CN107332913B - Optimized deployment method of service function chain in 5G mobile network - Google Patents

Abstract

The invention discloses an optimized deployment method of a service function chain in a 5G mobile network, and belongs to the field of mobile communication. When allocating the server resources and bandwidth resources of the underlying network to each service function chain request that arrives dynamically, the present invention proposes three deployment methods by considering the deployment methods of the VNF merge strategy, the virtual reuse strategy and the temporary link strategy during deployment. Different mapping schemes: Minimize the cost of computing resources: Limit the maximum mapping cost that each virtual link can be accepted by the service provider during deployment; Minimize the path length of the service function chain: Limit each virtual network function that can be served during deployment The maximum mapping cost accepted by the provider; compute resource cost and link resource cost are fairly optimized using a gender war game theory model. The three implementation schemes of the present invention can all achieve the minimum total mapping cost as much as possible while improving the mapping success rate of the service function chain request and the resource utilization rate of the underlying network.

Description

An optimized deployment method of service function chain in 5G mobile network

technical field

The invention belongs to the field of mobile communications, and in particular relates to the optimized deployment of service function chains in a 5G mobile network.

Background technique

With the substantial growth of wireless traffic and services, the fourth generation (4G) will not be able to meet future network needs, thus promoting the research of fifth generation (5G) mobile wireless networks to meet the great challenges of diversified and differentiated services in the future .

With the explosion of mobile wireless traffic, mobile operators are considering the use of cloud computing to expand their services and cope with the huge increase in mobile data traffic. In traditional telecom networks, network functions or middle boxes (such as packet data network gateways, service gateways, firewalls, content filters, proxy servers, WAN optimizers, intrusion detection systems, and intrusion prevention systems) are separated by some specialized physical devices and Facility realized. However, as users' demands for more diverse and new services continue to increase, service providers must accordingly purchase, store, and operate new physical equipment to meet user requirements. However, the purchase of new physical equipment will incur higher capex and opex. And these physical devices require specially trained personnel to deploy and maintain.

To address the aforementioned challenges, researchers have proposed Network Functions Virtualization (NFV), which aims to map the processing of packets from hardware middleboxes to software middleboxes running on commodity hardware. A network function that runs on a software middlebox is called a virtual network function (VNF). In network function virtualization, multiple virtual network functions are usually connected in a specific order to form a service function chain to provide different network services. For example, in a mobile network, the communication between mobile network users and service terminals needs to pass through the service function chain: user → service gateway → packet data network gateway → firewall → intrusion detection system → agent → terminal, this service function chain is generally deployed in A security policy for implementing traffic filtering between users and service terminals. Usually, the type and sequence of each virtual network function in the service function chain are determined according to the service classification, service level agreement and the provisioning policy of the operator.

In fact, core network virtualization and network functions virtualization express two key visions for future 5G architectures. As a key technology of 5G, network function virtualization has become an important direction in the evolution of wireless network architecture. As an emerging technology, network function virtualization has received extensive attention from industry, academia and standardization bodies. For service providers, effective deployment/mapping of service function chains to 5G mobile networks is critical. At present, the placement of virtual network functions has also become a research hotspot, and there have been some studies on the placement of virtual network functions.

There are currently proposed placement/deployment schemes for virtual network functions or service function chaining (SFC) requests, but most of the proposed deployment methods are not suitable for 5G mobile networks because the object of processing is virtual networks or federated clouds. For example, the Capacitated NFV Location algorithm, its main idea is to minimize the overall network cost when placing network functions, while satisfying the size constraints of network nodes. Although this method can realize the placement of virtual network functions, it is proposed for virtual networks or joint clouds without considering the characteristics and related constraints of 5G networks, so this method is not suitable for 5G mobile networks. For the placement/deployment solutions of virtual network functions or service function chains suitable for 5G mobile networks, only the deployment of virtual network functions (such as packet data network gateways and service gateways) of the wireless access network is considered, and there is no Consider the deployment of virtual network functions such as firewalls, content filters, proxy servers, WAN optimizers, intrusion detection systems, and intrusion prevention systems for core and data center networks. For example, the number of service gateways that need to be relocated and the length of the path can be compromised through the bidding Nash theory. Although the placement of virtual network functions in the 5G mobile network can be realized, only the deployment of virtual network functions of the wireless access network is considered. However, they do not consider the deployment of virtual network functions in core networks and data center networks.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to propose a known 5G mobile network (underlying network) and online service function chain request, the location of the mobile network user and the location of the service terminal. Each virtual network function and link connection in each service function chain request and meet the relevant constraints, with the goal of consuming the least server resources, bandwidth resources and reducing the blocking rate of the service function chain request, the service function chain request is processed. The deployment method for placement. The deployment scheme of the present invention comprehensively considers the particularity of the service function chain request. In addition to optimizing the configuration of common bandwidth resource requirements and server resource requirements, it also proposes a strict requirement for the communication delay of the service function chain request. corresponding solution strategies.

In the mapping process of the service function chain request, in order to effectively map/deploy the service function chain request to the 5G mobile network, the present invention introduces three effective strategies to reduce the cost of computing resources and link resources, thereby improving the service function chain request reception rate. .

(1) Virtual machine reuse strategy.

In the present invention, in order to improve the resource utilization rate of the server, a virtual machine reuse strategy is introduced, that is, when a virtual network function (VNF) is mapped, the existing/existing virtual machine that implements the same VNF can be reused. The virtual machine reuse strategy can not only improve the resource utilization of the server, but also improve the reception rate of service function chain requests when the server resources are limited. In order to improve the resource utilization of the server and reduce the cost of computing resources, when mapping a VNF to the server, if virtual machine reuse is considered, first determine whether there is a reusable virtual machine on the server where the current VNF is placed. The VNF is hosted on a reusable virtual machine on this server; otherwise, the current VNF is hosted on a new virtual machine on this server. The reusable virtual machines are: a virtual machine with a reuse identifier (indicating that the current virtual machine is allowed to be reused) that has hosted at least one VNF of the same type as VNF _i , and the managed VNF does not belong to the same VNF _i Service function chain request.

When the VNF is hosted on a reusable virtual machine, during the original service time of the virtual machine, the present invention does not consider the cost of computing resources, that is, the cost of computing resources is zero, but when the service time of the VNF exceeds the service time of the virtual machine The original service time, you need to calculate the computing resource cost for the additional time. When the VNF uses a new virtual machine, the present invention needs to calculate the computing resource cost for the entire service time of the VNF. Therefore, the calculation process of the computing resource cost of any VNF requested by a service function chain is as follows:

Cost(VNF _i →n _k )=p(n _k )×ε(VNF _i )×T _i ^p

Among them, Cost(VNF _i →n _k ) represents the computing resource cost of the i-th VNF (VNF _i ) being mapped to the server n _k , and p(n _k ) represents the resource unit cost of the server n _k , that is, the symbol used in the present invention p( ) represents the unit cost of the object in parentheses, and ε(VNF _i ) represents the resource constraints of VNF _i on the server, such as CPU, memory, and storage capacity. _{N V} = { ^VNF ₁ , _{VNF 2} , _. The paid time, Ti _{represents the service time of VNF i ,} T _o represents the original service time of the reused virtual machine, π _i represents the number of VNFs hosted on the virtual machine, π _i =1 means that the current virtual machine is not reused, π _i >1 means reused by multiple VNFs. Reusing existing virtual machines affects the performance of other VNFs, therefore, define δ as the maximum number of VNFs hosted on existing virtual machines, namely:

(2) VNF merge strategy.

