SDN (Software Defined Network) data center network energy-saving traffic scheduling method, equipment and storage medium

Abstract

The invention discloses an SDN (Software Defined Network) data center network energy-saving traffic scheduling method, equipment and a storage medium, and belongs to the technical field of computer networks. The method comprises the following steps: determining available paths through differential evolution according to real-time traffic by utilizing the global monitoring capability of SDN on network nodes, taking a result of differential evolution as input of a simulated annealing algorithm, determining a global optimal path from the available paths, selecting a path with the highest bandwidth utilization rate while meeting the current traffic demand, and selecting a path with the highest bandwidth utilization rate. The unused switches are closed or dormant, so that the transmission of the network flow is concentrated on the least links, the number of devices working simultaneously in the network is reduced, and the purpose of reducing the energy consumption of the data center network is achieved. According to the scheduling method, the differential evolution algorithm and the simulated annealing algorithm are fused, global scheduling of the flow is achieved on the premise that the network performance requirement is met, and therefore the purpose of saving energy is achieved.

Description

Technical Field

[0001] This invention relates to the field of computer network technology, and in particular to an energy-saving traffic scheduling method, device and storage medium for SDN data center networks. Background Technology

[0002] With the rapid development of information technology and the arrival of the digital age, data centers, as the core infrastructure supporting cloud computing, big data, and various online services, are gradually becoming an indispensable part of modern society. However, as the scale and workload of data centers continue to expand, their energy consumption is also increasing year by year. According to statistics, global data center energy consumption already accounts for a considerable portion of total energy consumption. In the past five years, my country's data center energy consumption has maintained a growth rate of 15%, reaching 270 billion kWh in 2023, accounting for approximately 3.1% of the country's total electricity consumption, enough to meet the annual electricity needs of two megacities.

[0003] Data center equipment includes peripheral devices and network equipment, with network equipment accounting for more than 20% of the total energy consumption of a data center. To ensure the reliability and availability of the data center, network devices such as switches provide network services almost without interruption, operating at nearly full power regardless of the actual load on the data center network, which results in significant energy waste.

[0004] Current data center networks utilize "rich connection" architectures, with Fat-trees being the most popular. These architectures are simple, offer high bandwidth, and provide multiple equal-cost routes to the end. However, compared to traditional data center network architectures, the introduction of "rich connection" architectures further increases network energy consumption due to the increased number of network devices required.

[0005] Currently, the most commonly used network traffic scheduling algorithm in data and central networks is the Equal-Cost Multi-Path routing (ECMP) algorithm. For the same destination node, if multiple equal-cost paths exist, the ECMP algorithm calculates all equal-cost paths and performs a hash operation on each data flow. Based on the calculation results, it distributes the data evenly across these paths, thereby achieving network load balancing and effectively preventing one or more paths from consuming excessive bandwidth and affecting network service quality. However, the purpose of the ECMP algorithm is to achieve network load balancing, and it does not consider energy consumption.

[0006] Software-defined networking (SDN), as an emerging network architecture, separates the network control plane from the data plane, giving network administrators greater control and flexibility, thus providing new opportunities for network optimization and energy saving. Traditional data center networks typically consist of static hardware switches and routers, which struggle to meet dynamically changing traffic demands, leading to unnecessary increases in energy consumption. The introduction of SDN technology enables data center networks to dynamically adjust and optimize based on real-time traffic conditions, thereby reducing energy consumption and improving network resource utilization efficiency. In SDN data center networks, administrators can easily implement sleep or open switching device ports.

[0007] With the development of SDN technology, many scholars have proposed applying SDN technology to data center networks and have put forward energy-saving methods based on SDN technology to address the energy consumption problem of data center networks. However, these energy-saving methods all have some shortcomings. For example, in energy-saving methods with traffic prediction, the traffic prediction accuracy is insufficient, resulting in untimely energy-saving scheduling of the network; the time complexity of the algorithms is high, the convergence speed is very slow, and they cannot find the globally optimal path within a certain time, which will increase network latency and reduce the quality of service of the network; some algorithms only consider one or two network parameters, or even do not use real-time network data, resulting in incomplete consideration and the path found is not the optimal solution. Summary of the Invention

[0008] To address the aforementioned problems, this invention aims to provide an energy-saving traffic scheduling method, device, and storage medium for SDN data center networks. This scheduling method integrates differential evolution and simulated annealing algorithms to achieve global traffic scheduling while meeting network performance requirements, thereby achieving the goal of energy saving.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0010] A method for energy-saving traffic scheduling in an SDN data center network, characterized by the following steps:

[0011] S1: Establish an SDN data center energy consumption model;

[0012] S2: Based on the differential evolutionary fusion simulated annealing algorithm, traffic optimization scheduling is performed on the SDN data center energy consumption model;

[0013] S3: Based on the traffic optimization scheduling scheme, shut down or put into hibernation any unused switches.

