CN106484512B - The dispatching method of computing unit - Google Patents
The dispatching method of computing unit Download PDFInfo
- Publication number
- CN106484512B CN106484512B CN201610875838.4A CN201610875838A CN106484512B CN 106484512 B CN106484512 B CN 106484512B CN 201610875838 A CN201610875838 A CN 201610875838A CN 106484512 B CN106484512 B CN 106484512B
- Authority
- CN
- China
- Prior art keywords
- computing unit
- individual
- physical machine
- scheduling
- genetic algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 230000002068 genetic effect Effects 0.000 claims abstract description 33
- 238000005457 optimization Methods 0.000 claims abstract description 28
- 238000013499 data model Methods 0.000 claims abstract description 3
- 108090000623 proteins and genes Proteins 0.000 claims description 19
- 230000035772 mutation Effects 0.000 claims description 18
- 238000011156 evaluation Methods 0.000 claims description 14
- 238000013507 mapping Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 230000001174 ascending effect Effects 0.000 claims description 3
- 238000009472 formulation Methods 0.000 abstract 1
- 239000000203 mixture Substances 0.000 abstract 1
- 230000006870 function Effects 0.000 description 19
- 210000000349 chromosome Anatomy 0.000 description 18
- 239000002699 waste material Substances 0.000 description 11
- 238000005265 energy consumption Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 201000004569 Blindness Diseases 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
- G06F9/5088—Techniques for rebalancing the load in a distributed system involving task migration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention relates to the dispatching methods of computing unit, comprising: S1. inputs physical machine resource；S2. the computing unit resource of user demand is inputted；S3. formulation description is carried out to optimization aim by data model；S4. a kind of computing unit deployment scheme is obtained using group genetic algorithm, it is overall minimum that the program makes power consumption, wasting of resources situation, SLA violate rate；S5. computing unit deployment scheme corresponding with optimal solution is exported.The dispatching method of computing unit of the present invention solves the uncertain problem of multiple optimization aim influence degrees during multiple-objection optimization by improved group genetic algorithm, has more practicability compared to single object optimization.The convergence rate that scheduling calculates obviously is accelerated, makes server cluster while significantly energy saving, also ensures running quality, significantly improve the harmony and computational efficiency of server resource.
Description
Technical Field
The invention relates to a computing unit scheduling technology under a cloud platform and an artificial intelligence scheduling method, in particular to a scheduling method of a computing unit capable of simultaneously optimizing multiple targets.
Background
The computing unit scheduling in distributed computing refers to mapping a group of computing units applied by a user to a physical machine (i.e. a server or a physical node) according to a certain scheduling algorithm, and simultaneously meeting necessary constraint conditions. The method comprises the steps of electric energy consumption, resource utilization rate, user experience, cloud provider benefits and the like caused by different mappings of a cloud data center computing unit and a physical machine. Therefore, it is very important to design an efficient scheduling algorithm. With the continuous enlargement of the scale of data centers and the rapid increase of the consumed electric energy, more and more researches focus on the energy consumption saving of the scheduling problem of the computing units, the existing mode is generally realized by a mode of aggregating servers, the computing units are placed on a small number of physical machines through a genetic algorithm, and the number of the activated physical machines is minimized to achieve the aim of saving energy. This approach is indeed effective in terms of energy saving. However, when the tasks on the physical machine are too aggregated, the physical machine may be overloaded and the user application performance may be degraded, resulting in a poor user experience. Therefore, the user quality of service cannot be ignored while considering saving power consumption. Meanwhile, the method improves the balanced utilization of the physical machine resources, and is an effective way for improving the efficiency of the data center and reducing the resource waste.
At present, the genetic algorithm solves the scheduling problem of the computing unit and has at least the following disadvantages: 1) the performance of the algorithm depends on the selection of parameters, and the quality of the solution is seriously influenced by improper parameter selection. 2) The cross probability and the variation probability are fixed and unchanged in the population evolution process, the convergence of the population is seriously influenced, and premature is easily caused to be incapable of reaching the global optimal solution. 3) In the algorithm, genes on chromosomes are randomly selected by cross operation and mutation operation, so that the blindness is realized, and the convergence speed of the algorithm is influenced. 4) When the fitness function is used for evaluating the multi-objective optimization problem, the linear summation of a plurality of objectives is converted into a single-objective problem, and the fitness function is not consistent with the fact that the influence degree of the plurality of objectives on the problem is uncertain in practice.
Disclosure of Invention
The invention provides a scheduling method of a computing unit, which is used for optimizing various operation data of a server for cloud data computing, so that a server cluster can ensure the operation quality while saving energy consumption, improve the resource balance and the computing efficiency of the server, and has higher practicability compared with the existing single-target optimization.