In the present invention, when mapping VNF _i , if the server hosting VNF _i-1 has enough available resources, it can be considered to map VNF _i to the server hosting VNF _i-1 , the present invention calls this strategy as VNF merge strategy . In the present invention, some VNFs are not merged before mapping this service function chain request, but during the mapping process, the i-th VNF is mapped to the server hosting the i-1-th VNF, if the server has enough available resources. In the VNF merge strategy, the i-th VNF can only be mapped to the server hosting the i-1-th VNF, but it cannot be mapped to the server hosting other VNFs requested by this service function chain, so as to avoid the ping-pong routing problem . For example, when VNF ₁ requested by the service function chain is mapped to physical node B and VNF ₂ is mapped to physical node F, when mapping VNF ₃ , it can be mapped to physical node F, if physical node F has enough available resources, that is, allows VNF ₂ and VNF ₃ to be merged together, so that VNF ₂ and VNF ₃ communication does not need to consume bandwidth resources, because VNF ₂ and VNF ₃ communicate inside physical node F. However, mapping VNF ₃ to Physical Node B is not allowed as this would result in ping-pong routing, which is not desired.

(3) Temporary link mapping strategy.

In the present invention, when mapping the i-th VNF, the i-th link e _i connecting the i-th VNF and the i-1-th VNF needs to be mapped at the same time, (virtual link) to ensure that the i-th VNF obtains Nearly optimal mapping scheme. The traditional approach is devoted to finding the locally optimal mapping solution for the ith VNF, but this cannot guarantee a near-optimal path for the entire service function chain. In order to ensure an approximate optimal path and improve the acceptance rate of the entire service function chain request, the present invention generates a temporary link te _i connecting the ith VNF and the service terminal when calculating the link resource cost of the virtual link e _i (virtual link), the bandwidth requirement of this temporary link is equal to the bandwidth requirement of the i+1th link, and this temporary link te _i is mapped to the underlying network to obtain a mapping path. When mapping the i-th VNF and the virtual link e _i , a temporary link mapping strategy is used to find an approximate optimal mapping scheme for the i-th VNF. In the temporary link mapping strategy, the temporary link does not need to consume actual link resources, it is only used to constrain the i-th VNF not to deviate too far from the service terminal, so as to ensure that the adjacent links of the physical server hosting the i-th VNF have Sufficient link resources to map the next virtual link, thereby increasing the service function link acceptance rate. The link resource cost of the i-th virtual link is calculated as follows:

表示连接VNF_i和VNF_i-1的第i条虚拟链路(e_i)的链路资源成本，p_ei表示虚拟链路e_i在底层网络的映射路径(底层路径)，p_tei表示临时链路te_i在底层网络的的映射路径(底层路径)，e_s表示底层网络的物理链路，x_i表示虚拟链路e_i的资源约束，如带宽等，而x_i+1则表示连接第VNF_i+1和VNF_i的第i+1条虚拟链路的资源约束。本发明中，默认连接第一个VNF与用户的虚拟链路为虚拟链路e₁。对服务功能链路请求的映射可分为两个部分。第一部分是放置和分配资源给服务功能链请求的VNF。第二部分是映射和分配带宽资源给服务功能链请求的虚拟链路。服务功能链的映射过程描述如下。in,

Represents the link resource cost of the _ith virtual link (ei) connecting VNF _i and VNF _i-1 , p _ei represents the mapping path (underlay path) of the virtual link e _i in the underlying network, p _tei represents the temporary chain The mapping path (underlayer path) of path te _i in the underlying network, es represents the physical link of the underlying network, _{xi represents the resource constraints of the virtual link e i ,} such as bandwidth, etc., and _xi+1 represents the connection first Resource constraints of VNF _i+1 and the i+1th virtual link of VNF _i . In the present invention, the virtual link connecting the first VNF and the user is the virtual link e ₁ by default. The mapping of service function link requests can be divided into two parts. The first part is to place and allocate resources to the VNF serving the function chain request. The second part is to map and allocate bandwidth resources to virtual links that serve function chain requests. The mapping process of the service function chain is described as follows.

(1) VNF mapping:

The VNF mapping process can be expressed as:

表示VNF_i能被映射到这个网络区域，

表示VNF_i不能被映射到这个网络区域，

表示服务器M(VNF_i)满足VNF_i的位置约束；若

则不满足。在5G移动网络中，服务网关和分组数据网关属于无线接入网络的功能，他们通常只部署在无线接入网络中，而防火墙、内容过滤器、代理服务器、广域网优化器、入侵检测系统和入侵防御系统是数据中心网络或核心网络的功能，他们通常只部署在核心网络和数据中心网络中。where N ^S1 represents the set of servers and routers allocated to the underlying network requested by the current service function chain, C ^N1 represents the server resources allocated to the current service function chain request, MN = { _M (VNF ₁ ), M(VNF ₂ ) ,...,M(VNF _n )} represents the mapping record of each VNF requested by the current service function chain. M(VNF _i ) represents the server hosting the VNF _i , and R(M(VNF _i )) represents the available resources of the server M(VNF _i ). C _N = {ε(VNF ₁ ),ε(VNF ₂ ),...,ε(VNF _n )} represents the resource constraint set of all virtual network functions, VM _i represents the virtual machine (existing) ε that hosts VNF _i (VM _i ) represents the computing resources of the existing virtual machine, y∈{0,1,2,...,Y} represents the number of the network area, L(M(VNF _i )) represents the server M(VNF _i ) The number of the network area where it is located, and a server can only belong to one network area,

Indicates that VNF _i can be mapped to this network area,

Indicates that the VNF _i cannot be mapped to this network area,

Indicates that the server M (VNF _i ) satisfies the location constraint of VNF _i ; if

is not satisfied. In 5G mobile network, service gateway and packet data gateway belong to the function of radio access network, they are usually only deployed in radio access network, while firewall, content filter, proxy server, WAN optimizer, intrusion detection system and intrusion detection system Defense systems are functions of the data center network or core network, and they are usually deployed only in the core network and the data center network.

(2) Link mapping of service function chain:

The link mapping of the service function chain is described as follows:

表示当前服务功能链请求的虚拟链路集合，|E_v|表示集合E_V的元素个数，即当前服务功能链请求的虚拟链路数量。

表示当前服务功能链请求的所有虚拟链路的资源约束集合。P¹表示当前服务功能链请求所映射的端到端的底层路径集合，并且P¹的每条底层路径

是底层网络的物理链路集合E^S的一个子集合。C^E1表示分配给这个服务功能链请求的链路资源。

表示底层路径

的可用带宽资源，b(e_s)表示物理链路e_s的可用带宽资源，

表示底层路径

的路径时延，d(e_s)表示物理链路e_s的时延。Among them, M _E ={M(e ₁ ),M(e ₂ ),...,M(e _|Ev| )} represents the mapping record of each virtual link requested by the current service function chain,

represents the virtual link set requested by the current service function chain, |E _v | represents the number of elements in the set _EV , that is, the number of virtual links requested by the current service function chain.

