CN116208567A - Method and system for flow scheduling of SDN network resources of cross-domain data center - Google Patents

Abstract

The present invention relates to the field of network communication technology, and in particular to a method and system for traffic scheduling of SDN network resources in a cross-domain data center. The method includes: adding the computing capability of the cross-domain data center and the network performance and energy consumption of the edge server Optimize the model NSGA‑II algorithm, and establish corresponding objective functions for the computing power of the cross-domain data center and the network performance and energy consumption of the edge server, and obtain the optimization target model of traffic scheduling; respectively establish for each SDN network characteristic of SDN network resources The corresponding objective function and constraint conditions are added to the traffic scheduling optimization target model; the non-dominated genetic algorithm with elite strategy is used to solve the traffic scheduling optimization mathematical model, and the traffic scheduling scheme for traffic scheduling of SDN network resources is obtained , realize the timely adjustment technology of network resources adapted to the dynamic network traffic, and save the traffic bandwidth cost inside the data center.

Description

Method and system for traffic scheduling of cross-domain data center SDN network resources

technical field

The present invention relates to the technical field of network communication, and in particular to a method and system for traffic scheduling of cross-domain data center SDN network resources.

Background technique

A cross-domain SDN network resource unified management and agile scheduling theory for computing and network integration is proposed. First of all, based on the differences in business characteristics and service quality requirements among different computing power businesses, a differentiated service system and evaluation method are established in accordance with the service principles of moderate on-demand and global optimization, and an efficient scheduling system for computing and network integration resources. The design and optimization of the system provide a theoretical basis.

Specifically, in the case of meeting strict delay requirements, collaborative scheduling is performed on multiple types of resources such as computing resources, storage resources, and transmission link communication resources in the entire network to provide strict service quality assurance for computing-network integration services, and realize Optimal allocation of multi-dimensional resources across the network. Then, taking time delay and energy consumption as performance indicators, by establishing different objective functions, taking edge node computing power, cloud computing node computing power, computing task allocation ratio, user allocation scheduling strategy, differentiated delay, etc. as constraint objectives, a mathematical Model, the reasonable allocation of network resources and computing resources, to ensure the optimization of computing task delay and system energy consumption under constraints such as limited resources and task priority.

Scheduling of network resources has always been a fundamental and key issue in Internet research. The emergence of SDN has brought new opportunities for research in this area. For traffic scheduling in the SDN environment, Cui et al. from Cornell University in the United States proposed a scalable multicast solution, namely dual-structure multicast, based on the analysis of the heterogeneity of multicast traffic in SDN data centers. A SDN resource optimization scheme applied to data centers is proposed. Experimental results show that, compared with the traditional scheme, this scheme can increase the link efficiency by 12%, and reduce the packet loss rate by 51%. The group members and related management information are controlled by the SDN controller, and a control strategy based on a greedy algorithm is proposed to dynamically allocate optical signal network resources. An Equal Cost Multiple Routing Scheme (ECMP) under a single controller is proposed, which optimizes the overall topology from the perspective of global network resources in the same network domain, thereby improving the utilization of network resources and reducing the packet loss rate. It is pointed out that the current network resource multi-path optimization algorithm can only be implemented for a fixed topology inside a data center, and it is difficult to meet the generalization requirements. Therefore, this document proposes an improved ant colony Down-aware topology status to meet the general requirements of multi-path optimization algorithms. Aiming at multi-path routing orchestration, some researchers are also trying to study, and proposed hierarchical multi-path orchestration optimization to improve network performance; another method uses congestion sensing and automatic identification technologies to optimize routing and improve resource utilization. . In 2019, multi-path adaptive routing algorithms for cross-domain distributed data centers also began to attract the attention of researchers. It can be seen that in terms of unified network resource scheduling, the unified scheduling theory of software-defined networks combined with future cloud-edge fusion cross-domain data centers is still in the initial stage of research. At present, there are mainly the following problems:

1) Facing the future cross-domain data center environment of cloud-edge integration, how to combine SDN network characteristic indicators to establish a software-defined network resource optimization objective function and adjust and plan corresponding network traffic and resource strategies is one of the technical problems solved by this patent one.

2) In the cross-domain data center environment oriented to cloud-edge integration, how to combine the computing power and storage capacity of sub-data centers and edge computing nodes to achieve low-latency, low-load global traffic scheduling based on global network conditions is also an issue of this topic. Another technical problem solved by the patent.

Contents of the invention

The technical problem to be solved by the present invention is to provide a method and system for traffic scheduling of cross-domain data center SDN network resources in view of the deficiencies of the prior art.

