Data center optical network bandwidth resource allocation method and system based on greedy genetic algorithm

Abstract

The invention discloses a data center optical network bandwidth resource allocation method and system based on a greedy genetic algorithm, and the method comprises the steps: obtaining an initial population of the genetic algorithm through the greedy algorithm, initializing configuration parameters, and building a corresponding relation between a genetic algorithm chromosome and a solution space through coding; carrying out fitness calculation on each chromosome of the obtained population; judging whether the genetic algorithm reaches a termination condition or not, if yes, ending the algorithm, and outputting an optimal solution; otherwise, carrying out chromosome selection according to a ranking selection method; randomly selecting a certain number of chromosomes from the population based on the chromosome fitness; selecting a plurality of chromosomes from the selected chromosomes to carry out crossover operation and mutation operation; performing optimal chromosome judgment and selection; then returning to the fitness calculation step to carry out iterative calculation; the method provided by the invention is suitable for a data center optical network with sudden high load flow, and has the advantage of higher bandwidth resource satisfaction compared with other schemes (such as ONU random migration and the like).

Description

Technical Field

[0001] The present invention relates to a data center optical network bandwidth resource allocation method and system, and in particular to a data center optical network bandwidth resource allocation method and system based on a greedy genetic algorithm, and belongs to the technical field of optical network communication. Background Art

[0002] Faced with new challenges brought by the diversified demands of future networks such as artificial intelligence and 5G, more and more traffic and services are pouring into data centers. Although technologies such as mobile edge computing (MEC) and fog computing have reduced the pressure on data centers in the core network, traffic is still growing rapidly, resulting in the continuous expansion of data center scale. As the scale of data centers increases, the traffic inside data centers also increases sharply. Two major problems facing data centers still need to be solved urgently. One is the huge traffic inside and between data centers, and the other is the huge energy consumed by data centers. Currently, more and more data centers use flexible passive optical network architectures to carry business data traffic. This architecture selects time-division wavelength-division multiplexing passive optical network (TWDM-PON) as the access technology to provide service flows within the data center, which can realize real-time unified and flexible management and control of server resources, while improving efficiency and reducing power consumption.

[0003] Due to the bursty and high-load nature of data center traffic, flexible bandwidth allocation in data center PON, especially bandwidth allocation under flexible ODN architecture, is a necessary demand for efficient management and control. In this technical field, there are methods to adjust the number of OLTs turned on according to real-time traffic, and use optical switches to flexibly change the association between optical network units ONU and OLT to reduce energy consumption; there are also methods to achieve load balancing between OLT ports by dynamically adjusting the allocation of wavelengths. However, these solutions do not improve the utilization rate of the overall OLT pool resources, and do not fully exploit the bandwidth fragmentation problem caused by OLT clusters under peak burst services, thereby causing insufficient bandwidth supply of data center optical network resources.

[0004] On the network access side, the diversity and dynamics of services on different racks will lead to dynamic changes in ONU bandwidth requirements, resulting in a mismatch between the bandwidth provision capacity of a single wavelength and the bandwidth requirements of the ONU, resulting in bandwidth fragmentation. When the granularity of the data stream is large, the problem of multi-OLT bandwidth fragmentation will be aggravated, resulting in reduced single-wavelength bandwidth utilization and affecting the total receiving capacity of the system. At this time, it is necessary to re-optimize the bandwidth resource allocation of multiple OLTs from the perspective of bandwidth fragmentation, while minimizing the consumption caused by bandwidth migration. This is a multi-objective optimization problem. The greedy algorithm is an algorithm idea that selects the current optimal solution at each step when solving a problem. Its basic idea is to gradually make local optimal choices in order to eventually obtain the global optimal solution. The greedy algorithm is applicable to some specific types of problems, usually requiring the problem to have greedy selection properties and optimal substructure properties. However, the greedy algorithm does not guarantee that the global optimal solution will be obtained, but in many cases it can obtain an approximate optimal solution. The genetic algorithm is an optimization algorithm based on the principle of biological evolution. It searches for the optimal solution by simulating natural selection and genetic mechanisms. The genetic algorithm evolves through the inheritance and mutation of individuals in the population to find the optimal solution to the problem.