Represents the set of resource constraints for all virtual links requested by the current service function chain. P ¹ represents the set of end-to-end underlying paths mapped by the current service function chain request, and each underlying path of P ¹

is a ^subset of the physical link set ES of the underlying network. C ^E1 represents the link resource allocated to this service function chain request.

Indicates the underlying path

The available bandwidth resources of , b( _es ) represents the available bandwidth resources of the physical link _es ,

Indicates the underlying path

The path delay of _d (es) represents the delay of the physical link _es .

Therefore, the deployment problem of each service function chain request in 5G mobile network, I) minimize the cost of link resources; II) minimize the cost of computing resources, which can be described according to the following linear programming (1):

s.t.

The purpose of the first goal is to reduce the cost of computing resources as much as possible. This will increase the probability that the service function chain will have a longer path. The purpose of the second goal is to minimize the cost of link resources, that is, to shorten the path of the entire service function chain as much as possible. At the same time, the constraints in linear programming (1) are used to ensure the following constraints:

Constraint 1 is used to ensure that the number of VNFs hosted on a virtual machine does not exceed the number δ given by the service provider.

Constraint 2 gives the time the VNF needs to pay.

Constraints 3 and 4 ensure that the servers being used meet the computing resource requirements of the VNF.

Constraints 5 and 6 ensure that the physical link used satisfies the constraints of the virtual link.

Constraints 7, 8, and 9 ensure that the server being used satisfies the VNF's location constraints.

Because the linear programming (1) is a multi-objective problem and cannot be solved directly, three solutions are proposed in the present invention to solve the multi-objective problem (1). The first solution proposed is: minimize the cost of computing resources; the second solution is: shorten the path of the entire service function chain; the third solution is: by utilizing the Battle of Sex Game (BOS) model for VNF Allocate resources and routes to find a fair solution for computational resource costs and link resource costs. The above three solutions are described in detail as follows:

(1) Minimize the cost of computing resources (MC scheme for short).

为服务功能链请求中每条虚拟链路能被服务提供商接受的最大映射成本，即

由服务提供商给定，其通常不超过每条虚拟链路的收费。这个优化模型以降低计算资源成本为目标，可描述为如下的线性规划(2)：In this solution, define

is the maximum mapping cost that can be accepted by the service provider for each virtual link in the service function chain request, namely

Given by the service provider, it usually does not exceed the charge per virtual link. This optimization model aims to reduce the cost of computing resources and can be described as the following linear programming (2):

s.t.

That is, when allocating the server resources and broadband resources of the underlying network to each dynamically incoming service function chain request, the optimal deployment scheme of the current service function chain request is obtained through the following steps:

Step 1: Starting from the first VNF of the virtual network function set N _V = {VNF ₁ , VNF ₂ ,..., VNF _n } requested by the service function chain to be mapped, sequentially for each of the current service function chain requests VNF _i (i=1,...,n) determines the set of alternative mapping schemes:

(1) Determine the candidate server set of the current VNF _i to be mapped:

中的各VNF选择过的服务器作为VNF_i的备选服务器集，即VNF_i可选的备选服务器包括：其他VNF未曾选择过的服务器、VNF_i-1所选择的服务器(VNF合并策略)，因此

也可以表示为：

From the set of available ^servers in the underlying network, US, will satisfy location constraints, computing resource requirements and have not been aggregated

The servers selected by each VNF in the VNF _i are used as the alternative server set of VNF _i , that is, the optional alternative servers of VNF i include: servers that have not been selected by other VNFs, servers selected by VNF _i-1 (VNF merging strategy), therefore

It can also be expressed as:

(2) Determine the set of alternative mapping schemes for VNF _i :

Use M(VNF _i ) to represent any candidate server where VNF _i is placed, and determine the virtual machine used to host VNF _i on server M(VNF _i ): determine whether there is a reusable virtual machine, and if so, host VNF _i on a reusable virtual machine; otherwise host the VNF _i on a new virtual machine;

其中VNF₀表示用户；Under the premise of satisfying the link resource requirements _xi and delay requirements of e _i (the virtual link connecting VNF _i-1 and VNF _i ), perform bottom-level path mapping on e _i to obtain the mapped path of e _i

Where VNF ₀ represents the user;

即通过临时链路映射策略来约束放置VNF_i的服务器不偏离用户终端太远；Generate a temporary link te _i connecting VNF _i and the user terminal, and use the link resource requirement x _i+1 of e _i+1 (the virtual link connecting VNF _i and VNF _i+1 ) as the link of te _i Resource requirements; on the premise of meeting the link resource requirements of te _i , perform bottom path mapping on te _i to obtain the mapping path of te _i

That is, the temporary link mapping strategy is used to constrain the server placing the VNF _i not to deviate too far from the user terminal;

映射路径

可能存在多条。where the mapping path

map path

There may be more than one.

的最短映射路径

作为对应当前M(VNF_i)的e_i的最终映射路径，从而保证对e_i的底层路径映射所对应的链路资源成本不会超过

will satisfy the condition

the shortest mapping path of

As the final mapping path of e _i corresponding to the current M(VNF _i ), so as to ensure that the link resource cost corresponding to the underlying path mapping to e _i will not exceed

Save the alternative mapping scheme corresponding to the current M(VNF _i ) into the set of alternative mapping schemes, where the alternative mapping scheme includes: M(VNF _i ), the virtual machine hosting the VNF _i , and the final mapping path of e _i ;

The combination of different alternative mapping schemes of all _VNFs requested by the current service function chain can obtain different mapping sets MN;

分别计算每个映射集合的总计算资源成本，由最小总计算资源成本对应的映射集合得到当前服务功能链请求的优化部署方案。Finally, according to the formula

The total computing resource cost of each mapping set is calculated separately, and the optimal deployment scheme for the current service function chain request is obtained from the mapping set corresponding to the minimum total computing resource cost.

(2) The path length of the service function chain is minimized (the SL scheme for short).

为服务功能链请求中每个VNF能被服务提供商接受的最大映射成本，即

由服务提供商给定，其通常不超过每个VNF的收费。这个优化模型旨在最短化服务功能链的路径长度，可描述为如下的线性规划(3)：In this solution, define

is the maximum mapping cost that each VNF in the service function chain request can be accepted by the service provider, namely

Given by the service provider, it usually does not exceed the charge per VNF. This optimization model aims to minimize the path length of the service function chain and can be described as the following linear programming (3):

s.t.