[0014] Furthermore, the SDN data center energy consumption model in step S1 is as follows:

[0015]

[0016] The constraints include:

[0017]

[0018]

[0019]

[0020]

[0021]

[0022] In the formula, the data center network topology is represented as graph G(S,L), where S is the set of all switches in the network, and s is the number of nodes. i ∈S (i=1,2,…,|S|); L represents the set of links connected to all switches in the network, and the bandwidth capacity of link l∈L is C. l Its link utilization rate is u l The set of all network flows in the network is E, and the network flow e∈E; L i and L' i These are subsets of the link set L, and switches. For link l i ∈L i The final node, For l' i ∈L' i The starting node; the data flow passes through l i Inflow to switch Data stream passes through l' i Inflow to switch Let E be the set of data flows in a data center network, and let b be the bandwidth of data flow e∈E. e The source and destination nodes of data stream e are represented as follows: and Indicates whether data stream e passes through link l; θ s θ l These indicate whether the switch and link are in working order; u l Indicates link utilization; u max Indicates maximum link utilization; Cost l Energy consumption of the link; Cost s This refers to the power consumption of the switch.

[0023] Furthermore, the specific operation of step S2 includes the following steps:

[0024] S201: Continuously collect data streams arriving at the data center network;

[0025] S202: Based on the differential evolution algorithm, each network flow data is used as the input of the differential evolution algorithm to calculate the alternative path that meets the bandwidth size of the current data flow;

[0026] S203: Use the candidate paths obtained in step S202 as the initial paths of the simulated annealing algorithm, and use the simulated annealing algorithm to determine the global optimal path;

[0027] S204: Repeat steps S202 and S203 until the iteration condition is met and stop. Convert the obtained global optimal path into a flow table entry and send it to each network node to reroute the data flow.

[0028] Furthermore, the data flow information mentioned in step S201 includes bandwidth size, source node, destination node, and bandwidth utilization of all nodes in the network; for newly arriving data flows in the data center network, traffic scheduling is performed using the ECMP algorithm.

[0029] Furthermore, the specific operation of step S202 includes the following steps:

[0030] S2021: Differential evolution algorithm initialization, determining the initial population Pop and fitness function;

[0031] S2022: Mutation operation;

[0032] S2023: Crossover operation;

[0033] S2024: Perform selection operation based on fitness function;

[0034] S2025: Determine if the iteration condition is met. If yes, exit the loop; otherwise, repeat steps S2022 to 2024 to finally obtain a population Pop containing individuals with multiple paths.

[0035] Furthermore, the specific operations of step S2021 include the following steps:

[0036] S20211: Set the loop control variable iter to 1 and set the maximum number of iterations Maxiter;

[0037] S20212: Each network flow data is used as input to the differential evolution algorithm, and the alternative paths of the same destination nodes generated according to the KSP algorithm are used as the initial solution space R of the differential evolution algorithm;

[0038] S20213: Select M individuals X from the solution space R. i Pop is the initial population, where each individual X i Each X is a complete, usable path that meets the constraints. iEach consists of n links l.

[0039] X i ={l1,l2,…,l n};X i ∈R, 1≤i≤M

[0040] S20214: Define the fitness function as follows

[0041] F(X i ) = dis i / (b i +u i )

[0042] Where: dis i Represents individual X i The length of , i.e., the path length; b i Represents individual X i Maximum bandwidth utilization across all links; u i Represents individual X i The number of nodes that are currently in operation.

[0043] Furthermore, the specific operation of step S3 includes the following steps:

[0044] S301: Use the population Pop obtained in step S2 as the solution space R of the simulated annealing algorithm. sa And set the initialization parameters;

[0045] S302: Randomly select M individuals x from the population Pop. i (i = 1, 2, ..., M), for each individual x i Perform a change operation to generate a new individual v' i (i = 1, 2, ..., M);

[0046] S303: Based on individual v' i With x i The energy difference is used to determine whether to accept a new individual;

[0047] S304: Cooling down, current temperature T = T1·α, where T1 is the original temperature and α is the cooling rate;

[0048] S305: Determine if the current temperature T has reached the termination temperature T. d If the target is not reached, proceed to step S302; otherwise, terminate the iteration and output the globally optimal path.