[0005] Therefore, how to combine the advantages of the greedy algorithm and the genetic algorithm, use the genetic algorithm to screen feasible solutions according to the optimization direction and continuously optimize in a loop to obtain the optimal solution, taking advantage of its parallelism and scalability; use the greedy algorithm to solve the problem by serializing the local optimal selection, taking advantage of its simplicity, speed and space saving. In this way, the purpose of maximizing the bandwidth resource satisfaction rate and minimizing the bandwidth fragmentation migration rate is a technical problem that technicians in this field need to solve urgently. Summary of the invention

[0006] Purpose of the invention: The purpose of the present invention is to provide a data center optical network bandwidth resource allocation method and system based on a greedy genetic algorithm that can improve the carrying capacity of sudden high-load business demands.

[0007] Technical solution: The method for allocating bandwidth resources of a data center optical network based on a greedy genetic algorithm of the present invention comprises:

[0008] (1) Using the greedy algorithm to obtain the initial population of the genetic algorithm and initialize the configuration parameters, the corresponding relationship between the genetic algorithm chromosome and the solution space is established through encoding;

[0009] (2) Calculate the fitness of each chromosome of the obtained population;

[0010] (3) Determine whether the genetic algorithm has reached the termination condition. If so, the algorithm ends and outputs the optimal solution; otherwise, proceed to the next step of chromosome selection;

[0011] (4) Randomly select a certain number of chromosomes from the population based on chromosome fitness;

[0012] (5) selecting a number of chromosomes from the selected chromosomes and performing a crossover operation on them in pairs, wherein the crossover operation generates new offspring from every two chromosomes through a crossover operation of partial gene sequences, thereby generating a new population;

[0013] (6) performing a single chromosome mutation operation on the chromosome after the crossover operation, wherein the mutation operation simulates gene mutation and generates a new chromosome by modifying the gene;

[0014] (7) Determine the best chromosome. If the best chromosome in the new population is inferior to the best chromosome in history, the best chromosome in history will be retained in the population. Otherwise, the new good chromosome will replace the inferior chromosome. Then return to step (2) to calculate the fitness.

[0015] Furthermore, the step (1) comprises:

[0016] Generate the initial population, use the relationship matrix to represent the randomly generated chromosomes, each column of the chromosome is a gene, a chromosome corresponds to the solution of the optical network unit migration problem, and the topological relationship between the optical network unit and the wavelength in the network is defined as a 0-1 matrix:

[0017]

[0018] The above formula indicates that there are n wavelengths and m optical network units in the network. Represents the jth optical network unit ONU j Use the i-th wavelength λ i ,in:

[0019]

[0020] The vector of bandwidth values ​​provided by each wavelength is defined as: in, Indicates wavelength λ i The maximum bandwidth value that can be provided;

[0021] The service load value vector under each ONU is defined as: Among them, M j yes The business load value of

[0022] Assume that the optical network unit directly connected to the high-load burst service is a special optical network unit, marked as ONU t , when connected to high-load burst services, the ONU t Corresponding The load from M t Increase to Mt ′,M′ t -M t The new bandwidth requirement after connecting the peak burst service; the vector formed by the sum of the bandwidth values ​​required by all optical network units under each wavelength is Among them, B i Indicates wavelength λ i The sum of the bandwidths occupied by the multiple optical network units passed through, and

[0023] When high-load burst services arrive and some optical network units are optimized and migrated, the correspondence between optical network units and wavelengths changes to The bandwidth resource vector required at each wavelength is changed to

[0024] Assume that the wavelength bandwidth requirement overflow value is b i , which is used to indicate that the bandwidth required by a single wavelength exceeds the bandwidth that can be provided; and when B′ i ≥B i 0 When b i =B′ i -B i 0 ; When B′ i <B i 0 When b i =0;

[0025] Based on the relationship matrix representing the randomly generated chromosomes, the fitness function F = f 1 ×p+f 2 ×(1-r), calculate the fitness value of each chromosome, and arrange them from high to low, select a preset proportion of the front row chromosomes to retain, and finally randomly generate the remaining chromosomes in the population and merge them to obtain the initial population.

[0026] Further, step (2) comprises:

[0027] The fitness objective function is established with the goal of maximizing the bandwidth resource satisfaction rate and minimizing the migration rate:

[0028] F=f 1 ×p+f 2 ×(1-r)

[0029]

[0030] Among them, f 1 and f 2 is the weight coefficient, f 1 +f 2=1, p is the bandwidth resource satisfaction rate, which indicates the degree to which the optical line terminal OLT cluster can provide bandwidth resources when facing a large bandwidth burst demand; r is the ONU migration rate, which indicates the ratio of the number of ONUs migrated in the same round of ONU optimization migration to the total number of ONUs;

[0031] Set the constraints as follows:

[0032] Overflow bandwidth resource constraints of the OLT cluster:

[0033]

[0034] The bandwidth resource constraints required for each wavelength are:

[0035] 0≤B i ≤B 0 i=1,2,...,n

[0036] Special optical network unit constraints:

[0037]

[0038] Used to indicate that a special optical network unit is not migrated.

