-target service function chain deployment and migration method, which mainly aims to improve the service function link yield as much as possible while reducing the end-to-end delay and cost of the service function chain of a mobile terminal in a mobile scenario.

[0056] The 2022 Mobile Internet White Paper shows that mobile smart terminals are growing across the board. In 2021, domestic smartphone shipments reached 343 million units, a year-on-year increase of 15.9%, and wearable device shipments reached 140 million units, a year-on-year increase of 25.4%. New mobile terminals have great development potential. By the end of 2021, the number of drones registered in the national drone real-name registration system totaled 830,000, an increase of 44.9% over 2020. my country's industrial robot shipments reached 256,000 units, a year-on-year increase of 49.5%, and VR headset shipments reached 3.65 million units, a year-on-year increase of 13.5%. In view of the development trend of mobile terminals, we have to consider how to provide services for the multi-target business function chain of mobile perception and how to ensure that mobile terminals can maintain communication while moving.

[0057] The mobile-aware multi-objective service function chain deployment and migration method provided by the present invention is based on deep reinforcement learning and comprehensively considers the reception rate of the service function chain, the end-to-end delay of the service function chain, the resource usage cost and the migration cost. In order to solve this multi-objective problem, the method described in the present invention is an efficient and convenient method for deploying and migrating service requests of mobile terminals. By making decisions on the deployment and migration locations based on the location information and resource information of all nodes in the physical network topology, the location information of the mobile terminal, and the service request information of the mobile terminal, as many service function chains as possible are received, lower end-to-end delay is met, and resource usage costs and service function chain migration costs are reduced.

[0058] The implementation process of the technical solution provided by the present invention is described in detail below.

[0059] The method to which the present invention belongs is to implement the deployment and migration decision of the service function chain. It mainly includes the control center, mobile terminal and physical network topology. The control center collects the current physical network topology and the service request information of the mobile terminal, and is responsible for the deployment and migration of the mobile terminal request service (service function chain); the mobile terminal submits a deployment or migration request to the control center, and waits for the nodes and links in the physical network topology to provide services for it; the nodes and links in the physical network topology are scheduled by the control center to provide deployment or migration services for the mobile terminal.

[0060] The main implementation process of the method of the present invention is as follows Figure 1 As shown, based on the above technical solution, further detailed description is made in the embodiment, which specifically includes the following steps:

[0061] (1) constructing a relational network model, including establishing a network topology structure of the interaction between a physical network topology, a mobile terminal service request, and a control center; the physical network topology provides services for the mobile terminal, and the control center provides a determined deployment or migration location for the mobile terminal;

[0062] (2) Determine the overall optimization goal based on the physical network topology service resource information and mobile terminal service request information;

[0063] (3) Construct and train a neural network model to simulate the deployment and migration of service function chains in T consecutive time slots. In each time slot t, the physical network topology information and service request information of the current time slot are collected as input. According to the overall optimization goal, the control center neural network is used to calculate and output the deployment or migration location of the service function chain and allocate deployment or migration resources. The service function chain request in the current time slot can only enter the next time slot after it is processed.

[0064] (4) The control center obtains the information output by the neural network and manages the physical network topology to provide resource services for business requests. When the business function chain of the mobile terminal can be fully deployed or migrated, the control center executes the information output by the neural network and calculates the remaining resource information of the physical network topology. Otherwise, the control center rejects the business request of the mobile terminal. The entire deployment and migration process is completed only when all business function chains in T consecutive time slots are processed.

[0065] In actual applications, for step (1), an interactive network is first established based on the area and geographical location of the mobile terminal and the area and geographical location of the nodes in the physical network topology, including the physical network topology, the mobile terminal and the control center. The network is used to establish a relationship network among the three and realize mutual communication, including the sending and receiving of instructions, and the establishment of deployment and migration mechanisms.