[0049] Furthermore, the specific operation of step S301 includes the following steps:

[0050] S3011: Use the population Pop obtained by the differential evolution algorithm as the solution space R. saSet the initial temperature T0 and the final temperature T d ;

[0051] S3012: From the solution space R sa Select M individuals X i Pop, as the initial population sa ;

[0052] S3013: Define the energy function of an individual as follows

[0053] F(X i )=(b i +u i ) / dis i

[0054] S3014: The formula for determining whether to accept a new individual is as follows:

[0055]

[0056] where dis i Represents individual X i Length; b i Represents individual X i Maximum bandwidth utilization across all links; u i Represents individual X i The number of nodes in the active state; ΔE is the original individual energy minus the new individual energy, and T is the current temperature.

[0057] Furthermore, the present invention also includes an SDN data center network energy-saving traffic scheduling device, characterized in that the device includes at least one processor and a memory communicatively connected to the processor, the memory storing instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the method described above.

[0058] Furthermore, the present invention also includes a non-transitory computer-readable storage medium storing computer instructions, characterized in that the computer instructions are used to cause the computer to perform the method described above.

[0059] The beneficial effects of this invention are:

[0060] 1. The SDN data center network energy-saving traffic scheduling method in this invention integrates differential evolution and simulated annealing algorithms. Utilizing SDN's global monitoring capabilities of network nodes, it determines available paths through differential evolution based on real-time traffic. The result of differential evolution is used as input to the simulated annealing algorithm to determine the globally optimal path from the available paths. While meeting current traffic demands, it selects the path with the highest bandwidth utilization, shutting down or putting unused switches to sleep. This concentrates network traffic transmission on the fewest links, reducing the number of devices operating simultaneously in the network, thereby reducing data center network energy consumption. Compared to traditional network traffic scheduling algorithms that mostly focus on load balancing without considering current network energy consumption, this invention aims to minimize the number of operating switches and links. Under the premise of meeting network performance requirements, it achieves global traffic scheduling, thus achieving energy savings.

[0061] 2. Simulation experiments in this invention verify that, compared with the traditional ECMP and Genetic Ant Colony Algorithm (GA-ACO), the scheduling method in this invention reduces network energy consumption to the lowest level and improves the maximum link utilization of the data center network under the same network load, thus achieving network energy saving. Attached Figure Description

[0062] Figure 1 This is a flowchart of the SDN data center network energy-saving traffic scheduling method in this invention.

[0063] Figure 2 This is a comparison of network energy consumption under the Random communication mode in the simulation experiment of this invention.

[0064] Figure 3 The results show the network energy consumption comparison in the Staggered(0.2,0.4) communication mode during the simulation experiment of this invention.

[0065] Figure 4 This is a comparison result of the maximum link utilization of the network in the Random mode in the simulation experiment of this invention. Detailed Implementation

[0066] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0067] Example 1:

[0068] Example 1 provides an SDN data center network energy-saving traffic scheduling method (DE-SASDN). This scheduling method integrates Differential Evolution (DE) and Simulated Annealing (SA) algorithms, and specifically includes the following:

[0069] 1.1 Establishing a Data Center Energy Consumption Model

[0070] This invention models the energy consumption problem of data center networks. After ensuring sufficient bandwidth for network flows, it optimizes the objective function through traffic scheduling to minimize the number of switches and links in operation, thereby achieving network energy savings. The specific modeling description is as follows:

[0071] The data center network topology is represented as a graph G(S,L), where S is the set of all switches in the network, and s is the number of nodes. i ∈S (i=1,2,…,|S|); L represents the set of links connected to all switches in the network, and the bandwidth capacity of link l∈L is C. l Its link utilization rate is u l The set of all network flows in the network is E, and network flow e∈E. i and L' i These are subsets of the link set L, and switches. For link l i ∈L i The final node, For l' i ∈L' i The starting node; the data flow passes through l i Inflow to switch Data stream passes through l' i Inflow to switch Let E represent the set of data flows in the data center network, and let b be the bandwidth of data flow e∈E. e The source and destination nodes of data stream e are represented as follows: and variable Indicates whether data stream e passes through link l; variable θ s θ l Indicates whether the switch and link are in working order. Link utilization u l Defined as the bandwidth b of all data streams e passing through link l. e The sum of the bandwidth capacity C of link l l The ratio of maximum link utilization (MLU). max This refers to the maximum bandwidth utilization rate across all links in a data center network.