[0039] Furthermore, the iteration termination condition in step (3) includes:

[0040] Calculate the overflow value at each wavelength if This means that the bandwidth demand of the OLT cluster has not overflowed, and the algorithm terminates;

[0041] If the algorithm falls into an infinite loop and reaches the maximum number of iterations, the algorithm terminates.

[0042] Furthermore, the step (4) includes: selecting the parent chromosome set from the obtained population by a ranking selection method to generate the next generation population; the stronger the chromosome fitness, the greater the probability of being selected.

[0043] Furthermore, the step (5) comprises:

[0044] Set the parameter crossover probability to decide whether to cross the two parent chromosomes. Set the crossover probability to a fixed value and select chromosomes with a corresponding proportion of the crossover probability for crossover operation. Specifically:

[0045] Randomly select a position column of the chromosome matrix and swap the two chromosomes as a whole from the current position column to the last position column. Specifically, when the parent chromosome is The exchange position is j, then the sub-chromosome is

[0046] Furthermore, the step (6) comprises:

[0047] On the basis of the crossover operation, the mutation probability is set, and the chromosome to be mutated is determined according to the mutation probability. Specifically:

[0048] Randomly select some columns of the parent chromosome matrix and change them to the wavelength corresponding to the optical network unit access. When the parent chromosome is And select mutation position from j to k, sub-chromosome is

[0049] Furthermore, the step (7) comprises

[0050] Find b i The minimum position b in min and the maximum position b max , randomly select the wavelength b max The optical network unit is migrated to b min ; If you select ONU j , then the corresponding column vector before migration is:

[0051]

[0052] The corresponding column vector after migration is:

[0053]

[0054] Based on the same inventive concept, the present invention also provides a data center optical network bandwidth resource allocation system based on a greedy genetic algorithm, comprising:

[0055] The initialization module is used to obtain the initial population of the genetic algorithm with the help of the greedy algorithm, initialize the configuration parameters, and establish the corresponding relationship between the genetic algorithm chromosome and the solution space through coding;

[0056] A calculation module, used for calculating the fitness of each chromosome of the obtained population;

[0057] The output module is used to determine whether the genetic algorithm has reached the termination condition. If so, the algorithm ends and outputs the optimal solution; otherwise, it enters the selection module;

[0058] A selection module, used to randomly select a certain number of chromosomes from the population based on chromosome fitness;

[0059] A crossover module is used to select a number of chromosomes from the selected chromosomes and perform a crossover operation on them in pairs. The crossover operation generates new offspring from every two chromosomes and produces a new population by crossover operation of part of the gene sequence.

[0060] A mutation module is used to perform a mutation operation on a single chromosome of the chromosome after the crossover operation. The mutation operation simulates gene mutation and generates a new chromosome by modifying the gene.

[0061] The judgment module is used to judge the best chromosome. If the best chromosome in the new population is inferior to the best chromosome in history, the best chromosome in history will be retained in the population; otherwise, the new good chromosome will be used to replace the existing inferior chromosome; then the calculation module is returned to perform fitness calculation.

[0062] Based on the same inventive concept, the present invention also provides a computing device, comprising: one or more processors, one or more memories and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the data center optical network bandwidth resource allocation method based on the greedy genetic algorithm are implemented according to any one of the above items.

[0063] Beneficial effects: Compared with the prior art, the present invention constructs the bandwidth demand of data center burst high-load traffic and the migration of optical network unit bandwidth fragments into a multi-objective optimization problem, namely, maximizing bandwidth resource satisfaction and minimizing optical network unit migration consumption, and converts it into a single objective function using a linear weighted method; by introducing a greedy algorithm to help the genetic algorithm generate some good solutions in the new population, the search range of the genetic algorithm can be narrowed, the convergence speed of the genetic algorithm can be accelerated, and the execution time of the algorithm can be reduced; then the optimal solution is obtained through continuous cycle optimization of the genetic algorithm, which has the advantages of parallelism and scalability; the method provided by the present invention is suitable for data center optical networks with high burst and large-scale traffic, and has the advantage of higher bandwidth resource satisfaction compared with other solutions (such as random migration of optical network units); in view of this problem, the present invention proposes a data center optical network bandwidth resource allocation method based on a greedy genetic hybrid algorithm, which can achieve the improvement of bandwidth resource satisfaction while ensuring the migration overhead of the optical network unit, thereby increasing the carrying capacity of the data center optical network for burst high-load business needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0065] Figure 2 A relationship diagram of bandwidth resource satisfaction / optical network unit mobility / objective function value according to an embodiment of the present invention;