[0066] Further combined with Figure 2 This step is based on the above relationship. Consider the relationship between a mobile terminal and the physical network topology that can provide services for it. The nodes and links in the physical network topology can provide services for the mobile terminal. The mobile terminal has mobility, and its movement trajectory is planned in advance. During the movement, the service previously provided by the physical network topology to the mobile terminal service request may not meet the service quality of the current service request. The decision that the control center needs to make is to provide a deployment strategy for the newly arrived service request and a migration strategy for the service request that has been successfully provided, so as to reduce the end-to-end delay and cost. The implementation process is described in detail below.

[0067] 1. Determine the service request information and location information of the mobile terminal

[0068] In the present invention, the control center provides services to mobile terminals according to their business needs. Under the assumption that each mobile terminal has a business request, the physical network topology must provide it with the resources required for this business request. Since the purpose of the service is to reduce the end-to-end delay, resource usage cost and migration cost of the business function chain, and to increase the reception rate of the business function chain, the increase in the reception rate of the business function chain will lead to an increase in the end-to-end delay. How to balance these three goals is the difficulty of this problem.

[0069] Use Ω={SFC 1 ,SFC 2 ,...,SFC |U|} represents the service request information of |U| mobile terminals, each Respectively represent the service function chain request information of the u-th mobile terminal, including all the service functions of the service function chain Each business function uh There is a fixed demand for resources. represents f uh The demand for computing and storage resources, represents the maximum end-to-end communication delay of the service function chain, b u represents the required bandwidth of the service function chain, L u A represents the logical link of the service function chain. The ability of the terminal to which the service function chain is attached to transmit data during wireless access is A. u , the life cycle of the business function chain is TTLu , indicating that the occupied resources will be released after its life cycle ends. Indicates the position of the mobile terminal to which the service function chain is attached in the current time slot and the position in the next time slot. Except for the position feature which changes with the movement of the mobile terminal, other attributes are the initial given values ​​in the whole process.

[0070] The information of the physical network topology is expressed as G = (N∪E), where N represents the nodes of the physical network topology and E represents the links of the physical network topology. The nodes include server nodes, routing nodes, and access nodes, and each node has its own pre-planned location. In addition, the server nodes have available computing resources that can provide services. and available storage resources There is bandwidth available on each link As the physical network topology provides services to mobile terminals, the available resources on the server nodes also change dynamically.

[0071] Considering the mobility of mobile terminals and the expectation to improve the service function link yield, reduce the end-to-end delay of the service function chain, and the deployment and migration costs, the overall optimization goals are given below, including the service function link yield, end-to-end delay, resource usage cost, and migration cost. The delay includes three parts: the wireless access delay of the mobile terminal accessing the physical network, the processing delay of the server node for the service function, and the delay of the link transmitting data, which are respectively and The cost consists of two parts: resource usage cost and migration cost, which are respectively and The calculation formulas are as follows:

[0072]

[0073] Where W is the channel bandwidth, ρ 2 is the signal-to-noise ratio;

[0074]

[0075] in, is a binary variable whose value is 1, indicating that in time slot t, f uh Placed on server node n, In time slot t, the service function chain F u The rate at which data arrives at server node n, is the server node n for business function f uh The rate of service provided;

[0076]

[0077]

[0078] in, is a binary variable whose value is 1, indicating that SFC u Logical link l pq Placed on the physical link ij superior, In time slot t, the service function chain F u Data arrives at the physical link e ij The rate on Is the physical link ij F u The service rate provided, D(t) is the end-to-end delay of all service function chains that are successfully deployed or migrated in time slot t;

[0079]

[0080] in, represents the unit computation cost on server node n in time slot t, represents the unit storage cost on server node n;

[0081]

[0082]

[0083] Among them, m fuh (t) is a binary variable, which determines whether the service function f uh Does it need to be migrated? Yes uh The placement position in time slot t-1, Yes uh The placement position in time slot t, is the shortest path between these two locations, and C(t) is the deployment and migration cost of all business function chains that are successfully deployed or migrated in time slot t.