The technical scheme of a method for traffic scheduling of cross-domain data center SDN network resources of the present invention is as follows:

The computing power of the cross-domain data center and the network performance and energy consumption of the edge server are added to the multi-objective optimization model NSGA-II algorithm, and the computing power of the cross-domain data center and the network performance and energy consumption of the edge server are respectively established The corresponding objective function is used to obtain the objective model of traffic scheduling optimization;

Establishing corresponding objective functions and constraint conditions for each SDN network feature of the SDN network resources respectively, and adding them to the traffic scheduling optimization target model to obtain a traffic scheduling optimization mathematical model;

The non-dominated genetic algorithm with elitist strategy is used to solve the traffic scheduling optimization mathematical model to obtain a traffic scheduling scheme for traffic scheduling of the SDN network resources.

The technical scheme of a system for traffic scheduling of cross-domain data center SDN network resources of the present invention is as follows:

Including a first addition building module, a second addition building module and a solving module;

The first addition building module is used to: add the computing power of the cross-domain data center and the network performance energy consumption of the edge server to the multi-objective optimization model NSGA-II algorithm, and respectively calculate the computing power of the cross-domain data center and The network performance and energy consumption of the edge server establishes a corresponding objective function to obtain a traffic scheduling optimization objective model;

The second adding and establishing module is used to: respectively establish corresponding objective functions and constraint conditions for each SDN network characteristic of the SDN network resource, and add them to the traffic scheduling optimization target model to obtain a traffic scheduling optimization mathematical model;

The solving module is used to solve the traffic scheduling optimization mathematical model by using a non-dominated genetic algorithm with an elitist strategy, and obtain a traffic scheduling scheme for traffic scheduling of the SDN network resources.

The beneficial effects of the present invention are as follows:

In the future-oriented cross-domain data center environment of cloud-edge integration, it can effectively realize low-latency and low-load global traffic scheduling based on global network conditions, reveal the internal relationship between computing, storage, network and other resource scheduling, and optimize software definition Network resource utilization and user experience, realize timely network resource adjustment technology that adapts to network traffic dynamics, save traffic bandwidth costs within the data center, thereby saving costs for related data centers and services, and obtaining indirect economic benefits.

Description of drawings

1 is a schematic flow diagram of a method for traffic scheduling of cross-domain data center SDN network resources according to an embodiment of the present invention;

Fig. 2 is a schematic flow diagram of solving the flow scheduling optimization mathematical model by using a non-dominated genetic algorithm with an elitist strategy;

FIG. 3 is a schematic structural diagram of a system for traffic scheduling of cross-domain data center SDN network resources according to an embodiment of the present invention.

Detailed ways

Some software-defined network features, that is, SDN network features, can be directly detected, but some need to be calculated (ie, measured). In addition, feature measurement of SDN network characteristics usually requires detection process. Feature detection adopts the following technical route:

1) Analyze the network state-related characteristics of the OpenFlow protocol, such as statistics of switch port information (number of received bytes, number of lost packets, duration, etc.), and guide the collection of characteristic data of each SDN network;

2) Investigate currently commonly used sampling algorithms, such as uniform sampling, adaptive random sampling, group sampling, and threshold sampling, analyze the advantages and disadvantages of each sampling algorithm, and further combine the characteristics of SDN networks and different network characteristic metrics to be studied. Propose a sampling algorithm suitable for different SDN network characteristics;

3) Study and analyze the qualitative and quantitative relationship between sampling frequency, load and accuracy in the sampling algorithm, and reduce the load overhead brought by data sampling to the controller. Pre-install flow entries in the network to form detection routes covering the entire network. According to the research and pre-research results of the project team on the mathematical model of network characteristics, optimize each mathematical model and propose a new model. Introduce the characteristic mathematics of each SDN network initially defined in the present invention below, specifically:

① Link bandwidth:

Symbol interpretation: where UB represents the link bandwidth, Bt ₁ and Bt ₂ represent the initial number of bytes and the number of ending bytes in the sampling time interval, respectively, and p represents the sampling time interval.

②Packet loss rate:

Symbol interpretation: the sampling interval is [i-1,i], the sampling time interval is [time(i-1),time(i)], and dropped indicates the number of discarded bytes contained in the switch response information.

③Delay:

Delay _s1&s2 ＝Totaltime-Delay _c&s1 -Delay _c&s2

Symbol interpretation: Delay _s1&s2 represents the delay between switches s1 and s2, Totaltime represents the total time from the origin of the data packet to the termination of the data packet, Delay _c&s1 and Delay _c&s2 represent the delay between the controller and switches s1 and s2, respectively.