[0066] Figure 3 The figure is a schematic diagram showing that the bandwidth resource satisfaction rate and the optical network unit mobility rate vary with load according to an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0068] In the present invention, the migration of wavelength bandwidth fragments in the ONU cluster adopts a genetic algorithm, which establishes a group of initial feasible solutions (initialization), selects feasible solutions according to the optimization direction (selection), forms a group of new feasible solutions based on the selected feasible solutions (crossover, mutation), and selects feasible solutions according to the optimization direction for continuous cycle optimization to obtain the optimal solution, which has the advantages of parallelism and scalability. A part of the good solutions in the iterative population of the genetic algorithm is generated by a greedy algorithm, which can narrow the search range of the genetic algorithm, accelerate the convergence speed of the genetic algorithm, and reduce the execution time of the algorithm.

[0069] As attached Figure 1 As shown, the data center optical network bandwidth resource allocation method based on the greedy genetic algorithm of this embodiment includes:

[0070] (1) Using the greedy algorithm to obtain the initial population of the genetic algorithm and initialize the configuration parameters, the corresponding relationship between the genetic algorithm chromosome and the solution space is established through encoding;

[0071] (2) Calculate the fitness of each chromosome of the obtained population;

[0072] (3) Determine whether the genetic algorithm has reached the termination condition. If so, the algorithm ends and outputs the optimal solution; otherwise, proceed to the next step of chromosome selection;

[0073] (4) Randomly select a certain number of chromosomes from the population based on chromosome fitness;

[0074] (5) selecting a number of chromosomes from the selected chromosomes and performing a crossover operation on them in pairs, wherein the crossover operation generates new offspring from every two chromosomes through a crossover operation of partial gene sequences, thereby generating a new population;

[0075] (6) performing a single chromosome mutation operation on the chromosome after the crossover operation, wherein the mutation operation simulates gene mutation and generates a new chromosome by modifying the gene;

[0076] (7) Determine the best chromosome. If the best chromosome in the new population is inferior to the best chromosome in history, the best chromosome in history will be retained in the population. Otherwise, the new good chromosome will replace the inferior chromosome. Then return to step (2) to calculate the fitness.

[0077] Specifically, in step (1), the initial population and configuration parameters of the genetic algorithm are initialized;

[0078] The topological relationship between the optical network unit (ONU) and the wavelength in the network is defined as a 0-1 matrix: That is, there are n wavelengths and m ONUs in the network. Device ONU j Use wavelength λ i It can be expressed as: In the embodiment of the present invention, it is assumed that there are 20 wavelengths and 80 ONU devices in the optical network.

[0079] The vector of bandwidth values ​​provided by each wavelength is defined as: in, Indicates wavelength λ i The maximum bandwidth value that can be provided;

[0080] The service load value vector under each ONU is defined as: Among them, M j It's ONU j business load value.

[0081] After connecting to the peak burst service, the ONU load will change dramatically, from M t Increase to M t ′,M t ′ -M t is the new bandwidth requirement. The vector formed by the sum of the bandwidth values ​​required by all ONUs at each wavelength is defined as:

[0082] When a burst of high-load services arrives and some ONUs are optimized and migrated, the correspondence between ONUs and wavelengths changes to Accordingly, the bandwidth resource vector required at each wavelength is changed to

[0083] Assume that the wavelength bandwidth requirement overflow value is b i , which is used to indicate that the bandwidth required by a single wavelength exceeds the bandwidth available; and when B i ′ ≥B i 0 When b i =B′ i -B i 0 ; When B′ i <B i 0 When b i =0.

[0084] The initialization population is encoded to establish the correspondence between the GA chromosome and the solution space; the relationship matrix is ​​used as the chromosome, each column of the chromosome is called a gene, and the chromosome is a method to solve the ONU migration problem.

[0085] like Figure 1 As shown, a greedy algorithm is used to randomly generate a certain proportion of individuals in the population. In this example, 30% of the individuals are selected, and then the remaining individuals in the population are randomly generated, and finally the complete population is obtained by merging. Specifically:

[0086] Based on the relationship matrix representing the randomly generated chromosomes, the fitness function (y = F = f 1 ×p+f 2 ×(1-r),y∈[0,1]) calculate the fitness value of each chromosome. Then arrange them from high to low. Select the top 30% chromosomes to keep. The local optimal solution is calculated by the greedy algorithm. Finally, randomly generate (20*80)*70% chromosomes as the remaining random samples.