[0072] The total energy consumption of a data center network is defined as the energy consumption of all switches and links in the network that are currently in operation. The energy consumption of a link is defined as Cost. l This represents the energy consumption at both ends of the link; the energy consumption of a switch is defined as Cost. s , refers to the energy consumption of the switch. The optimization objective of the data center network energy-saving algorithm (i.e., the SDN data center energy consumption model) can be defined as Equation (1):

[0073]

[0074] When the above optimization objectives are achieved, traffic scheduling should satisfy the following constraints (2) to (7):

[0075]

[0076]

[0077]

[0078]

[0079]

[0080] Equation (2) states that the total bandwidth of the data flow e passing through link l cannot exceed the bandwidth capacity C of link l. l Equation (3) indicates that if switch s i If it is an intermediate node, then the data flow into that node is equal to the sum of the outflow data flow; Equation (4) indicates that if the switch If the source node of data flow e is the node from which data flow e originates, then data flow e will only originate from that node; Equation (5) indicates that if the switch If the destination node of data flow e is the node from which data flow e will flow, then data flow e will only flow into the node from which it will flow; Equation (6) defines the variable θ s ,θ l , The type and range of values.

[0081] 1.2 Network Energy Saving Algorithm DE-SASDN

[0082] 1.2.1 Network Energy-Saving Traffic Scheduling Mechanism

[0083] The traffic scheduling algorithm in this invention first uses a differential evolution algorithm to obtain multiple candidate paths that meet the conditions. To avoid the differential evolution algorithm getting trapped in local optima, the obtained paths are input into a simulated annealing algorithm in each algorithm iteration to obtain the globally optimal scheduling path. The flowchart of the proposed DE-SASDN algorithm is as follows: Figure 1 As shown, the specific steps are as follows.

[0084] 1) Use the ECMP algorithm to schedule newly arriving data streams and take advantage of the global monitoring capabilities of the SDN network to continuously collect parameters such as bandwidth size, source node and destination node of all data streams, as well as bandwidth utilization of all nodes in the network through the controller.

[0085] 2) Traverse the obtained network streams and use each network stream data as input for the differential evolution algorithm.

[0086] 3) Invoke the differential evolution algorithm to select 50 available paths identical to the current source and destination nodes as the initial paths for rerouting. Based on the relevant network parameters collected by the controller, select multiple alternative paths that match the current data flow bandwidth.

[0087] 4) Use multiple available alternative paths as the initial paths for the simulated annealing algorithm. Taking advantage of the fact that the simulated annealing algorithm can escape local optima with a high probability, the global optimal path can be selected.

[0088] 5) Repeat steps 3) and 4) until the iteration condition is met. Then, convert the obtained global optimal path into a flow table entry through the controller and send it to each of the relevant switches.

[0089] 6) Perform steps 4) to 5) on all network flows in the network.

[0090] 1.2.2 Call the Differential Evolutionary Algorithm to calculate available paths

[0091] The Differential Evolution Algorithm (DE), proposed by Storm et al. in 1997, is based on a genetic algorithm. It uses iterative mutation and crossover to select individuals with good fitness to survive. The DE algorithm is simple in principle, converges quickly, and involves few parameters, making it a highly efficient global optimization algorithm.

[0092] When the DE algorithm is called to optimize paths, the DE-SASDN energy-saving algorithm selects multiple available paths that meet the constraints (4) and (5) based on the current network topology and the utilization rate of each link. The algorithm takes the network topology, node status, and alternative paths of the same source and destination nodes as inputs and outputs multiple available paths. The specific steps of the algorithm are as follows:

[0093] Step 1 Algorithm Initialization

[0094] 1) Set the loop control variable iter to 1, and set the maximum number of iterations Maxiter.

[0095] 2) The alternative paths of the same source destination nodes generated by the K-Shortest-Paths (KSP) algorithm are used as the initial solution space R of the DE algorithm.

[0096] 3) Select M individuals X from the solution space R. i Pop serves as the initial population. Each individual X... i Each X is a complete, usable path that meets constraints (4) and (5). i Each consists of n links l, as shown in equation (7).

[0097] X i ={l1,l2,…,l n};X i ∈R, 1≤i≤M (7)

[0098] 4) To achieve the goal of minimizing network energy consumption optimization as stated in equation (1), this invention defines the fitness function of the DE algorithm as equation (8):

[0099] F(X i ) = dis i / (b i +u i (8)

[0100] Where: dis i Represents individual X i The length of , i.e., the path length; b i Represents individual X i The maximum bandwidth utilization among all links, i.e., the component path X i Maximum bandwidth utilization of all links in the system; u i Represents individual X i The fitness level is determined by the number of active nodes in the path, i.e., the number of switches that are not in sleep mode. A shorter path length results in higher bandwidth utilization, and a higher number of active switches in the current path leads to higher fitness. Individual X i fitness function F(X) i This reduces the optimization objective to a minimization problem, making it easier to call the DE-SASDN algorithm for optimization later.