④ Jitter:

Symbol interpretation: The sampling interval is [i-1, i], the delay of this interval is Delay(i-1, i), and θ is the noise effect.

At this stage, when optimizing and constraining the computing power resources of the data center, generally only consider the index factors such as delay, bandwidth, and packet loss rate. In the model, time delay, energy consumption, bandwidth, etc. are used as performance indicators. By establishing different objective functions, cloud edge node computing power, differentiated delay, and storage capacity are used as constraint targets to establish a mathematical model. Network resources and computing resources Reasonable allocation of computing tasks to ensure that computing task delays and system energy consumption are optimized under constraints such as limited resources and task priorities.

The multi-objective optimization problem is composed of decision variable parameters, objective functions and constraints. The optimization objective functions of the entire network are divided into minimizing service delay, maximizing the experience quality of sub-data centers, minimizing energy consumption of edge servers, and minimizing link loss. Packet rate, maximize network path bandwidth, minimize link jitter rate:

minf ₁ = T _average (1)

maxQoE(2)

min∑ _l g _l (x _l )＝C (3)

D ^T x _uv ≤ d (13)

J ^T x _uv ≤ j (14)

∏LossRate(i)≤L (15)

The explanations to the above formula (1) to formula (15) are as follows:

De^m表示应用m上传到云数据中心的时延，

表示在第i个边缘服务器上上传到云计算数据中心的应用m的请求的总数，还包括Delay_s1&s2表示交换机s1和s2之间的延迟。公式(1)中，f₁表示时延函数，minf₁表示最小化时延目标函数，T_average表示网络流调度总平均时延。1) Formula (1) expresses the minimum average delay of user services, including the delay uploaded to the cloud computing sub-data center, expressed as

De ^m represents the delay when application m is uploaded to the cloud data center,

Indicates the total number of requests of the application m uploaded to the cloud computing data center on the i-th edge server, and Delay _s1&s2 also indicates the delay between switches s1 and s2. In formula (1), f ₁ represents the delay function, minf ₁ represents the objective function of minimizing the delay, and T _average represents the total average delay of network flow scheduling.

2) Equation (2) expresses that maximizing the quality of experience Qoe of the sub-data center is related to computing power. In the formula (2), QoE represents the quality of experience index function of the network service of the sub-data center, maxQoE represents the maximum quality of experience QoE, and maxQoE is an evaluation standard of the network quality of experience set artificially;

3) Equation (3) means to minimize the energy consumption of the edge server network performance, which is related to the storage capacity of the edge server. In the formula (3), ∑ _l g _l (x _l ) means that the energy consumption on the link obeys the function, min∑ _l g _l (x _l ) represents the minimized link energy consumption, and C represents the total power consumption in the entire edge network.

表示最小化链路中采样区间i到n的丢包率，4) Formula (4) represents the minimum link packet loss rate, and list represents the link set. In the formula (4), L _list represents the packet loss rate function of the link list,

Indicates to minimize the packet loss rate of the sampling interval i to n in the link,

表示最大化网络路径l_e的带宽利用率，UB表示带宽计算公式，l_e表示某一链路；5) Formula (5) is to maximize the bandwidth utilization of the network path, and UB represents the link bandwidth. In the formula (5), U _list represents the bandwidth calculation objective function of the link List,

Indicates to maximize the bandwidth utilization of the network path l _e , UB indicates the bandwidth calculation formula, and l _e indicates a certain link;

表示最小化链路List的抖动率，Jitter(i)为抖动率计算公式，l_e表示某一链路。6) Equation (6) represents the link packet loss rate. In the formula (6), J _list represents the calculation of the jitter rate of the link List,

Indicates the jitter rate of the minimized link List, Jitter(i) is the formula for calculating the jitter rate, and l _e indicates a certain link.

表示时间T内每个t时间节点内N个计算节点的平均能耗计算函数，其中，Q是分数据中心计算能力限制，/>

为时间t节点的能耗，T为总时间，N为算力节点。7) Equation (7) expresses the relationship between the maximum computing capacity limit of the sub-data center and the energy consumption of the node at time t. In formula (7),

Indicates the average energy consumption calculation function of N computing nodes in each t time node in time T, where Q is the computing capacity limit of the sub-data center, />

is the energy consumption of the node at time t, T is the total time, and N is the computing power node.

表示存储K节点内的总存储能力计算，/>

表示时间节点t的数据中心计算任务，C_n表示存储容量限制，c_k表示k存储节点的容量表示。8) Formula (8) represents the limitation constraint condition of the maximum storage capacity of the sub-data center; in formula (8),

Indicates the calculation of the total storage capacity in the storage K node, />

Represents the data center computing task at time node t, C _n represents the storage capacity limit, and c _k represents the capacity representation of k storage nodes.