That is, when allocating the server resources and broadband resources of the underlying network to each dynamically incoming service function chain request, the optimal deployment scheme of the current service function chain request is obtained through the following steps:

Step 1: Starting from the first VNF of the virtual network function set N _V = {VNF ₁ , VNF ₂ ,..., VNF _n } requested by the service function chain to be mapped, sequentially for each of the current service function chain requests VNF _i (i=1,...,n) determines the set of alternative mapping schemes:

中的各VNF选择过的服务器作为VNF_i的初始备选服务器集U_i′；101: From the pool of available servers in the underlying network, will satisfy location constraints, computing resource requirements and have not been pooled

The server selected by each VNF in the VNF _i is used as the initial candidate server set U _i ′ of VNF i;

Determine the virtual machine used to host the VNF _i on each initial candidate server: determine whether there is a reusable virtual machine, if so, host the VNF _i on the reusable virtual machine; otherwise, host the VNF _i on a new virtual machine. on the virtual machine;

比较，将小于或等于

的初始备选服务器n_k作为VNF_i的备选服务器n_m，并记录备选服务器n_m上托管VNF_i的虚拟机；由所有备选服务器n_m得到VNF_i的备选服务器集U_i。For each initial candidate server n _k ∈ U _i ′, determine the virtual machine used to host VNF _i on it: determine whether there is a reusable virtual machine, and if so, host VNF _i on the reusable virtual machine ; otherwise host the VNF _i on a new virtual machine. and calculate its resource cost Cost(VNF _i →n _k ) with

Compare, will be less than or equal to

The initial candidate server n _k is used as the candidate server n _m of the VNF _{i, and the virtual machines hosting the VNF i on} the candidate server n _m are recorded; the candidate server set U _i of the VNF _i is obtained from all the candidate servers n _m .

102: Determine the set of alternative mapping schemes for VNF _i :

其中VNF₀表示用户，即对于同一个服务器n_m来说，VNF_i-1存在多少个备选服务器，则就存在多少条其关于e_i的映射路径

For each candidate server n _m ∈ U _i , on the premise of satisfying the link resource requirement x _i and the delay requirement of e _i , find the underlying path from server n _m to the candidate server of VNF _i-1 . A shortest path as the mapping path of e _i

Among them, VNF ₀ represents the user, that is, for the same server n _m , as many alternative servers exist in VNF _i-1 , there are as many mapping paths for e _i as there are.

Generate a temporary link te _i connecting the VNF _i and the user terminal, and take the link resource requirement x _i+1 of e _i+1 as the link resource requirement of te _i ; on the premise of satisfying the link resource requirement of te _i In the bottom path from the server n _m to the user terminal, find a shortest path as the mapping path of te _i

Save the alternative mapping scheme corresponding to the current n _m to the set of alternative mapping schemes of the VNF _i , where the alternative mapping scheme includes: n _m , the virtual machine hosting the VNF _i , the mapping path

Step 2: Obtain different mapping sets by combining different alternative mapping schemes of all VNFs requested by the current service function chain;

分别计算每个映射集合所对应的总链路资源成本，由最小总链路资源成本对应的映射集合得到当前服务功能链请求的优化部署方案。According to the formula

Calculate the total link resource cost corresponding to each mapping set separately, and obtain the optimal deployment scheme of the current service function chain request from the mapping set corresponding to the minimum total link resource cost.

(3) Fairly optimize computing resource cost and link resource cost using gender war game theory model (FOCL scheme for short).

The Gender Wars game theory model describes a game scenario in which two players have some common interests, but the common interests have different outcomes and have conflicting preferences. For example, the couple would rather watch the same TV show than watch their respective TV shows alone, and the couple also prefer to watch their favorite shows.

Therefore, in the present invention, computing resource cost and link resource cost are considered as two players in the gender war game theory model, which is based on two strategies: I) reuse existing virtual machines when mapping VNFs, II) when mapping VNFs when using a new virtual machine. An existing virtual machine is used to map the VNF, which can reduce the cost of computing resources, but it can lead to a longer path in the service function chain. A new virtual machine is used to map the VNF, it is easier to find a near-optimal path of the service function chain, but it may lead to higher cost of computing resources. Therefore, the mapping process of each VNF is a game process. The two players, the computing resource cost and the link resource cost, play a game to decide whether to use an existing virtual machine. In the present invention, an existing virtual machine will be reused as the first strategy (ie, 1-S), and a new virtual machine will be used as the second strategy (ie, 2-S), that is, the computing resource cost will be used as the first strategy. One player (ie 1-P), takes the link resource cost as the second player (ie 2-P). This game strategy is shown in Table 1.

Table 1 Game strategy

The expressions and parameters involved in Table 1 are annotated as follows:

当使用一个存在的虚拟机映射VNF_i时的计算资源收益；

Computational resource gain when mapping VNF _i with an existing virtual machine;

当使用一个存在的虚拟机映射VNF_i时的链路资源收益；

Link resource benefits when using an existing virtual machine to map VNF _i ;

Cost(ME _{(VNF i} ⁾ ): the cost of computing resources when using an existing virtual machine to map VNF _i ;

Cost(p ^E (e _i )): link resource cost when using an existing virtual machine to map VNF _i ;

M ^E (VNF _i ): the mapping scheme of VNF _i when mapping VNF _i with an existing virtual machine;

p ^E (e _i ): the mapping path of virtual link e _i when using an existing virtual machine to map VNF _i ;

当使用一个新虚拟机映射VNF_i时的计算资源收益；

Computational resource gain when mapping VNF _i with a new virtual machine;

当使用一个新虚拟机映射VNF_i时的链路资源收益；

Link resource gains when using a new virtual machine to map VNF _i ;

Cost(M ^N (VNF _i )): the cost of computing resources when mapping VNF _i with a new virtual machine;

Cost(p ^N (e _i )): link resource cost when using a new virtual machine to map VNF _i ;

MN ( ^VNF _i ): the mapping scheme of VNF _i when mapping VNF _i with a new virtual machine;

p ^N (ei ) _{: the mapping path of virtual link e i when} a new virtual machine is used to map VNF _i ;

where Cost(M ^E (VNF _i )), Cost(M ^N (VNF _i )), Cost(p ^E (e _i )) and Cost(p ^N (e _i )) are calculated according to the following formulas:

Cost(M ^λ (VNF _i ))=p(M ^λ (VNF _i ))×ε(VNF _i )×T _i ^p ,

Among them, the superscript λ∈{E,N}, p ^E (te _i ), p ^N (te _i ) represent the corresponding temporary link te _i when an existing virtual machine and a new virtual machine are used to map VNF _i , respectively. map path.

和

为了获得VNF_i的近似最优映射方案，在本发明中选择总收益最高的纯策略纳什均衡点为聚焦均衡点。聚焦均衡点是本发明中VNF_i的映射方案。与每个VNF的映射过程是一个重复博弈的过程。上述关于公平优化计算资源成本和链路资源成本，可以通过以下线性规划(4)描述：In this model, there are two pure-strategy Nash equilibria, namely

and

In order to obtain the approximate optimal mapping scheme of VNF _i , in the present invention, the pure-strategy Nash equilibrium point with the highest total return is selected as the focus equilibrium point. The focus equalization point is the mapping scheme of VNF _i in the present invention. The mapping process with each VNF is an iterative game process. The above-mentioned fair optimization computing resource cost and link resource cost can be described by the following linear programming (4):

s.t.