[0101] Step 2 Mutation Operation

[0102] Considering the availability of paths in data center networks, i.e., the reachability of switches and the acyclic nature of paths, the mutation operation of the DE algorithm in this invention involves randomly selecting three individuals from the population Pop as the original individuals for mutation. In the iter-th iteration, individual X is traversed... i Among all links in (iter), select the link with the lowest link utilization and use link v. iReplacement, forming a new individual V with a mutated path. i (iter). Where link v i The solution space R' formed by multiple paths generated by the KSP algorithm has the highest fitness, and its combination can make the mutated path V i (iter) represents the path to a feasible solution.

[0103] Step 3 Cross Operation

[0104] Randomly select an individual W from the population Pop. i (iter), and with the variant individual V i (iter) performs a crossover operation, which involves traversing the two individuals and swapping all links after the first common node to obtain the crossover path C. i (iter).

[0105] In the iter-th iteration, a random number r is generated in the range [0,1]. If r is greater than the given crossover factor cf, the mutated individual V is directly retained. i (iter) as C i (iter), and proceed to step 4. Otherwise, continue to step 3.

[0106] Step 4: Select Operation

[0107] Calculate the crossover individual C according to equation (8). i The fitness of (iter), if C i (iter) has a higher fitness than the original individual X. i (iter), then the crossover individual C i (iter) replaces the original individual X i (iter) is retained in the next generation population Pop; otherwise, the original individual X is retained. i (iter) is retained in the next generation of population Pop.

[0108] If the iteration condition is met in step 5, exit the loop and obtain a population Pop containing 50 individuals along the path; otherwise, jump to step 2 and continue executing the DE algorithm.

[0109] 1.2.3 Using the simulated annealing algorithm to select the globally optimal path

[0110] While the Differential Evolutionary Algorithm (DE) is an efficient global optimization algorithm with fast convergence, it is prone to getting trapped in local optima. To address this, Simulated Annealing (SA) is introduced. This invention first uses the population Pop obtained by the fast-converging DE algorithm as the initial population Pop for the SA algorithm. saUsing the states of links and nodes in the network as input, this algorithm can output a globally optimal path more quickly and efficiently. The algorithm steps are as follows:

[0111] Step 1 Algorithm Initialization

[0112] 1) Set the initial temperature T0 and the final temperature T d Given the current temperature T and the cooling rate α, the population Pop input into the differential evolution algorithm is used as the solution space R. sa .

[0113] 2) From the solution space R sa Select M individuals X i Pop, as the initial population sa Among them, each individual X i Each X is a complete, usable path that meets constraints (4) and (5). i Each consists of n links l, as shown in equation (7).

[0114] 3) This invention defines the energy function of an individual as Equation (9). The formula for determining whether to accept a new individual is shown in Equation (10):

[0115] F(X i )=(b i +u i ) / dis i (9)

[0116]

[0117] where dis i Represents individual X i The length of , i.e., the path length; b i Represents individual X i The maximum bandwidth utilization among all links in the path; u i Represents individual X i The number of nodes in the path that are active, i.e., the number of switches that are not in sleep mode. ΔE is the difference between the original individual's energy and the new individual's energy, and T is the current temperature.

[0118] Step 2: Generate new individuals

[0119] To ensure the usability of the newly generated individuals, this invention modifies the original SA algorithm's operation of generating new individuals to randomly select M individuals x from the population Pop. i (i = 1, 2, ..., M), for each individual x i Perform the change operation, specifically: iterate through individual x iThe link with the lowest link utilization is selected. Among the multiple available paths generated by the KSP algorithm, the path with the highest bandwidth utilization and that meets the current flow size is selected to replace link l and form a new individual v'. i (i = 1, 2, ..., M).

[0120] Step 3: Determine whether to accept the new individual.

[0121] According to equation (9), calculate v' for all individuals. i With x i The energy is calculated, and the corresponding energy difference is determined. According to equation (10), it is determined whether to accept the new individual. If ΔE < 0, the new individual v' i Replace individual x i Preserved for the next generation Pop sa If ΔE > 0, calculate P(ΔE) according to equation (11) and generate a random number r in the range (0,1). If r < P(ΔE), the new individual v' i Replace individual x i Preserved for the next generation Pop sa Otherwise, retain the original individual x. i .

[0122] Step 4: Cooling

[0123] The original temperature was T1, and the current temperature was T = T1·α.

[0124] Step 5: Determine if the current temperature T has reached the termination temperature T. d If the target is not reached, proceed to step 2; otherwise, terminate the iteration, output the current best individual to the controller, and the controller will generate flow table entries and reroute the flow.