表示时间t节点时，数据中心a节点和b节点任务之间的带宽需求，

表示n个节点网络带宽的最大限制。9) Equation (9) represents the maximum limit constraint condition of the network bandwidth of the sub-data center. In formula (9),

Indicates the bandwidth requirement between the tasks of node a and node b in the data center at time t node,

Indicates the maximum limit of the network bandwidth of n nodes.

为能耗服从函数(3)中∑_lg_l(x_l)的自变量，x_l为存储能力，/>

表示i和j节点在l任务集合下的k服务节点资源利用率，i和j表示边缘服务器，l表示某个任务，E为任务集合，k为服务节点。10) Formulas (10), (11) and (12) represent the minimized edge server network performance and energy consumption costs, which are related to conditions such as edge server storage capacity and network node performance. In formula (10),

is the independent variable of ∑ _l g _l (x _l ) in the energy consumption obedience function (3), x _l is the storage capacity, />

Indicates the utilization rate of k service node resources of i and j nodes under the l task set, i and j represent the edge server, l represents a certain task, E is the task set, and k is the service node.

表示x_l为最大存储能力限制为c_l。In formula (11),

Indicates that x _l is the maximum storage capacity limited to c _l .

表示边缘服务器i和j在k下的网络节点的性能评估表示，d_ij为性能评估值。In formula (12),

Indicates the performance evaluation representation of the network nodes of edge servers i and j under k, and d _ij is the performance evaluation value.

11) Equation (13) indicates that d represents the boundary of the network application on the delay constraint, D represents the column vector of the delay of each side, and x _uv represents a link. In the formula (13), D ^T x _uv ≤ d, which means that d represents the boundary of the delay constraint of the network application, D represents the column vector of the delay of each side, and x _uv represents a link.

12) Formula (14) link jitter is lower than the constraint limit, j represents the boundary of the network application on jitter constraints, and J represents the column vector of jitter on each side. In formula (14), J ^T x _uv ≤ j, the link jitter is lower than the constraint limit, j represents the boundary of the network application on jitter constraints, and J represents the column vector of jitter on each side.

13) Formula (15) L represents the maximum requirement on the packet loss rate. In formula (15), ∏LossRate(i)≤L, LossRate(i) represents the sum of the packet loss rate, and L represents the maximum requirement for the packet loss rate.

As shown in Figure 1, a method for traffic scheduling of cross-domain data center SDN network resources according to an embodiment of the present invention includes the following steps:

S1. Add the computing power of the cross-domain data center and the network performance and energy consumption of the edge server to the multi-objective optimization model NSGA-II algorithm, and respectively establish the corresponding calculation capabilities for the computing power of the cross-domain data center and the network performance and energy consumption of the edge server Objective function to get the traffic scheduling optimization objective model;

Among them, the calculation process of the cross-domain data center is as follows:

S010. Establish a first constraint condition according to the performance parameters of the cross-domain data center;

Among them, the performance parameters of the cross-domain data center include cache space and network link bandwidth. The established first constraint condition includes formula (7), formula (8) and formula (9), namely:

网络带宽的最大限制，/>

时间t节点的能耗。Among them, Q is the computing power limit, C _n is the maximum storage capacity,

The maximum limit of network bandwidth, />

Energy consumption at time t node.

S011. Using the multi-objective optimization model with the first constraint added, combined with the objective function established for the computing capability of the cross-domain data center, to calculate the computing capability of the cross-domain data center;

Among them, the interpretation of the multi-objective optimization model is as follows:

At this stage, when constraining the computing resources of data centers, scheduling across data centers through a pure network generally considers factors such as delay and bandwidth. Comprehensively consider the computing power resources and network performance of the sub-data center and edge server. The sub-data center and the general dispatching are based on the SDN architecture, and the openflow protocol needs to be extended; individual sub-data centers are connected to the general dispatching through the traditional Internet. For data collection, this special situation needs to be considered. On the basis of traditional computing resource regulation , adding the computing power of the cross-domain data center and the network performance, energy consumption and storage capacity of the edge server into the multi-objective optimization model NSGA-II algorithm.

Among them, the objective function set for the computing power of the cross-domain data center is formula (2);

In the algorithm implementation process of the multi-objective optimization model with the first constraint added, the computing capability of the cross-domain data center can be calculated through the formula (2) and the first constraint.