[0084] In order to increase the success rate of business function chain acceptance, a fixed reward r will be given to business function chains that can be successfully deployed or migrated. base , which will increase the reception rate of the service function chain. Therefore, the goal of the present invention is to minimize the end-to-end delay and cost of the service function chain while maximizing the reception rate of the service function chain. The goal can be expressed as follows:

[0085] max R=[R base ,-D,-C]

[0086] Where num is the number of business function chains that are successfully deployed or migrated, Rbase =num·r base , denote the reward, end-to-end delay and cost of successful deployment of all successfully deployed or migrated service function chains in the entire time slot T, R base The larger the -D and -C are, the more business function chains are successfully received and the smaller the end-to-end delay and cost are.

[0087] 2. Build a network model based on deployment, migration algorithms and Advantage-Actor-Critic (A2C) algorithms

[0088] The algorithm model is based on a multi-objective problem. The neural network needs to make a trade-off between maximizing the service function link yield and minimizing the service function chain end-to-end delay, resource usage cost and migration cost. It is expected that when the receiving rate is high, the end-to-end delay and resource usage and migration cost can be guaranteed to be low. The problem is divided into two sub-problems: deployment and migration. The same neural network is constructed to solve the two sub-problems.

[0089] The two sub-problems are linked by the migration cost. In each time slot t, the new service function chain is deployed first, and then the service function chain to be migrated is migrated. The ultimate goal is to make the service function link yield high, the end-to-end delay low, and the deployment and migration cost low in all time slots T. The input of the network is the physical network topology information and the mobile terminal service request information, and the output is the location of service function chain deployment and migration.

[0090] The present invention finally obtains the deployment and migration strategy through multiple iterations, completes the deployment and migration of the service function chains requested by all mobile terminals in all time slots T, and after one iteration, uses the neural network parameters after the iteration as the initial parameters of the neural network in the next iteration, thereby accelerating the training of the network model.

[0091] 3. Use the trained network model to obtain the deployment and migration strategy of the business function chain

[0092] After the neural network is trained, it is applied to the deployment and migration of the business function chain to obtain the deployment and migration strategy.

[0093] 4. The control center manages the physical network topology and provides services for mobile terminals

[0094] The control center obtains the deployment or migration position of the business function chain in each time slot, and needs to determine whether the current business function chain can be fully deployed or migrated. If so, it sends instructions to the mobile terminal and physical network topology to allow the two to interact, be served, and provide services.

[0095] In order to verify the actual effect of the present invention, a simulation experiment was carried out in this embodiment. In order to better illustrate the effect of the present invention, an algorithm based on the greedy idea was used for comparison. The Uunet network topology (49 nodes, 84 edges) in TopologyZoo was used for the experiment, and the performance comparison under different numbers of business requests was given. The effects mainly include the following points:

[0096] (1) Figure 3 As shown, the method of the present invention analyzes the change of the average business function link yield as the number of business function chains increases. As the number of business function chains increases, the average reception rate of the two methods shows a downward trend, but the overall number of receptions is increasing, which is in line with expectations. Compared with the algorithm based on the greedy idea, the average reception rate of the method of the present invention is generally higher. When the number of requested business function chains is 200, the reception rates of the two methods are the same; but as the number of requested business function chains increases, the reception rate of the method of the present invention is higher. When the number is 240, the method proposed in the present invention is about 10.71% higher than the comparison method in terms of reception rate, when the number is 280, it is about 16.33% higher, when the number is 320, it is about 21.95% higher, and when the number is 360, it is about 16.28% higher.

[0097] (2) Figure 4 As shown, the method of the present invention analyzes the change of the average end-to-end communication delay of the service function chain as the number of service function chains increases. As the number of service function chains increases, more service function chains are received in the physical network, and the average end-to-end delay of the two methods shows an overall upward trend, which is in line with expectations. Compared with the algorithm based on the greedy idea, the average end-to-end delay of the method of the present invention is lower. When the number of requested service function chains is 200, the method proposed by the present invention is about 11.71% lower than the comparison method in terms of end-to-end delay, when the number is 240, it is about 9.55% lower, when the number is 280, it is about 9.55% lower, when the number is 320, it is about 8.75% lower, when the number is 320, it is about 9.92% lower, and when the number is 360, it is about 10.73% lower.