λ={E,N} (4)

表示托管VNF_i的虚拟机

的计算资源，上标λ用于区分虚拟机是已存在的，还是新的，其中E表示已存在，N表示新的，下同。

分别表示底层路径

的时延、可用带宽资源。in,

Represents the virtual machine hosting the VNF _i

The superscript λ is used to distinguish whether the virtual machine is existing or new, where E means existing and N means new, the same below.

respectively represent the underlying path

delay and available bandwidth resources.

Step 1: Starting from the first VNF of the virtual network function set N _V = {VNF ₁ , VNF ₂ ,..., VNF _n } requested by the service function chain to be mapped, sequentially for each of the current service function chain requests The final mapping scheme determined by VNF _i (i=1,...,n):

101: Determine the candidate server set of the VNF _i currently to be mapped

中的各VNF选择过的服务器作为VNF_i的备选服务器集

From the pool of available servers in the underlying network, will satisfy location constraints, computing resource requirements and have not been pooled

The servers selected by the _VNFs in the

中的每个服务器n_k，将服务器n_k上的一个新的虚拟机作为托管VNF_i的虚拟机；102: For alternate server sets

For each server n _k in , use a new virtual machine on server n _k as a virtual machine hosting VNF _i ;

其中M(VNF₀)表示用户端，即用户所在的物理节点；On the premise of satisfying the link resource requirements _xi and delay requirements of e _i , find a shortest path from the server where VNF _i-1 is placed (ie M(VNF _i-1 )) to the underlying path of server n _k map path as e _i

Among them, M(VNF ₀ ) represents the user terminal, that is, the physical node where the user is located;

Generate a temporary link te _i connecting the VNF _i and the user terminal, and take the link resource requirement x _i+1 of e _i+1 as the link resource requirement of te _i ; on the premise of satisfying the link resource requirement of te _i In the bottom path from the server n _k to the user terminal, find a shortest path as the mapping path of te _i

之和得到对应服务器n_k的总映射成本，将总映射成本最小的服务器n_k记为

By the computing resource cost of the corresponding server n _k CostVNF ^N (VNF _i →n _k ), the link resource cost

The sum gets the total mapping cost of the corresponding server n _k , and the server n _k with the smallest total mapping cost is recorded as

中的每个服务器n_j，计算服务器n_j的总映射成本，查找总映射成本最小的n_j并记为

103: For alternate server sets

For each server n _j in , calculate the total mapping cost of server n _j , find n _j with the smallest total mapping cost and record it as

where the total mapping cost of server n _j is calculated as:

Determine whether there is a reusable virtual machine on server n _j , if not, set the total mapping cost of server n _j to infinity;

其中VNF₀表示用户；以及生成一条连接VNF_i与用户终端的临时链路te_i，并将连接VNF_i与VNF_i+1的虚拟链路e_i+1的链路资源需求x_i+1作为te_i的链路资源需求；在满足te_i的链路资源需求的前提下，从服务器n_j到用户终端的底层路径中，查找一条最短路径作为te_i的映射路径

由服务器n_j的计算资源成本CostVNF^E(VNF_i→n_j)、链路资源成本

之和得到对应服务器n_j的总映射成本；If yes, host VNF _i on a reusable virtual machine, and under the premise of meeting the link resource requirements x _i and delay requirements of the virtual link e _i connecting VNF _i-1 and VNF _i , place the VNF from In the underlying path from the server of _i-1 to the server n _k , find a shortest path as the mapping path of e _i

Wherein VNF ₀ represents the user; and a temporary link te _i connecting VNF _i and the user terminal is generated, and the link resource requirement x _i+ 1 of the virtual link e _i+1 connecting VNF _i and VNF _i+1 is taken as The link resource requirements of te _i ; on the premise that the link resource requirements of te _i are met, find a shortest path in the underlying path from the server n _j to the user terminal as the mapping path of te _i

By the computing resource cost of server n _j CostVNF ^E (VNF _i →n _j ), link resource cost

The sum gets the total mapping cost of the corresponding server n _j ;

中总映射成本最小的服务器作为放置VNF_i的服务器M(VNF_i)，基于服务器M(VNF_i)确定VNF_i的最终映射方案，包括服务器M(VNF_i)、服务器M(VNF_i)上的托管VNF_i的虚拟机、对应服务器M(VNF_i)的e_i的映射路径；104: will

The server with the smallest total mapping cost is used as the server M (VNF _i ) where the VNF _i is placed, and the final mapping scheme of the VNF _i is determined based on the server M (VNF _i ), including the server M (VNF _i ) and the server M (VNF _i ) on the server M (VNF i ). The virtual machine hosting VNF _i and the mapping path of e _i corresponding to server M (VNF _i );

为定值，所以对

求最大则可以直接取转换为取Cost(M^λ(VNF_i))与

之和最小的映射方案，其中λ∈{N,E}。because

is a fixed value, so

To find the maximum, it can be directly converted to take Cost(M ^λ (VNF _i )) and

Sum-minimum mapping scheme, where λ∈{N,E}.

Step 2: Obtain the optimized deployment scheme requested by the current service function chain from the final mapping scheme of all VNFs requested by the current service function chain.

To sum up, due to the adoption of the above-mentioned technical solutions, the beneficial effects of the present invention are:

(1) Wide range of applications. Most of the traditional virtual network function or service function chain mapping algorithms are proposed for virtual networks and data center networks, or do not consider the complete service function chain in a complete 5G network. The method proposed in the present invention can be applied to the complete service function chain request in the 5G network, so compared with the traditional mapping algorithm, the present invention has a wider scope of application.

(2) The mapping cost is low. Since the present invention proposes three schemes for the multi-objective problem of linear programming (1), the deployment of low mapping cost for service function chain requests is obtained on the basis of combining VNF merging strategy, virtual machine reuse strategy and temporary link mapping strategy Schemes, especially the third scheme, comprehensively consider the cost of computing resources and link resources, and the cost of the mapping scheme is lower.

(3) The resource utilization rate is high. Since the virtual machine reuse strategy, the VNF merge strategy and the temporary link mapping strategy used in the present invention can reduce the consumption of resources, the utilization rate of the resources can be improved.

(4) The mapping blocking rate is small. Since the virtual machine reuse strategy, the VNF merging strategy and the temporary link mapping strategy used in the present invention can reduce resource consumption, the probability of successful mapping is greater and the blocking rate is smaller. .

Description of drawings

Figure 1 is a schematic diagram of a service function chain request, wherein the numbers in the rectangular box above the virtual network function represent server resource requirements, and the numbers above the virtual link represent virtual link resource requirements and delays;

Figure 2 is a schematic diagram and a comparison diagram of a temporary link mapping strategy, wherein Figure 2-a is a mapping scheme that does not consider the temporary link mapping strategy, and Figures 2-b and 2-c are mapping schemes that consider the temporary link mapping strategy , A to G in the figure represent different physical nodes (ie servers), the dashed line represents the mapping path of the virtual link, and the dashed line represents the mapping path of the temporary link.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

Taking the SDN-based 5G network as the implementation object, the network operator can deploy the optimized deployment scheme (method of mapping service function chain) proposed by the present invention on the control layer in the SDN control router, and the SDN control router can schedule its own bandwidth. Some control and management functions collect network-wide information, obtain resource information of all nodes in the network, and information such as link resources and delays. Through this centralized control method, the router can obtain the topology of the entire network and corresponding resource information. When a service function chain request arrives, the SDN control router can schedule the service function chain-based mapping method deployed on its control layer according to the network-wide information it has, and calculate key parameters such as mapping cost and rejection rate. feedback to the operator.