[0125] Simulation experiment:

[0126] To verify the performance of the proposed DE-SASDN energy-saving algorithm, the ECMP and GA-ACS algorithms were selected as comparison algorithms. The ECMP algorithm is the most widely used traffic scheduling algorithm in current data center networks. It schedules network flows to multiple equal-cost paths by performing a hash operation. The GA-ACS algorithm is an algorithm that uses a genetic algorithm combined with an ant colony algorithm for energy-saving traffic scheduling in SDN data center networks. First, the controller collects network information, which is used as input to the genetic algorithm. Through optimization by the genetic algorithm, a usable population is output as the initial population for the ant colony algorithm. The ant colony algorithm continues to search for the global optimum, finally outputting the optimal path. The controller then issues a rerouting flow table to complete the energy-saving operation.

[0127] This paper uses network energy consumption and maximum network bandwidth utilization as indicators to measure the performance of the algorithm. Network energy consumption refers to the number of switches currently in operation in the network and the total energy consumption of the links, as shown in Equation (11).

[0128]

[0129] Where θ s θ l Cost is a variable representing whether the switch / link is in a working state. s and Cost l This represents the energy consumption of a switch / link, with specific values ​​shown in Table 1 below. Under the same load, lower network energy consumption indicates higher network energy efficiency and better energy-saving effect.

[0130] Table 1 Network Equipment Power Consumption (W)

[0131] hierarchy switch link Core layer 176 2.18 edge layer 135 1.80

[0132] Experimental environment

[0133] Network Topology and SDN Controller: This invention uses the Mininet simulation platform and Python to build a k=4 Fat-tree network topology. All switches are Open vSwitch (OVS), totaling 20 units. Each access layer OVS connects to two hosts, totaling 16 hosts. All link bandwidths in the network are set to 10Mb / s. The effectiveness of the proposed algorithm is verified using a Python 3.0-based Ryu controller.

[0134] Communication Modes: This invention utilizes the Iperf traffic generation tool built into the Mininet simulation platform for secondary development, generating data streams with two different communication modes. These two communication modes are described below:

[0135] 1) Random mode. The source and destination hosts are randomly selected in the network.

[0136] 2) Staggered mode (p1, p2). Each host transmits data to hosts belonging to the same access layer OVS with probability p1; transmits data to hosts belonging to the same Pod with probability p2; and sends data to hosts of other Pods with probability (1-p1-p2).

[0137] In this simulation experiment, the data stream size follows an exponential distribution, and the data stream duration follows a Poisson distribution. Each experiment lasts for 60 seconds, and data from the 20-40 second time interval is collected. This process is repeated 10 times, and the average value of the obtained data is taken as the valid experimental data.

[0138] Algorithm parameter settings: As a heuristic optimization algorithm, the control parameter settings directly affect the optimization effect. This invention aims at energy-saving traffic scheduling and, based on existing research results and multiple simulation experiments, selects the main parameters shown in Table 2 as the relevant parameter settings for this experiment.

[0139] Table 2 Main parameter values ​​of DE-SASDN energy-saving algorithm

[0140] parameter value Maximum number of iterations (Maxiter) 30 Number of individuals M in the population 50 Cross factor cf 0.3 <![CDATA[Initial temperature T0]]> 100 <![CDATA[Termination temperature T d > 10 Cooling rate α 0.9

[0141] Through multiple experimental tests, this simulation experiment selected 30 iterations and a final temperature of 100 degrees Celsius. This paper sets the maximum path length between two points to 11 hops. In a Fat-tree network topology with k=4, the probability of a path exceeding 11 hops being the optimal solution is very small. Therefore, the maximum number of switches required for the optimal solution in the experimental network is set to 10.

[0142] Network energy consumption comparison results:

[0143] Due to the massive and complex data flow in data center networks, this invention uses two communication modes for simulation experiments. Network energy consumption is calculated according to equation (11). To ensure the authenticity of the experimental data, this invention conducts experiments on three traffic scheduling algorithms under the two communication modes, with 10 sets of simulation experiments for each algorithm. The final experimental results are obtained by averaging the experimental data.

[0144] In Random mode, traffic scheduling was implemented using three different traffic scheduling algorithms, and the energy consumption results are shown in the attached figure. Figure 2 As shown. From Figure 2 The results show that, in terms of network energy consumption, the DE-SASDN energy-saving algorithm reduces energy consumption by 4.46%–51.47% and 2.05%–32.95% compared to the ECMP and GA-ACS energy-saving scheduling algorithms, respectively, and the average energy consumption is reduced by 25.15% and 16.18% compared to ECMP and GA-ACO, respectively.

[0145] Because the initial load is low and network energy consumption is low after algorithm optimization, network traffic peaks after 24 seconds. Although the energy-saving algorithm has been optimized, network energy consumption will still be high while meeting network load requirements. Since the DE-SASDN algorithm converges faster than the GA-ACO algorithm, it completes network flow rerouting earlier, reducing network energy consumption. This results in a phenomenon where energy consumption is high in the middle and low at both ends.