[0098] (3) Figure 5 As shown, the method of the present invention analyzes the change of the average resource usage cost as the number of business function chains increases. The average resource usage cost does not continue to rise or fall, because the business function chain and deployment process involve the release of physical resources, and the resource usage cost is related to the amount of remaining resources. Even though the average resource usage cost varies for different business function chains, the average resource usage cost of the method of the present invention is lower than that of the greedy-based algorithm.

[0099] (4) Figure 6 As shown, the method of the present invention analyzes the change of the average migration cost as the number of business function chains increases. Compared with the algorithm based on the greedy idea, when the number of business function chain requests is small, the average migration cost is low, because the number of business function chains to be migrated is small at this time. As the number of requested business function chains increases, the average business function chain migration cost of the method of the present invention is higher, because it is reasonable to sacrifice the migration cost in the process of business function chain migration to make the overall optimization goal better.

[0100] In summary, in the experimental results, compared with the greedy algorithm, the method of the present invention is up to 21.95% higher in average service function link yield, up to 11.71% lower in average end-to-end delay, and always maintains a low level in average resource usage cost. Although it is slightly insufficient in average service function chain migration cost, it is reasonable and improves the overall optimization goal. Therefore, the method proposed by the present invention is better overall.

Claims

1. A mobile-aware multi-target service function chain deployment and migration method, wherein the method is based on a mobile terminal submitting service request information to a control center, and the control center neural network calculates the deployment and migration locations provided by the physical network topology, thereby improving the service function link yield and reducing the end-to-end delay of the service function chain, the resource usage cost during the deployment process, and the migration cost, and is characterized by: The steps include: (1) Constructing a relational network model, including establishing the physical network topology, the network topology structure of the interactive relationship between the mobile terminals and the control center; The mobile terminal, the physical network topology and the control center communicate with each other, and the mobile terminal makes a business request to the control center, and the physical network topology provides services to the mobile terminal according to the decision of the control center; (2) determining an overall optimization target based on the physical network topology information and the mobile terminal service request information, wherein the overall optimization target is determined by the service function link yield, end-to-end latency, resource usage cost, and migration cost; The objective function expression corresponding to the overall optimization objective is as follows: max R=[R base ,-D,-C] The end-to-end delay expression of the service function chain is as follows: in is the wireless access delay of mobile terminal u accessing the physical network in time slot t, The processing delay of the server node for business functions, The delay of link transmitting data; Where W is the channel bandwidth, ρ 2 is the signal-to-noise ratio; in, is a binary variable whose value is 1, indicating that in time slot t, f uh Placed on server node n, In time slot t, the service function chain F u The rate at which data arrives at server node n, is the server node n for business function f uh The rate of service provided; in, is a binary variable whose value is 1, indicating that SFC u Logical link l pq Placed on the physical link ij superior, In time slot t, the service function chain F u Data arrives at the physical link e ij The rate on Is the physical link ij F u The rate of service provided; The deployment and migration cost of the business function chain is expressed as follows: in is the resource usage cost when the business function chain is deployed, It is the cost of migrating the business function chain; in, represents the unit computation cost on server node n in time slot t, represents the unit storage cost on server node n; in, is a binary variable, which determines whether the service function f uh Does it need to be migrated? Yes uh The placement position in time slot t-1, Yes uh The placement position in time slot t, is the shortest path between these two locations, using the business function f uh exist and The link transmission delay between uh Migration costs; Convert the receiving rate of the business function chain into a numerical reward, and receive a business function chain reward of r base , the reward corresponding to the receiving rate in the entire time slot T is as follows: R base =number·r base Where num is the number of times the business function chain has been successfully placed or migrated; (3) Construct and train a neural network model to simulate the deployment and migration of service function chains in T consecutive time slots. In each time slot t, the physical network topology information and service request information of the current time slot are collected as input. According to the overall optimization goal, the control center neural network is used to calculate and output the deployment or migration location of the service function chain and allocate deployment or migration resources. The service function chain request in the current time slot can only enter the next time slot after it is processed. (4) The control center manages the physical network topology to provide resource services for business requests and calculates the remaining resource information of the physical network topology. The entire deployment and migration process is completed only when all business function chains in T consecutive time slots are processed.