[0087] In summary, the 30% obtained by the greedy algorithm plus the 70% randomly generated constitute the initial population of the entire algorithm. This not only ensures the diversity of the initial population but also speeds up the convergence of the algorithm. It has a compromise effect.

[0088] Step (2) includes calculating the fitness of each chromosome of the initial population or the iterative population;

[0089] Fitness can measure the quality of chromosomes. The smaller the fitness, the worse the chromosome's viability and the further away from the optimal solution. The present invention defines the ONU optimization migration problem as a multi-objective optimization problem that needs to maximize the bandwidth resource satisfaction rate and minimize the mobility rate, and uses a linear weighted method to convert it into a single objective function. The formula is as follows:

[0090] F=f 1 ×p+f 2 ×(1-r)

[0091]

[0092]

[0093] Among them, p is the bandwidth resource satisfaction rate, which is used to measure the degree to which the OLT cluster can provide bandwidth resources when facing large bandwidth burst demands. r is the ONU migration rate, which represents the ratio of the number of ONUs migrated in the same round of ONU optimization migration to the total number of ONUs, reflecting the overhead caused by migration.

[0094] The relevant constraints are as follows:

[0095] In an embodiment of the present invention, f 1 ,f 2 is the weight coefficient, f 1 +f 2 =1;

[0096] Since the sum of overflow bandwidth resources of the OLT cluster cannot be higher than the total number of newly added bandwidth resources, the following constraints apply:

[0097]

[0098] Because before the service arrives, the bandwidth resources required by any wavelength cannot be greater than the bandwidth resources provided by the wavelength, so the constraint is: 0≤B i ≤B 0 i=1,2,...,n.

[0099] In addition, special ONUs will not be migrated, so the constraints are:

[0100] In step (3), it is determined whether the genetic algorithm meets the termination condition, that is, whether the maximum number of iterations is reached. If it is satisfied, the algorithm ends and the optimal solution is output; if not, it goes to step (4) to continue the calculation.

[0101] The conditions for the algorithm to terminate include the following:

[0102] Calculate the difference between the bandwidth resources required for each wavelength and the bandwidth resources available at the current wavelength if This means that the bandwidth demand of the OLT cluster has not overflowed, and the algorithm terminates;

[0103] If the algorithm falls into an infinite loop and reaches the maximum number of iterations, which is set to 100 in this example, the algorithm terminates.

[0104] In step (4), a certain number of individuals are randomly selected from the population based on their individual fitness to perform crossover and mutation operations.

[0105] The parent chromosome set is selected from the old population through ranking selection to generate the next generation population. The stronger the chromosome fitness, the greater the probability of being selected.

[0106] Step 500: Generate new offspring from two selected individuals through a crossover operation of partial gene sequences. These offspring are genetically different from the original individuals, thereby generating a new population.

[0107] Generate new chromosomes by exchanging genes, that is, generate new individuals; decide whether to cross two parent chromosomes based on the crossover probability. According to the individual fitness function y = F = f 1 ×p+f 2×(1-r),y∈[0,1] calculates the fitness value of each individual in the population and ranks them from high to low, which is the "ranking selection method of genetic algorithm". At this time, the "crossover probability" is introduced. In this method, the crossover probability is set to a fixed value. The specific value depends on the size of the population and the maximum number of iterations. The general range is between (0.5 and 0.9). For example, it is set to 0.7, then the top 70% of individual chromosomes are selected, and two of them are randomly selected for crossover operation. The specific crossover operation details are:

[0108] Randomly select a position column of the chromosome matrix and swap the two chromosomes as a whole from the current position column to the last position column. For example, the parent chromosome is The exchange position is j, then the sub-chromosome is

[0109] In this embodiment, the number of wavelengths is 20, the number of ONUs is 80, and with such a population size, the maximum iteration is 100 times, and the selection crossover probability is 0.7.

[0110] In step (6), gene mutation is simulated and new chromosomes are generated by modifying genes, which can increase the diversity of the population and prevent the algorithm from falling into local extreme points.