Among them, the calculation process of the optimal network performance and energy consumption of the edge server is as follows:

S020. Establish a second constraint condition according to the performance parameters of the edge server;

The performance parameters of the edge server include storage capacity and network performance, etc., and the established second constraints include formula (10), formula (11) and formula (12) as follows:

S021. Using the energy consumption obedience function under the second constraint condition and the objective function corresponding to the network performance and energy consumption of the edge server, calculate the optimal network performance and energy consumption of the edge server.

Among them, the objective function established for the network performance and energy consumption of the edge server is formula (3);

The energy consumption obedience function is g _l (x _l ) = μ _l x _l α, where μ _l and α are parameters related to the computing power of the edge server network node equipment, c _l is the storage capacity, and d _ij is the performance of the network node Evaluation, in the implementation process of the algorithm, the optimal solution of energy consumption after the objective function trade-off can be obtained through the calculation of formula (3), that is, the optimal network performance energy consumption of the edge server can be obtained.

Among them, the specific explanation of the multi-objective optimization model NSGA-II algorithm is as follows:

The multi-objective optimization model NSGA-II algorithm is a fast non-dominated multi-objective optimization algorithm with an elite retention strategy. It is a multi-objective optimization algorithm based on the Pareto optimal solution. It reduces the complexity of the non-inferior sorting genetic algorithm and has The running speed is fast, the convergence of the solution set is good, and the crowding degree and the crowding degree comparison operator are introduced, which not only overcomes the defect of manually specifying the shared parameters in the NSGA algorithm, but also uses the crowding degree as a comparison criterion between individuals in the population, The population individuals in the quasi-Pareto domain can be evenly expanded to the entire Pareto domain, thereby ensuring the diversity of the population.

In the multi-objective optimization solution, some optimization objective functions seek the maximum value and some seek the minimum value. For the convenience of calculation, the maximum value of all objective functions is usually converted to the minimum value. The mathematical description of the multi-objective optimization problem in this paper is:

MinF(X)＝[f ₁ (X),f ₂ (X),…,f _m (X)]

stg _i (X)≤0,i=1,2.,…k

h _j (X)=0,j=1,2...l

Among them, F(X)=[f ₁ (X),f ₂ (X),…,f _m (X)] ^T is the goal of the problem, and the space where it is located is called the target space, which is the number of sub-goals, X= [x ₁ ,x ₂ ,…,x _n ] ^T is an n-dimensional vector on a given R ⁿ space, the space where X is called is the decision space of the problem, h _j (X)=0, j=1,2 ,...l is an equality constraint, g _i (X)≤0, i=1, 2...k is an inequality constraint.

S2. Establish corresponding objective functions and constraint conditions for each SDN network characteristic of the SDN network resources, and add them to the traffic scheduling optimization target model to obtain the traffic scheduling optimization mathematical model;

Among them, the SDN network characteristics of SDN network resources include: link bandwidth, packet loss rate, delay and jitter. specifically:

The objective function established for link bandwidth is formula (5), the constraint condition established for link bandwidth is formula (9), the objective function established for packet loss rate is formula (4), and the constraint condition established for packet loss rate is formula (15), the objective function established for delay is formula (1), the constraint condition established for delay is formula (13), the objective function established for jitter is formula (6), and the constraint condition established for link bandwidth is formula (14).

S3. Using a non-dominated genetic algorithm with an elite strategy to solve the traffic scheduling optimization mathematical model, and obtain a traffic scheduling scheme for traffic scheduling of SDN network resources, as shown in Figure 2, specifically:

The purpose of optimization is to use the multi-path allocation optimization strategy to deploy the global network path with the smallest packet loss rate, transmission delay, and edge server energy consumption compared to the network before optimization, and the business will tend to flow to lighter links In general, the network has a higher acceptance capacity for incoming connection requests without rerouting existing connections. The specific process of solving the traffic scheduling optimization mathematical model by using the non-dominated genetic algorithm with elite strategy is as follows:

分数据中心数量D_n，/>

表示数据中心链路带宽限制，表示控制器数量K等。S31. The initial definition parameter is the SDN network topology graph G(V, E, C), where V represents the set of switch nodes, E represents the set of detected links, the set of bandwidth C, the number of user hosts is P, and the number of edge servers is N , the edge server storage capacity x _l , the computing capacity limit Q of the sub-data center, and the upload delay of the sub-data center

The number of sub-data centers D _n , />

Indicates the data center link bandwidth limit, indicates the number of controllers K, etc.

S32. Initialization of the population. Randomly create a parent population P with a population size of N. Each individual contains L chromosomes, and L represents the number of candidate data streams. Each individual can be represented by the following genes:

1 2 3 4 ... N

The number in each chromosome represents the path selected by the candidate path, and the number and the candidate path have the following simple correspondence as shown in Table 1 below.