When the SDN control router deploying the FOCL scheme of the present invention performs online mapping processing of multiple service function chain requests, first, the set of outgoing service function chain requests is defined as ExpiredSFC, and the arrival queue of service function chain requests is defined as ArrivedSFC. In the queue ArrivedSFC, each service function chain request is mapped one after the other. Define the service function chain request set blocked due to insufficient underlying network resources as SFC _blo (referred to as the blocked mapping request set SFC _blo ). The online processing process for multiple service function chain requests is as follows:

enter:

1. The bottom layer network resources, including the bottom layer network G ^S =(N ^S , ^{E S} ), the resource constraints of the bottom layer network SC=(CE , ^CN ,L ^N ), and the set P of all end-to-end bottom layers. Among them, N ^S represents the server set and router set of the underlying network, ^ES represents the physical link set of the underlying network, C ^E represents the attributes of the physical link, such as bandwidth, delay, unit cost, etc., and ^CN represents the server of the underlying network. and router attributes, where the server attributes include the unit cost of server resources, server resources (such as CPU, memory, and storage capacity), etc. The router attributes mainly refer to the processing capacity of the router, and L ^N represents the location of the underlying network server and router;

2. An arrived service function chain request queue ArrivedSFC, wherein each service function chain request includes its virtual network (virtual network function set N _V , virtual link set E _V ) and placement constraints (resource constraint set _CN of each VNF) , the virtual link resource constraint set CE, the virtual link maximum delay constraint set _CD , the _VNF placement location constraint set _LN , the user's location _LU , and the serving terminal's location _LT ).

和被阻塞的映射请求集合SFC_blo。output: mapping cost

and the blocked map request collection SFC _blo .

Step 1: Initial

就执行步骤3；否则，转到步骤10。Step 2: If

Go to step 3; otherwise, go to step 10.

就更新底层网络资源，使

否则，转步骤4。Step 3: If

update the underlying network resources so that

Otherwise, go to step 4.

Step 4: Take out the service function chain request SFC _k of the head of the team from the ArrivedVDC, where the subscript k is the identifier of the service function chain request.

Step 5: Invoke the SFCM algorithm to map SFC _k .

就执行步骤7；否则，转步骤8。Step 6: If a mapping scheme for SFC _k is found

Go to step 7; otherwise, go to step 8.

并更新底层网络资源，然后转步骤9，其中

表示映射方案

的映射代价，即SFC_k的所有VNF的总映射成本(计算资源成本+链路资源成本)之和。Step 7: Order

and update the underlying network resources, then go to step 9, where

Representation mapping scheme

The mapping cost is the sum of the total mapping cost (computing resource cost + link resource cost) of all VNFs of SFC _k .

Step 8: Update SFC _blo = SFC _blo ∪ {SFC _k }.

Step 9: Update ArrivedSFC=ArrivedSFC-SFC _k , then go to step 2 .

SFC_blo。Step 10: Go Back

SFC _blo .

这个映射方案

包括映射成本、VNF的映射记录(放置的服务器、托管的虚拟机)、和服务功能链链路的映射记录(连接当前VNF和上一个VNF的虚拟链路的映射路径)。The SFCM algorithm involved in step 5 above is used to find a mapping scheme for each VNF in a service function chain request, find the mapping path of the service function chain request, and allocate resources to each VNF and each virtual chain road. The SFCM algorithm finds a mapping scheme for each VNF using an existing virtual machine and a new virtual machine, and then decides the final mapping scheme through a gender war game model. The SFCM algorithm finds the mapping scheme with the smallest mapping cost as the final mapping scheme

this mapping scheme

Including mapping cost, mapping record of VNF (placed server, hosted virtual machine), and mapping record of service function chain link (mapping path of virtual link connecting current VNF and previous VNF).

enter:

1. The bottom layer network G ^S =(N ^S , ^{E S} ), the resource constraints of the bottom layer network SC =(CE , ^CN ,L ^N ), and the set P of all end-to-end bottom layers.

2. A service function chain requests G _V = (N _V , E _V ) and placement constraints PC = (C _N , _CE , _CD , L _N , L _U , L _T ), as shown in FIG. 1 .

output: mapping scheme

Step 501: Store all available ^servers in the US.

Step 502: Starting from the first VNF, traverse each _{VNF i} in the NV in turn, and execute the next step. If each _VNF in the NV has been traversed, go to step 512;

Step 503: Traverse each server n _k ^∈ US in the underlying network, and execute the next step. If the servers in the underlying network have been traversed, go to Step 507 ;

且n_k未被这个服务功能链请求中的其他VNF使用或满足VNF合并策略，执行步骤505至步骤506；Step 504: If

And n _k is not used by other VNFs in the service function chain request or satisfies the VNF merging policy, execute steps 505 to 506;

Step 505: Map the current VNF _i to a new virtual machine on the server n _k , and calculate and record the CostVNF ^N (VNF _i →n _k ) according to formula (5);

Step 506: Based on the server n _k where the current VNF _i is placed, generate a temporary link te i from n _k to the server where the user terminal is located (known from the location _LT of the service terminal), the virtual link te _i of the temporary link te _i . The link resource constraint is x _i+1 ;

和

计算和记录

根据公式(6)，计算和记录VNF_i的总映射成本TCostVNF^N(VNF_i→n_k)根据公式(9)，再转至步骤50 3；Then use Dijkstra's algorithm (shortest path algorithm), based on the underlying path set P, find the link e _i (the virtual link connecting VNF _i and VNF _i-1 ) and the temporary link te under the condition of satisfying the bandwidth and delay constraints the shortest underlying path of _i

and

Calculate and record

According to formula (6), calculate and record the total mapping cost TCostVNF ^N of VNF _i (VNF _i →n _k ) according to formula (9), then go to step 503;

Step 507: Traverse each server n _j ^∈ US in the underlying network, and execute Step 508 , if the servers in the underlying network have been traversed, go to Step 511 ;

和n_j未被这个服务功能链请求中的其他VNF使用或满足VNF合并策略，执行步骤509至步骤510；Step 508: If

and n _j are not used by other VNFs in the service function chain request or satisfy the VNF merging policy, perform steps 509 to 510;

Step 509: Determine whether there is a reusable virtual machine on the server n _j , if not, set the total mapping cost of the server n _j to infinity, and go to step 507; if so, go to step 510;

Step 510: Host VNF _i on a reusable virtual machine, and calculate and record CostVNF ^E (VNF _i →n _j ) according to formula (7);

At the same time, based on the server n _j where the current VNF _i is placed, a temporary link te _i from n _j to the server where the user terminal is located (known from the location _LT of the service terminal) is generated. The virtual link of the temporary link te _i The road resource constraint is x _i+1 ;