[0111] By modifying genes to generate new chromosomes to simulate gene mutations, the diversity of the population can be increased and the algorithm can be prevented from falling into local extreme points. Determine whether to mutate the parent chromosome based on the mutation probability. On the basis of the crossover operation in the previous step, based on the mutation probability (the mutation probability depends on the size of the population, large scale with low probability, small scale with high probability, the purpose is to ensure population diversity. The general range is set at 0.01~0.1, and the specific adjustment needs to be made through experiments), select a part of the individuals for gene mutation operations. The main operation is to randomly select several position columns of the individual chromosomes and randomly move 1 to other different positions. The actual meaning is to change the wavelength corresponding to the ONU access. Randomly select certain columns of the parent chromosome matrix and change the position of "1" in the selected column vector (that is, change the wavelength corresponding to the ONU access). For example, the parent chromosome is And select mutation position from j to k, sub-chromosome is

[0112] In the embodiment, the number of wavelengths is 20, the number of ONUs is 80, and the population size is 100 times at most. Through various experimental attempts and adjustments, the mutation probability is positioned at 0.1, which has the best effect.

[0113] In step (7), the best individual is retained, that is, if the best individual in the new group is inferior to the best individual in history, the best individual in history is retained in the group; otherwise, the new excellent individual is used to replace the existing inferior individual; to ensure that the optimal solution of each generation of chromosomes is not inferior to that of the previous generation, go to step (2).

[0114] The specific optimization method is: find b i The minimum position b in min and the maximum position b max , randomly select the wavelength b max The optical network unit is migrated to b min ; If you select ONU j , then the corresponding column vector before migration is:

[0115]

[0116] The corresponding column vector after migration is:

[0117]

[0118] In this embodiment, the line overhead of the passive optical network makes the actual effective load rate lower than the line rate, but this feature does not affect the problem characteristics and algorithm performance. Therefore, the present invention does not analyze the overhead separately, but refers to the service acceptance capacity provided by a specific wavelength through bandwidth resource satisfaction. In this embodiment, the single wavelength bandwidth resource is set to 10Gbit / s, and the background traffic load of the server rack flow is defined as the ratio of the bandwidth required by the ONU in the network before the server request arrives to the sum of the bandwidth that the OLT cluster can provide.

[0119] The objective function of the present invention is a multi-objective function, in which the bandwidth resource satisfaction reflects the quality of the network status, and the ONU migration rate reflects the size of the overhead caused by the migration. 1 ,f 2 The values ​​of the weight coefficients represent the bandwidth resource satisfaction rate and the weight of the ONU migration. Figure 2 The number of wavelengths is 20, the number of ONUs is 80, and the load is 0.94 (heavy load condition). The total bandwidth demand brought by the server is about 90% of the bandwidth resources owned by a single wavelength, that is, 9 Gbit / s.

[0120] Weight coefficient f 1 The change of will affect the bandwidth resource satisfaction rate p, ONU mobility rate r and objective function value F.

[0121] With the f 1 As the weight of bandwidth resource satisfaction increases, the larger the p is, the better the network resource status is. At the same time, the smaller the weight of ONU mobility is, the larger r is, and the greater the migration loss is.

[0122] When f 1 When f decreases, r has a greater impact on F among the two sub-goals. 1 increases, r increases, and F decreases; when f 1 When it is larger, p has a greater impact on F. 1 As f increases, p increases, and F increases. The total objective function value F increases with f 1 It decreases first and then increases with the increase of 1 When the bandwidth resource satisfaction and ONU mobility rate are between 0.5 and 0.9, both can reach relatively good values. 1 =0.8,f 2 =0.2 to evaluate the effect of algorithm optimization.

[0123] The schematic diagram of the bandwidth resource satisfaction rate and ONU migration rate of the ONU optimization migration mechanism based on the greedy genetic hybrid algorithm changing with load is shown in the figure. Figure 3 As shown. As the load increases, the bandwidth resource satisfaction shows a downward trend. Because as the load increases, the remaining bandwidth in the network will decrease, and the resources used to accommodate non-special ONUs will also decrease, reducing the bandwidth resource satisfaction. For the migration mechanism described in the present invention, when the load is low (such as lower than 0.94), the migration mechanism of the present invention can be directly used to make the network reach a good state; when the load is high (such as higher than 0.94), as the load increases, the bandwidth resource satisfaction decreases significantly, indicating that defragmentation is also difficult to provide sufficient bandwidth for non-special ONUs. At this time, wavelength sharing should be considered as a system to increase wavelengths to reduce the impact of overload.