[0146] The simulation experiments were conducted in Staggered(p1,p2) mode, with p1 = 0.2 and p2 = 0.4. Ten sets of experiments were performed for each algorithm, and the average of the experimental data was taken as the final experimental data. The results are shown in the appendix. Figure 3 As shown. From Figure 3 The results show that, under the three traffic load conditions, the DE-SASDN energy-saving algorithm has lower energy consumption than the comparison algorithms. It reduces energy consumption by up to 37.42% compared to the GA-ACS energy-saving scheduling algorithm and by up to 47.79% compared to the ECMP algorithm. The average energy consumption is reduced by 18.34% and 25.70%, respectively.

[0147] Comparison results of maximum network link utilization:

[0148] To more comprehensively and specifically verify the performance of the DE-SASDN energy-saving scheduling algorithm, this invention conducted comparative experiments on the DE-SASDN energy-saving scheduling algorithm, the ECMP algorithm, and the GA-ACS energy-saving algorithm based on the maximum link utilization in the network. When network traffic is concentrated on fewer links, the maximum link utilization of the network will also increase, thereby reducing the number of devices used and achieving energy-saving effects. Therefore, the maximum link utilization is also an important indicator for evaluating energy-saving algorithms.

[0149] Maximizing link utilization can potentially cause network congestion. This invention addresses this issue by implementing the following measures:

[0150] 1) During the optimization process, a 5% buffer bandwidth is set for each link.

[0151] 2) When the optimization algorithm selects the optimal solution, it compares the current flow size with the remaining bandwidth of the path and selects a suitable path as the optimal solution.

[0152] The experiment selected the Random mode and conducted ten sets of experiments on each of the three scheduling algorithms. Each set of data streams lasted 60 seconds, with data from 20 to 40 seconds collected. The average value of the ten sets of experiments was used as the final experimental result, as shown in the attached figure. Figure 4 As shown. From Figure 4 The results show that the maximum bandwidth utilization of the DE-SASDN energy-saving algorithm network is higher than that of the ECMP and GA-ACS algorithms, with the highest improvement of 33.17% over ECMP and the highest improvement of 58.87% over GA-ACS.

[0153] In summary, under both communication modes, the DE-SASDN energy-saving algorithm exhibits lower network energy consumption than the ECMP and GA-ACO algorithms, while achieving higher maximum link bandwidth utilization. This is because the ECMP algorithm distributes flows evenly across different equivalent paths based on hash values, aiming for load balancing but neglecting energy conservation, resulting in high energy consumption. The GA-ACO energy-saving algorithm has a slow convergence speed and may not find the optimal solution within a certain number of iterations and time, leading to higher energy consumption and worse experimental data compared to DE-SASDN. The proposed DE-SASDN energy-saving algorithm first utilizes the fast-converging differential evolution algorithm to calculate multiple available paths, then inputs them into a simulated annealing algorithm that can escape local optima, thus approximating the global optimum. Furthermore, the DE-SASDN algorithm incorporates the number of active switches in the current individual into its fitness function, reducing the number of switches that need to be restarted. Therefore, the results are superior to the ECMP and GA-ACO algorithms, effectively reducing network energy consumption.

[0154] Example 2:

[0155] Example 2 provides an SDN data center network energy-saving traffic scheduling device. The device includes at least one processor and a memory communicatively connected to the processor. The memory stores instructions that can be executed by the processor. The instructions are executed by the processor to enable the processor to perform the method described in Example 1.

[0156] Example 3:

[0157] Embodiment 3 provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform the method described in Embodiment 1.

[0158] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for energy-saving traffic scheduling in an SDN data center network, characterized in that, Includes the following steps, S1: Establish an SDN data center energy consumption model; S2: Based on the differential evolutionary fusion simulated annealing algorithm, traffic optimization scheduling is performed on the SDN data center energy consumption model; S3: Based on the traffic optimization scheduling scheme, shut down or put into hibernation any unused switches.

2. The energy-saving traffic scheduling method for SDN data center networks according to claim 1, characterized in that, The SDN data center energy consumption model in step S1 is as follows: The constraints include: In the formula, the data center network topology is represented as graph G(S,L), where S is the set of all switches in the network, and s is the number of nodes. i ∈S (i=1,2,…,|S|); L represents the set of links connected to all switches in the network, and the bandwidth capacity of link l∈L is C. l Its link utilization rate is u l The set of all network flows in the network is E, and the network flow e∈E; L i and L' i These are subsets of the link set L, and switches. For link l i ∈L i The final node, For l′ i ∈L' i The starting node; the data flow passes through l i Inflow to switch Data stream passes through l′ i Inflow to switch Let E be the set of data flows in a data center network, and let b be the bandwidth of data flow e∈E. e The source and destination nodes of data stream e are represented as follows: and Indicates whether data stream e passes through link l; θ s θ l These indicate whether the switch and link are in working order, respectively; u l Indicates link utilization; u max Indicates maximum link utilization; Cost l Energy consumption of the link; Cost s This refers to the power consumption of the switch.