Table 1:

label path 1 candidate path 1 2 candidate path 2 ... ... m Candidate path M

S33. The model evaluates each individual based on the multiple constraints proposed above, and calculates the minimum average delay through the defined algorithm parameters, maximizes the experience quality of the sub-data center, minimizes the network performance and energy consumption on the edge server, and minimizes The value of the six objective functions of link packet loss rate, jitter rate and maximizing network path bandwidth utilization.

S34. Perform a fast non-dominated sorting on the population in combination with the objective function, and assign a corresponding non-dominated fitness value to each individual. According to the fitness function:

f=Af ₁ +BQoE+C∑ _l g _l (x _l )+DL _list +EU _list +JH _list These six optimization objectives are constructed, where A, B, C, D, E, and F are weight coefficients, different The weight coefficients are used to indicate the importance of the optimization of these six objectives. In order to reflect the overall balance, the weight coefficients are all set to 1. The smaller the value of the objective function, the better the individual.

Fast non-dominated sorting performs non-dominated sorting on each chromosome in the initial solution space according to different objective functions, and generates different levels of solution sets according to the sorting results. For example, the different grades obtained after performing non-dominated sorting on the partial initial population given in step S32 are:

Rank0={Label2, Label3}, Rank1={Label1, Label3}, Rank2={Label3}, Rank3={Label5, Label7}...

S35. Generate a first-generation sub-population of the population using a binary tournament algorithm, which includes performing selection, crossover, and mutation operations on the population. The tournament method selection strategy first randomly selects k individuals from the population, where k is N/2. Then take 2 individuals from the population composed of k individuals each time, select the best individual to enter the pairing pool, and repeat this step until the number of individuals in the pairing pool reaches the set value, here is N. The crossover operator and mutation operator used are two-point crossover and single-point mutation respectively, and these two operators are used to operate on the individuals in the matching pool to generate offspring populations. The crossover rate is 0.8 and the mutation rate is 0.1.

S36. Merge the populations of the parent generation and the child generation to generate a new population, the number of the merged population is the sum of the populations of the parent generation and the child generation.

S37. Calculate the non-dominated level again through the fast non-dominated sorting algorithm, and calculate the crowding degree of individuals in the population at the same time; select excellent individuals in the population to form a new parent population according to the non-dominated sorting and the calculated crowded degree.

S38. Judging the number of iterations, which can be set to T=50, 100, 150..., when the preset algebraic value is reached, cluster the generated population, and select an individual in each class to strengthen, and then combine with the parent Compared with the generation population, select the best and leave it as the new parent population; when the number of iterations of the preset value is not reached, transfer to S39 for execution;

S39. Use the binary tournament algorithm to perform selection, crossover, and mutation operations on the newly generated population to generate a new subpopulation, and judge whether the maximum number of iterations is reached. If it is reached, the algorithm is terminated. If it is not reached, continue to iterate and jump to S36 for execution.

S40. Analyze the importance of each goal according to the actual situation, obtain a final set of Pareto optimal solutions, and select a suitable solution from the Pareto optimal solution set as the final optimal scheduling scheme.

Optionally, in the above technical solution, the acquisition process of the computing capability of the cross-domain data center includes:

Since the network traffic scheduling problem is a multi-objective optimization problem, the final solution sought is not the only solution, but an optimal solution set for multiple choices. It is necessary to analyze the importance of each objective according to the actual situation and choose the best solution. Scheduling scheme, the optimization result designed by the present invention should first satisfy the first, second and third objective functions. In order to control the energy consumption and transmission delay of the edge server, when selecting the optimization strategy, it is necessary to take into account the fourth, fifth and third objective functions to a certain extent. Six objective functions. The model comprehensively considers a variety of situations for different user needs. When the real-time requirements for traffic scheduling are relatively high, the importance of the delay function can be increased, and the scheduling scheme with the smallest delay function value can be selected to ensure the real-time performance of data transmission. When the number of services deployed on the edge network increases, more user requests will be processed on the edge nodes connected to it. In this case, there will be no transmission delay and delay in uploading to the cloud computing sub-data center, so Users can get a lower network delay and better user experience. However, the corresponding power consumption will increase. This is because in order to obtain a lower network delay, more deployed services will increase the number of edge servers powered on, and the load will also increase, resulting in increased power consumption. Therefore, this algorithm can improve the utilization efficiency of network resources and save the cost of traffic bandwidth inside the data center, thereby saving costs for related data centers and services and obtaining indirect economic benefits.