和

计算和记录

根据公式(8)，计算和记录VNF_i的总映射成本TCostVNF^E(VNF_i→n_j)根据公式(10)，转至步骤507；Then use Dijkstra's algorithm to find the shortest underlying path of link e _i and temporary link te _i

and

Calculate and record

According to formula (8), calculate and record the total mapping cost TCostVNF ^E of VNF _i (VNF _i →n _j ) According to formula (10), go to step 507;

返回步骤502，其中

的初始值为空集；Step 511: Find a mapping scheme with the smallest total mapping cost TCostVNF ^N (VNF _i →n _k ) among all mapping schemes that use the new virtual machine to map VNF _i ; find a mapping scheme that uses the existing virtual machine to map VNF _i A mapping scheme with the minimum total mapping cost TCostVNF ^E (VNF _i →n _j ); find the scheme with the minimum mapping cost from these two schemes according to formula (11) as the final mapping scheme M ^* (VNF _i ) of VNF _i , and Update the mapping scheme for the current service function chain request

Returning to step 502, where

The initial value of is the empty set;

Step 512: Return

When a new virtual machine on server n _k is used to map VNF _i , the mapping cost of VNF _i can be calculated according to formula (5), and the mapping cost of virtual link e _i can be calculated according to formula (6):

CostVNF ^N (VNF _i →n _k )=P(n _k )ε(VNF _i )T _i (5)

When using an existing virtual machine on server n _j to map VNF _i , the mapping cost of VNF _i can be calculated according to formula (7), and the mapping cost of virtual link e _i can be calculated according to formula (8):

CostVNF ^E (VNF _i →n _j )=P(n _j )ε(VNF _i )max{T _i -T _o ,0} (7)

When mapping VNF _i with a new virtual machine on server n _k , the minimum total mapping cost can be calculated according to equation (9):

When mapping VNF _i using an existing virtual machine on server n _j , the minimum total mapping cost can be calculated according to formula (10):

The minimum total mapping cost of VNF _i can be calculated according to formula (11):

TCostVNF(VNF _i →n _m )=min{TCostVNF ^N (VNF _i →n _k ),TCostVNF ^E (VNF _i →n _j )} (11)

Among them, n _m is n _k or n _j . If the calculation results of TCostVNF ^N (VNF _i →n _k ) and TCostVNF ^E (VNF _i →n _j ) are the same, either one is selected.

Referring to Figure 2, in the service function chain request including 2 VNFs, the user (User) is placed on the physical node A, and the service terminal (Terminal) is placed on the physical node G. When placing VNF ₁ , it is assumed that the The selected physical nodes are B and D. If only the local optimal mapping is considered, the two alternatives are satisfied. When physical node B is selected (Figure 2-a), it will cause VNF ₁ to be far away from the service terminal. Therefore, this The invention constrains the VNF ₁ not to deviate too far from the service terminal through the temporary link te _i . When calculating the link resource cost between the VNF ₁ and the user, the mapping path _of the temporary link te _i is also taken into account. In terms of link resource cost, it is: T ₁ ×((unit cost of physical path A→D)×x ₁ +(unit cost of physical path D→G)×x ₂ ). Therefore, under the condition of the same unit cost, the link resource cost of selecting physical node D (Fig. 2-b) is smaller. Then combined with the VNF merging strategy and the virtual reuse strategy, the physical node of VNF ₂ is determined by considering the rule with the smallest total mapping cost, so as to obtain the mapping scheme of the service function chain request, as shown in Figure 2-c, the corresponding underlying path is : A→D→E→G. If the temporary link te _i is not passed, the underlying path as shown in Figure 2-a may be obtained, that is, A→B→C→F→I→H→G, which will significantly increase the link resource cost.

Therefore, in the present invention, when allocating the server resources and bandwidth resources of the underlying network to each service function chain request that arrives dynamically, by considering the deployment methods of the VNF merge strategy, the virtual reuse strategy and the temporary link strategy during deployment, it achieves While improving the mapping success rate of service function chain requests and the resource utilization of the underlying network, the total mapping cost is minimized as much as possible.

The above descriptions are only specific embodiments of the present invention, and any feature disclosed in this specification, unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes; all the disclosed features, or All steps in a method or process, except mutually exclusive features and/or steps, may be combined in any way.

Claims (3)

1. an optimized deployment method of service function chain in a 5G mobile network, is characterized in that, comprises the following steps: Step 1: Starting from the first VNF of the virtual network function set N _V = {VNF ₁ , VNF ₂ ,..., VNF _n } requested by the service function chain to be mapped, sequentially for each of the current service function chain requests VNF _i determines the set of alternative mapping schemes, where i=1,...,n, n represents the number of VNFs requested by the current service function chain: 101: Determine the candidate server set of the VNF _i currently to be mapped: From the pool of available servers in the underlying network, will satisfy location constraints, computing resource requirements and have not been pooled

The servers selected by each VNF in the VNF are used as the alternative server set of VNF _i , that is, the optional alternative servers of VNF _i include: servers that have not been selected by VNF ₀ to VNF _i-2 , servers selected by VNF _i-1 , Where VNF ₀ represents the user; 102: Determine the set of alternative mapping schemes for VNF _i : Let M(VNF _i ) represent any candidate server where VNF _i is placed, and M(VNF ₀ ) represent the user terminal; Determine the virtual machine used to host the VNF _i on the server M (VNF _i ): determine whether there is a reusable virtual machine, if so, host the VNF _i on the reusable virtual machine; otherwise, host the VNF _i on a new virtual machine On a virtual machine; the reusable virtual machine is: a virtual machine with a reuse identifier that has hosted at least one VNF of the same type as VNF _i , and the hosted VNF and VNF _i do not belong to the same service function chain request ; Under the premise of meeting the link resource requirements _xi and delay requirements of the virtual link e _i connecting VNF _i-1 and VNF _i , perform the underlying path mapping on e _i to obtain the mapping path of e _i

Generate a temporary link te _i connecting VNF _i and the user terminal, and use the link resource requirement x _i+ 1 of the virtual link e _i+1 connecting VNF _i and VNF _i+1 as the link resource requirement of te _i ; Under the premise of meeting the link resource requirements of te _i , perform bottom path mapping on te _i to obtain the mapping path of te _i

will satisfy the condition

the shortest mapping path of

As the final mapping path of e _i corresponding to the current M( _{VNF i} ), where es represents the physical link of the underlying network, the symbol p( ) represents the unit cost of the object in brackets, T _i represents the service time of VNF _i ,

represents the preset maximum mapping cost of e _i ; Save the alternative mapping scheme corresponding to the current M(VNF _i ) into the set of alternative mapping schemes, where the alternative mapping scheme includes: M(VNF _i ), the virtual machine hosting the VNF _i , and the final mapping path of e _i ; Step 2: Obtain different mapping sets by combining different alternative mapping schemes of all VNFs requested by the current service function chain; According to the formula