[0124] Based on the same inventive concept, this embodiment also provides a data center optical network bandwidth resource allocation system based on a greedy genetic algorithm, including:

[0125] The initialization module is used to obtain the initial population of the genetic algorithm with the help of the greedy algorithm, initialize the configuration parameters, and establish the corresponding relationship between the genetic algorithm chromosome and the solution space through coding;

[0126] A calculation module, used for calculating the fitness of each chromosome of the obtained population;

[0127] The output module is used to determine whether the genetic algorithm has reached the termination condition. If so, the algorithm ends and outputs the optimal solution; otherwise, it enters the selection module;

[0128] A selection module, used to randomly select a certain number of chromosomes from the population based on chromosome fitness;

[0129] A crossover module is used to select a number of chromosomes from the selected chromosomes and perform a crossover operation on them in pairs. The crossover operation generates new offspring from every two chromosomes and produces a new population by crossover operation of part of the gene sequence.

[0130] A mutation module is used to perform a mutation operation on a single chromosome of the chromosome after the crossover operation. The mutation operation simulates gene mutation and generates a new chromosome by modifying the gene.

[0131] The judgment module is used to retain the best chromosome. If the best chromosome in the new population is inferior to the best chromosome in history, the best chromosome in history will be retained in the population; otherwise, the new excellent chromosome will replace the inferior chromosome, and the calculation module will be returned to perform fitness calculation.

[0132] Based on the same inventive concept, this embodiment also provides a computing device, including: one or more processors, one or more memories and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the data center optical network bandwidth resource allocation method based on the greedy genetic algorithm are implemented according to any one of the above items.

[0133] Based on a flexible ODN architecture and from the perspective of bandwidth fragmentation, the present invention proposes an ONU optimization migration and bandwidth resource allocation mechanism based on a greedy genetic hybrid algorithm, which solves the access problem of sudden large-bandwidth services on server racks, reduces the delay caused by service congestion, and improves the utilization efficiency of bandwidth resources.

Claims

1. A data center optical network bandwidth resource allocation method based on greedy genetic algorithm, characterized in that: include: (1) Using the greedy algorithm to obtain the initial population of the genetic algorithm and initialize the configuration parameters, the corresponding relationship between the genetic algorithm chromosome and the solution space is established through encoding; (2) Calculate the fitness of each chromosome of the obtained population; (3) Determine whether the genetic algorithm has reached the termination condition. If so, the algorithm ends and outputs the optimal solution; otherwise, proceed to the next step of chromosome selection; (4) Randomly select a certain number of chromosomes from the population based on chromosome fitness; (5) selecting a number of chromosomes from the selected chromosomes and performing a crossover operation on them in pairs, wherein the crossover operation generates new offspring from every two chromosomes through a crossover operation of partial gene sequences, thereby generating a new population; (6) performing a single chromosome mutation operation on the chromosome after the crossover operation, wherein the mutation operation simulates gene mutation and generates a new chromosome by modifying the gene; (7) Determine the best chromosome. If the best chromosome in the new population is inferior to the best chromosome in history, the best chromosome in history will be retained in the population. Otherwise, the new good chromosome will replace the inferior chromosome. Then return to step (2) to calculate the fitness.

2. The data center optical network bandwidth resource allocation method based on greedy genetic algorithm according to claim 1 is characterized in that: The step (1) comprises: Generate the initial population, use the relationship matrix to represent the randomly generated chromosomes, each column of the chromosome is a gene, a chromosome corresponds to the solution of the optical network unit migration problem, and the topological relationship between the optical network unit and the wavelength in the network is defined as a 0-1 matrix: The above formula indicates that there are n wavelengths and m optical network units in the network. Represents the jth optical network unit ONU j Use the i-th wavelength λ i ,in: The vector of bandwidth values ​​provided by each wavelength is defined as: in, Indicates wavelength λ i The maximum bandwidth value that can be provided; The service load value vector under each optical network unit is defined as: Among them, M j It's ONU j The business load value; Assume that the optical network unit directly connected to the high-load burst service is a special optical network unit, marked as ONU t , when connected to high-load burst services, the ONU t Corresponding The load from M t Increase to M t ′,M t ′ -M t The new bandwidth requirement after connecting the peak burst service; the vector formed by the sum of the bandwidth values ​​required by all optical network units under each wavelength is Among them, B i Indicates wavelength λ i The sum of the bandwidths occupied by the multiple optical network units passed through, and When high-load burst services arrive and some optical network units are optimized and migrated, the correspondence between optical network units and wavelengths changes to The bandwidth resource vector required at each wavelength is changed to Assume that the wavelength bandwidth requirement overflow value is b i , which is used to indicate that the bandwidth required by a single wavelength exceeds the bandwidth available; and when B i ′ ≥B i 0 When b i =B i ′ -B i 0 When B i ′ i 0 When b i =0;​ Based on the relationship matrix representing the randomly generated chromosomes, the fitness value of each chromosome is calculated through the fitness function F=f1×p+f2×(1-r), and they are arranged from high to low. A preset proportion of the front row chromosomes are selected for retention, and finally the remaining chromosomes in the population are randomly generated and merged to obtain the initial population.