In practical applications, one of the non-dominated solutions obtained from NSGA-II can be selected for deployment as required. For example, you can choose some compromise solutions, and try to choose some solutions with lower power consumption for actual traffic scheduling deployment under the condition of satisfying the network delay of users and applications, so as to obtain the maximum benefit.

The inventor's beneficial effects are as follows:

The application scope of software-defined network not only includes cloud computing and enterprise network fields, but also backbone network operators will partially apply SDN technology. Therefore, SDN will have a profound impact on the business model of network applications, can further promote the industrial innovation of Internet applications, and has broad and far-reaching application value and industrialization prospects. This patent focuses on the future cross-domain data center of cloud-edge integration, providing cloud computing and edge computing integration services. This cloud-edge integration service model combines the complementary advantages of cloud computing and edge computing. With the development of Internet applications, With continuous deepening, the research results of this patent have definite market development potential.

In terms of economic benefits, the technical achievements of this project can directly provide software-defined network deployment solutions and technical support for cross-domain data centers with cloud-edge integration in the future, so as to obtain relevant direct economic benefits. In addition, the unified management of software-defined network resources and traffic scheduling theory studied in this patent can improve the utilization efficiency of network resources and save the cost of traffic bandwidth inside the data center, thereby saving costs for related data centers and services and obtaining indirect economic benefits.

In each of the above-mentioned embodiments, although the steps are numbered S1, S2, etc., they are only specific embodiments provided by the application, and those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation. Within the protection scope of the present invention, it can be understood that, in some embodiments, part or all of the foregoing implementation manners may be included.

As shown in FIG. 3 , a system 200 for traffic scheduling of cross-domain data center SDN network resources according to an embodiment of the present invention includes a first addition and establishment module 210, a second addition and establishment module 220, and a solution module 230;

The first addition building module 210 is used to: add the computing power of the cross-domain data center and the network performance and energy consumption of the edge server to the multi-objective optimization model NSGA-II algorithm, and respectively add the computing power of the cross-domain data center and the network performance and energy consumption of the edge server The network performance and energy consumption establishes the corresponding objective function, and obtains the optimization objective model of traffic scheduling;

The second addition and establishment module 220 is used to: respectively establish corresponding objective functions and constraint conditions for each SDN network characteristic of the SDN network resource, and add them to the traffic scheduling optimization target model to obtain the traffic scheduling optimization mathematical model;

The solving module 230 is used for: adopting the non-dominated genetic algorithm with elitist strategy to solve the flow scheduling optimization mathematical model, and obtain the flow scheduling scheme for flow scheduling of SDN network resources.

Optionally, in the above technical solution, a first acquisition module is also included, and the first acquisition module is used for:

Establishing the first constraint condition according to the performance parameters of the cross-domain data center;

The computing capability of the cross-domain data center is calculated by using the multi-objective optimization model with the first constraint and combining with the objective function established for the computing capability of the cross-domain data center.

Optionally, in the above technical solution, a second acquisition module is also included, and the second acquisition module is used for:

establishing a second constraint condition according to the performance parameter of the edge server;

Using the energy consumption obedience function under the second constraint condition, combined with the objective function corresponding to the network performance and energy consumption of the edge server, the optimal network performance and energy consumption of the edge server is calculated.

Optionally, in the above technical solution, the objective function of the storage capability of the edge server is: optimal network performance and energy consumption of the edge server.

Optionally, in the above technical solution, the SDN network characteristics of the SDN network resources include: link bandwidth, packet loss rate, delay and jitter.

For the steps of realizing the corresponding functions of each parameter and each unit module in the system of traffic scheduling of a cross-domain data center SDN network resource of the present invention, please refer to the above-mentioned traffic scheduling of a cross-domain data center SDN network resource The parameters and steps in the embodiment of the method will not be repeated here.

An electronic device according to an embodiment of the present invention includes a memory, a processor, and a program stored in the memory and run on the processor. When the processor executes the program, any one of the above-mentioned implementations can be realized. Steps of a method for traffic scheduling of cross-domain data center SDN network resources.

Among them, the electronic device can be a computer, a mobile phone, etc. Correspondingly, its program is a computer software or a mobile phone APP, etc., and the above-mentioned parameters and steps in an electronic device of the present invention can refer to the above-mentioned cross-domain The parameters and steps in the embodiment of the method for traffic scheduling of data center SDN network resources will not be repeated here.

Those skilled in the art know that the present invention can be implemented as a system, method or computer program product.