Calculate the total computing resource cost of each mapping set separately, and obtain the optimal deployment scheme for the current service function chain request from the mapping set corresponding to the minimum total computing resource cost, where ε(VNF _i ) represents the computing resource requirement of VNF _i ; The payment time T _i ^p of VNF _i is: if the number of VNFs hosted on the virtual machine where VNF _i is located is equal to 1, then T _i ^p =T _i ; if the number of VNFs hosted on the virtual machine where VNF _i is located is greater than 1 , then T _i ^p =max{T _i -T _o ,0}, where T _o represents the original service time of the virtual machine hosting the VNF _i . 2. The method of claim 1, wherein steps 101-102 and step 2 are replaced by: 101: From the pool of available servers in the underlying network, will satisfy location constraints, computing resource requirements and have not been pooled

The server selected by each VNF in the VNF _i is used as the initial candidate server set of VNF i; Determine the virtual machine used to host the VNF _i on each initial candidate server: determine whether there is a reusable virtual machine, if so, host the VNF _i on the reusable virtual machine; otherwise, host the VNF _i on a new virtual machine. The reusable virtual machine is: a virtual machine with a reuse identifier that has hosted at least one VNF of the same type as VNF _i , and the hosted VNF and VNF _i do not belong to the same service function chain ask; The computing resource cost of each initial candidate server corresponding to VNF _i and the preset maximum mapping cost of VNF _i

Compare, will be less than or equal to

The initial alternative server is used as the alternative server of VNF _i , and saves the virtual machine hosting VNF _i on the alternative server; The computing resource cost of each initial candidate server corresponding to VNF _i is: the unit cost of the initial candidate server×the computing resource requirement of the VNF _i ×the payment time T _i ^p of the VNF _i on the initial candidate server; The payment time T _i ^p is: if the number of VNFs hosted on the virtual machine where VNF _i is located is equal to 1, then T _i ^p =T _i ; if the number of VNFs hosted on the virtual machine where VNF _i is located is greater than 1, then T _i ^p =max{T _i -T _o ,0}, where T _i represents the service time of VNF _i , and T _o represents the original service time of the virtual machine hosting VNF _i ; 102: Determine the set of alternative mapping schemes for VNF _i : Let M(VNF _i ) represent any candidate server for placing VNF _i . On the premise of meeting the link resource requirements _xi and delay requirements of the virtual link e _i connecting VNF _i-1 and VNF _i , the slave server M In the underlying path from (VNF _i-1 ) to M(VNF _i ), find a shortest path as the mapping path of e _i

Wherein M(VNF ₀ ) represents the user terminal; Generate a temporary link te _i connecting VNF _i and the user terminal, and use the link resource requirement x _i+ 1 of the virtual link e _i+1 connecting VNF _i and VNF _i+1 as the link resource requirement of te _i ; Under the premise of satisfying the link resource requirements of te _i , from the bottom path from the server M (VNF _i ) to the user terminal, find a shortest path as the mapping path of te _i

Save the alternative mapping scheme corresponding to the current M(VNF _i ) into the set of alternative mapping schemes, where the alternative mapping scheme includes: M(VNF _i ), the virtual machine hosting the VNF _i , the mapping path

Step 2: Obtain different mapping sets by combining different alternative mapping schemes of all VNFs requested by the current service function chain; According to the formula

Calculate the total link resource cost corresponding to each mapping set separately, and obtain the optimal deployment plan of the current service function chain request from the mapping set corresponding to the minimum total link resource cost, where E _V represents the virtual link requested by the current service function chain Set, _es represents the physical link of the underlying network. 3. A method for optimizing deployment of service function chains in a 5G mobile network, comprising the following steps: Step 1: Starting from the first VNF of the virtual network function set N _V = {VNF ₁ , VNF ₂ ,..., VNF _n } requested by the service function chain to be mapped, sequentially for each of the current service function chain requests The final mapping scheme determined by VNF _i , where i=1,...,n, n represents the number of VNFs requested by the current service function chain: 101: Determine the candidate server set of the VNF _i currently to be mapped

From the pool of available servers in the underlying network, will satisfy location constraints, computing resource requirements and have not been pooled

The servers selected by the _VNFs in the

102: For alternate server sets

For each server n _k in , use a new virtual machine on server n _k as a virtual machine hosting VNF _i ; On the premise of meeting the link resource requirements _xi and delay requirements of the virtual link e _i connecting VNF _i-1 and VNF _i , find a path from the server where VNF _i-1 is placed to the server n _k The shortest path as the mapped path of e _i

Where VNF ₀ represents the user; Generate a temporary link te _i connecting VNF _i and the user terminal, and use the link resource requirement x _i+ 1 of the virtual link e _i+1 connecting VNF _i and VNF _i+1 as the link resource requirement of te _i ; On the premise of meeting the link resource requirements of te _i , find a shortest path in the bottom path from the server n _k to the user terminal as the mapping path of te _i

By the computing resource cost of the corresponding server n _k CostVNF ^N (VNF _i →n _k ), the link resource cost

The sum gets the total mapping cost of the corresponding server n _k , and the server n _k with the smallest total mapping cost is recorded as

where CostVNF ^N (VNF _i →n _k )=P(n _k )ε(VNF _i )T _i ,

The symbol p( ) represents the unit cost of the object in parentheses, ε(VNF _i ) represents the computing resource requirement of VNF _i , Ti represents the service time of _{VNF i} , and es _represents the physical link of the underlying network; 103: For alternate server sets

For each server n _j in , calculate the total mapping cost of server n _j , find n _j with the smallest total mapping cost and record it as

where the total mapping cost of server n _j is calculated as: Determine whether there is a reusable virtual machine on server n _j , if not, set the total mapping cost of server n _j to infinity; the reusable virtual machine is: a virtual machine with a reuse identifier that has hosted at least one and VNF _i is a virtual machine of the same type of VNF, and the managed VNF and VNF _i do not belong to the same service function chain request; If so, host VNF _i on a reusable virtual machine, and under the premise of meeting the link resource requirements x _i and delay requirements of the virtual link e _i connecting VNF _i-1 and VNF _i , place the VNF from In the underlying path from the server of _i-1 to the server n _k , find a shortest path as the mapping path of e _i

where VNF ₀ represents the user; a temporary link te _i connecting VNF _i and the user terminal is generated, and the link resource requirement x _i+ 1 of the virtual link e _i+1 connecting VNF _i and VNF _i+1 is taken as te The link resource requirements of _i ; on the premise that the link resource requirements of te _i are met, find a shortest path from the bottom path from the server n _j to the user terminal as the mapping path of te _i

By the computing resource cost of server n _j CostVNF ^E (VNF _i →n _j ), link resource cost

The sum gets the total mapping cost of the corresponding server n _j ; where CostVNF ^E (VNF _i →n _j )=P(n _j )ε(VNF _i )max{T _i -T _o ,0},

es _represents the physical link of the underlying network, and _To represents the raw service time of the virtual machine hosting the VNF _i ; 104: will

The server with the smallest total mapping cost is used as the server M (VNF _i ) where the VNF _i is placed, and the final mapping scheme of the VNF _i is determined based on the server M (VNF _i ), including the server M (VNF _i ) and the server M (VNF _i ) on the server M (VNF i ). The virtual machine hosting VNF _i and the mapping path of e _i corresponding to server M (VNF _i ); Step 2: Obtain the optimized deployment scheme requested by the current service function chain from the final mapping scheme of all VNFs requested by the current service function chain.

Images