3. The SDN data center network energy-saving traffic scheduling method according to claim 2, characterized in that, Step S2 includes the following steps: S201: Continuously collect data streams arriving at the data center network; S202: Based on the differential evolution algorithm, each network flow data is used as the input of the differential evolution algorithm to calculate the alternative path that meets the bandwidth size of the current data flow; S203: Use the candidate paths obtained in step S202 as the initial paths of the simulated annealing algorithm, and use the simulated annealing algorithm to determine the global optimal path; S204: Repeat steps S202 and S203 until the iteration condition is met and stop. Convert the obtained global optimal path into a flow table entry and send it to each network node to reroute the data flow.

4. The energy-saving traffic scheduling method for SDN data center networks according to claim 3, characterized in that, The data flow information mentioned in step S201 includes bandwidth size, source node, destination node, and bandwidth utilization of all nodes in the network; for newly arriving data flows in the data center network, traffic scheduling is performed using the ECMP algorithm.

5. The SDN data center network energy-saving traffic scheduling method according to claim 3, characterized in that, The specific operation of step S202 includes the following steps: S2021: Differential evolution algorithm initialization, determining the initial population Pop and fitness function; S2022: Mutation operation; S2023: Crossover operation; S2024: Perform selection operation based on fitness function; S2025: Determine if the iteration condition is met; if so, exit the loop. Otherwise, repeat steps S2022 to S2024 to finally obtain a population Pop containing individuals from multiple paths.

6. The energy-saving traffic scheduling method for SDN data center networks according to claim 5, characterized in that, The specific operation of step S2021 includes the following steps: S20211: Set the loop control variable iter to 1 and set the maximum number of iterations Maxiter; S20212: Each network flow data is used as input to the differential evolution algorithm, and the alternative paths of the same destination nodes generated according to the KSP algorithm are used as the initial solution space R of the differential evolution algorithm; S20213: Select M individuals X from the solution space R. i Pop is the initial population, where each individual X i Each X is a complete, usable path that meets the constraints. i Each consists of n links l. X i ={l1,l2,…,l n }; X i ∈R,1≤i≤M S20214: Define the fitness function as follows F(X i )=dis i / (b i +u i ) Where: dis i Represents individual X i The length of , i.e., the path length; b i Represents individual X i Maximum bandwidth utilization across all links; u i Represents individual X i The number of nodes that are currently in operation.

7. The energy-saving traffic scheduling method for SDN data center networks according to claim 6, characterized in that, Step S3 includes the following steps: S301: Use the population Pop obtained in step S2 as the solution space R of the simulated annealing algorithm. sa And set the initialization parameters; S302: Randomly select M individuals x from the population Pop. i (i = 1, 2, ..., M), for each individual x i Perform a change operation to generate a new individual v′ i (i = 1, 2, ..., M); S303: Based on individual v′ i With x i The energy difference is used to determine whether to accept a new individual; S304: Cooling down, current temperature T = T1·α, where T1 is the original temperature and α is the cooling rate; S305: Determine whether the current temperature T has reached the termination temperature T. d If the target is not reached, proceed to step S302; otherwise, terminate the iteration and output the globally optimal path.

8. The SDN data center network energy-saving traffic scheduling method according to claim 6, characterized in that, The specific operation of step S301 includes the following steps: S3011: Use the population Pop obtained by the differential evolution algorithm as the solution space R. sa Set the initial temperature T0 and the final temperature T d ; S3012: From the solution space R sa Select M individuals X i Pop, as the initial population sa ; S3013: Define the energy function of an individual as follows F(X i )=(b i +u i ) / dis i S3014: The formula for determining whether to accept a new individual is as follows: where dis i Represents individual X i Length; b i Represents individual X i Maximum bandwidth utilization across all links; u i Represents individual X i The number of nodes in the active state; ΔE is the original individual energy minus the new individual energy, and T is the current temperature.

9. An energy-saving traffic scheduling device for SDN data center networks, characterized in that, The device includes at least one processor and a memory communicatively connected to the processor, the memory storing instructions executable by the processor to enable the processor to perform the method of any one of claims 1-8.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-8.

Images