Therefore, the present disclosure can be specifically implemented in the following forms, that is: it can be complete hardware, it can also be complete software (including firmware, resident software, microcode, etc.), and it can also be a combination of hardware and software. Called a "circuit", "module" or "system". Furthermore, in some embodiments, the present invention can also be implemented in the form of a computer program product embodied in one or more computer-readable media having computer-readable program code embodied therein.

Any combination of one or more computer readable medium(s) may be utilized. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more leads, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), Erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention, those skilled in the art can make the above-mentioned The embodiments are subject to changes, modifications, substitutions and variations.

Claims (10)

1. A method for traffic scheduling of SDN network resources in a cross-domain data center, comprising:

adding the computing capacity of a cross-domain data center and the network performance energy consumption of an edge server into a multi-objective optimization model NSGA-II algorithm, and respectively establishing corresponding objective functions for the computing capacity of the cross-domain data center and the network performance energy consumption of the edge server to obtain a flow dispatching optimization objective model;

establishing a corresponding objective function and constraint conditions for each SDN network feature of SDN network resources respectively, and adding the objective function and constraint conditions into the flow scheduling optimization objective model to obtain a flow scheduling optimization mathematical model;

and solving the flow scheduling optimization mathematical model by adopting a non-dominant genetic algorithm with elite strategy to obtain a flow scheduling scheme for performing flow scheduling on the SDN network resources.

2. The method for traffic scheduling of SDN network resources of a cross-domain data center of claim 1, wherein the process of obtaining computing power of the cross-domain data center comprises:

establishing a first constraint condition according to the performance parameters of the cross-domain data center;

and calculating the computing capacity of the cross-domain data center by utilizing a multi-objective optimization model added with the first constraint condition and combining an objective function established for the computing capacity of the cross-domain data center.

3. The method for traffic scheduling of SDN network resources of a cross-domain data center of claim 2, wherein the process of obtaining the optimal network performance energy consumption of the edge server includes:

establishing a second constraint condition according to the performance parameters of the edge server;

and calculating the optimal network performance energy consumption of the edge server by utilizing the energy consumption obeying function under the second constraint condition and combining an objective function corresponding to the network performance energy consumption of the edge server.

4. A method for traffic scheduling of SDN network resources of a cross-domain data center according to claim 3, wherein an objective function of network performance energy consumption of the edge server is: and the optimal network performance energy consumption of the edge server.

5. A method of traffic scheduling of SDN network resources of a cross-domain data center according to any of claims 1 to 4, characterized in that SDN network characteristics of the SDN network resources comprise: link bandwidth, packet loss rate, delay, and jitter.

6. The system for flow scheduling of SDN network resources of the cross-domain data center is characterized by comprising a first adding and establishing module, a second adding and establishing module and a solving module;

the first adding and establishing module is used for: adding the computing capacity of a cross-domain data center and the network performance energy consumption of an edge server into a multi-objective optimization model NSGA-II algorithm, and respectively establishing corresponding objective functions for the computing capacity of the cross-domain data center and the network performance energy consumption of the edge server to obtain a flow dispatching optimization objective model;

the second adding and establishing module is used for: establishing a corresponding objective function and constraint conditions for each SDN network feature of SDN network resources respectively, and adding the objective function and constraint conditions into the flow scheduling optimization objective model to obtain a flow scheduling optimization mathematical model;

the solving module is used for: and solving the flow scheduling optimization mathematical model by adopting a non-dominant genetic algorithm with elite strategy to obtain a flow scheduling scheme for performing flow scheduling on the SDN network resources.

7. The system for traffic scheduling of SDN network resources of a cross-domain data center of claim 6, further comprising a first acquisition module configured to:

establishing a first constraint condition according to the performance parameters of the cross-domain data center;

and calculating the computing capacity of the cross-domain data center by utilizing a multi-objective optimization model added with the first constraint condition and combining an objective function established for the computing capacity of the cross-domain data center.

8. The system for traffic scheduling of SDN network resources of a cross-domain data center of claim 7, further comprising a second acquisition module configured to:

establishing a second constraint condition according to the performance parameters of the edge server;

and calculating the optimal network performance energy consumption of the edge server by utilizing the energy consumption obeying function under the second constraint condition and combining an objective function corresponding to the network performance energy consumption of the edge server.

9. The system for traffic scheduling of SDN network resources of a cross-domain data center of claim 8, wherein an objective function of network performance energy consumption of the edge server is: and the optimal network performance energy consumption of the edge server.

10. A system for traffic scheduling of SDN network resources of a cross-domain data center according to any of claims 6 to 9, characterized in that SDN network characteristics of the SDN network resources comprise: link bandwidth, packet loss rate, delay, and jitter.

Images