2. The method for deploying and migrating a mobile-aware multi-target service function chain according to claim 1, characterized in that: The physical network topology information described in step (2) includes the location information of server nodes, routing nodes, access nodes, computing resource capacity information, storage resource capacity information and related link bandwidth capacity information; the mobile terminal service request information includes the type of service request, the location information of the mobile terminal, the demand for computing resources, storage resources, bandwidth resources of the mobile terminal service request, and the limit requirements for end-to-end delay.

3. The method for deploying and migrating a multi-target service function chain based on mobile awareness according to claim 1 or 2, characterized in that: In step (2), Ω = {SFC1, SFC2, ..., SFC |U| } represents the service request information of |U| mobile terminals, each Respectively represent the service function chain request information of the u-th mobile terminal, including all the service functions of the service function chain Each business function uh There is a fixed demand for resources. represents f uh The demand for computing and storage resources, represents the maximum tolerable end-to-end communication delay of the service function chain, b u represents the required bandwidth of the service function chain, L u A represents the logical link of the service function chain. The ability of the terminal to which the service function chain is attached to transmit data during wireless access is A. u , the life cycle of the business function chain is TTL u , indicating that the occupied resources will be released after its life cycle ends. Indicates the position of the mobile terminal to which the service function chain is attached in the current time slot and in the next time slot. Except for the position feature which changes with the movement of the mobile terminal, other attributes are initially given values ​​throughout the process. The information of the physical network topology is expressed as G = (N∪E), where N represents the nodes of the physical network topology and E represents the links of the physical network topology. The nodes include server nodes, routing nodes, and access nodes. Each node has its own pre-planned location. In addition, the server nodes have available computing resources that can provide services. and available storage resources There is bandwidth available on each link As the physical network topology provides services to mobile terminals, the available resources on server nodes and physical links also change dynamically.

4. The method for deploying and migrating mobile-aware multi-target service function chains according to claim 1, characterized in that: The neural network described in step (3) sets weights for end-to-end delay and cost based on the overall optimization objective function. The structure and training process of the neural network are as follows: (31) Set a set of weights, w = (w1, w2), which represent the weights of end-to-end delay, deployment and migration costs respectively. The objective function is expressed as: maxR base -w1·D-w2·C; (32) The business function chain deployment and migration problem is divided into two sub-problems: business function chain deployment and migration. The two sub-problems are linked by the migration cost, and the Advantage-Actor-Critic algorithm is used to train the network. (33) According to (32), the objective function solution of the two sub-problems is further converted into T optimization sub-problems. These optimization sub-problems make decisions through neighbor-based strategies. During the training process, the training parameters of the previous neural network will be used as the initial parameters of the next neural network. After iterative learning, more business function chains are received, resulting in lower end-to-end latency, deployment cost, and migration cost. The following conditions need to be met during the iteration process: In the current time slot t, the service resources available to any server node should be higher than or equal to the resources used by all service function chains using the node. Similarly, the bandwidth resources available to any link should be higher than or equal to the bandwidth used by all service function chains using the link. The constraints are as follows: in, represents the computing resources available to server node n in time slot t, represents the storage resources available to server node n in time slot t; in, represents the physical link e in time slot t ij Available bandwidth; In the current time slot t, the delay of the service function chain that can be successfully deployed or migrated must be less than or equal to its maximum tolerable delay, and the relationship is as follows: Where D u (t) is the business function chain F u The end-to-end latency of current communications.

5. The method for deploying and migrating mobile-aware multi-target service function chains according to claim 1, characterized in that: In step (4), the control center obtains the deployment or migration position of the business function chain in each time slot, and needs to determine whether the current business function chain can be fully deployed or migrated, that is, whether all business functions of the business function chain have been provided with deployment resources or migration resources. If so, it sends instructions to the mobile terminal and the physical network topology to allow the two to interact, be served and provide services.

Images