3. The data center optical network bandwidth resource allocation method based on greedy genetic algorithm according to claim 2 is characterized in that: Step (2) comprises: The fitness objective function is established with the goal of maximizing the bandwidth resource satisfaction rate and minimizing the migration rate: F=f1×p+f2×(1-r) Among them, f1 and f2 are weight coefficients, f1+f2=1, p is the bandwidth resource satisfaction rate, which indicates the degree to which the optical line terminal OLT cluster can provide bandwidth resources when facing large bandwidth burst demand; r is the optical network unit migration rate, which indicates the ratio of the number of optical network units migrated in the same round of optical network unit optimization migration to the total number of optical network units; Set the constraints as follows: Overflow bandwidth resource constraints of the OLT cluster: The bandwidth resource constraints required for each wavelength are: 0≤B i ≤B 0 i=1,2,...,n Special optical network unit constraints: Used to indicate that a special optical network unit is not migrated.

4. The data center optical network bandwidth resource allocation method based on greedy genetic algorithm according to claim 3 is characterized in that: The iteration termination conditions in step (3) include: Calculate the overflow value at each wavelength If b i <0, This means that the bandwidth demand of the OLT cluster has not overflowed, and the algorithm terminates; If the algorithm falls into an infinite loop and reaches the maximum number of iterations, the algorithm terminates.

5. The data center optical network bandwidth resource allocation method based on greedy genetic algorithm according to claim 4 is characterized in that: The step (4) comprises: selecting the parent chromosome set from the obtained population by ranking selection method to generate the next generation population; the stronger the chromosome fitness, the greater the probability of being selected.

6. The method for allocating bandwidth resources of a data center optical network based on a greedy genetic algorithm according to claim 5, characterized in that: The step (5) comprises: Set the parameter crossover probability to decide whether to cross the two parent chromosomes. Set the crossover probability to a fixed value and select chromosomes with a corresponding proportion of the crossover probability for crossover operation. Specifically: Randomly select a position column of the chromosome matrix and swap the two chromosomes as a whole from the current position column to the last position column. Specifically, when the parent chromosome is The exchange position is j, then the sub-chromosome is 7. The data center optical network bandwidth resource allocation method based on greedy genetic algorithm according to claim 6, characterized in that: The step (6) comprises: On the basis of the crossover operation, the mutation probability is set, and the chromosome to be mutated is determined according to the mutation probability. Specifically: Randomly select some columns of the parent chromosome matrix and change them to the wavelength corresponding to the optical network unit access. When the parent chromosome is And select mutation position from j to k, sub-chromosome is 8. The method for allocating bandwidth resources of a data center optical network based on a greedy genetic algorithm according to claim 7, characterized in that: The step (7) comprises Find b i The minimum position b in min and the maximum position b max , randomly select the wavelength b max The optical network unit is migrated to b min ; If you select ONU j , then the corresponding column vector before migration is: The corresponding column vector after migration is:

9. A data center optical network bandwidth resource allocation system based on greedy genetic algorithm, characterized in that: include: The initialization module is used to obtain the initial population of the genetic algorithm with the help of the greedy algorithm, initialize the configuration parameters, and establish the corresponding relationship between the genetic algorithm chromosome and the solution space through coding; A calculation module, used for calculating the fitness of each chromosome of the obtained population; The output module is used to determine whether the genetic algorithm has reached the termination condition. If so, the algorithm ends and outputs the optimal solution; otherwise, it enters the selection module; A selection module, used to randomly select a certain number of chromosomes from the population based on chromosome fitness; A crossover module is used to select a number of chromosomes from the selected chromosomes and perform a crossover operation on them in pairs. The crossover operation generates new offspring from every two chromosomes and produces a new population by crossover operation of part of the gene sequence. A mutation module is used to perform a mutation operation on a single chromosome of the chromosome after the crossover operation. The mutation operation simulates gene mutation and generates a new chromosome by modifying the gene. The judgment module is used to judge the best chromosome. If the best chromosome in the new population is inferior to the best chromosome in history, the best chromosome in history will be retained in the population; otherwise, the new good chromosome will be used to replace the existing inferior chromosome; then the calculation module is returned to perform fitness calculation.

10. A computing device, characterized in that include: One or more processors, one or more memories, and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the method for allocating bandwidth resources of a data center optical network based on a greedy genetic algorithm are implemented according to any one of claims 1 to 8.

Images