The invention discloses a scheduling method of a computing unit, which comprises the following steps:
s1, inputting physical machine resources: determining the number of the physical machines, the types of resources owned by the physical machines and the capacity of the resources, wherein the resources, such as a CPU (central processing unit), a memory, a bandwidth, a storage and the like, can be regarded as resources owned by the physical machines;
s2, inputting computing unit resources required by a user;
s3, performing formulaic description on the optimization target through a data model;
s4, scheduling the computing unit through a grouping genetic algorithm: firstly, determining a gene coding mode, then generating a plurality of groups of computing unit deployment schemes by a random method, enabling each computing unit deployment scheme to correspond to one individual in a packet genetic algorithm, converting a mapping relation between a computing unit and a physical machine into a gene code in the packet genetic algorithm by using the packet coding mode, and enabling all the individuals corresponding to the computing unit deployment schemes to form an initial population in the packet genetic algorithm; calculating the fitness value of each individual in the initial population, selecting the individual inherited to the next generation according to the selection probability of each individual, performing cross operation and variation operation on the corresponding physical machine according to the cross probability and the variation probability of each individual, and then generating a progeny population, wherein the individual with the minimum fitness value in the progeny population is the optimal solution of the grouping genetic algorithm;
and S5, outputting a calculation unit deployment scheme corresponding to the optimal solution.
Through improvement of the existing grouping genetic algorithm, the computing unit is more balanced in scheduling of the physical machine, the overall energy consumption of the whole physical machine cluster is minimized, and good computing efficiency is guaranteed.
Further, the computing unit resource described in step S2 includes the requested number of computing units and the resource type and resource capacity of each computing unit. Because network attached storage is usually adopted in distributed computing, only the CPU and the memory of a physical machine are concerned in the resource category, and no disk is considered, but corresponding adjustment can be made according to actual situations.
On the basis of the above, a preferred method is to calculate the fitness value of each individual in the initial population in a manner of maximizing the minimum satisfaction through the fuzzy logic theory in step S4. And the selection probability of each individual can be calculated through a roulette algorithm, the cross probability and the variation probability of each individual are determined through an adaptive algorithm, the evaluation value of each physical machine in each individual is calculated, the evaluation values are arranged in an ascending order, and the physical machine with the largest evaluation value is selected to perform cross operation and variation operation.
The fitness value of each individual is calculated according to a fitness function, the fitness function is a selection standard of high or low in a genetic algorithm, in order to overcome the uncertainty of the influence degree of a plurality of optimization targets on the problem to be solved in practical application, the plurality of optimization targets are integrated and quantitatively evaluated through a fuzzy logic theory, and a maximum and minimum satisfaction fuzzy logic model is adopted to perform fuzzification processing on the three optimization targets. And then, the genetic algorithm is used for simulating the biological evolution principle to adopt selection operation to eliminate the individuals, the probability that the individuals with high fitness are transmitted to filial generations is higher, and the probability that the individuals with low fitness are transmitted to the filial generations is lower. However, if only the individuals with high fitness are selected for inheritance, the local optimal solution is involved, so the invention adopts the roulette algorithm to select the inherited individuals.
In order to optimize the final scheduling scheme, after the child population is generated in step S4, it is first determined whether the current iteration number reaches the maximum number set by the algorithm, and if so, the iteration is stopped, and step S5 is executed; and if the maximum number is not reached, returning to the individual step of selecting inheritance to the next generation for loop execution.
Specifically, the optimization objectives described in step S3 include minimizing power consumption of the physical machine, minimizing resource utilization, and minimizing SLA (service level agreement) violation rate, and may also include other corresponding optimization objectives according to actual requirements.
The scheduling method of the computing unit solves the problem of uncertainty of the influence degree of a plurality of optimization targets in the multi-target optimization process through an improved grouping genetic algorithm, and has higher practicability compared with single-target optimization. The convergence rate of scheduling calculation is obviously accelerated, so that the server cluster ensures the operation quality while greatly saving energy consumption, and the balance and the calculation efficiency of server resources are obviously improved.
The present invention will be described in further detail with reference to the following examples. This should not be understood as limiting the scope of the above-described subject matter of the present invention to the following examples. Various substitutions and alterations according to the general knowledge and conventional practice in the art are intended to be included within the scope of the present invention without departing from the technical spirit of the present invention as described above.
Drawings
FIG. 1 is a flow chart of a method for scheduling computing units according to the present invention.
FIG. 2 is a flow chart of the packet genetic algorithm of FIG. 1.
Fig. 3 is a schematic diagram of a block coding scheme according to the present invention.
Detailed Description
As shown in fig. 1, the method for scheduling a computing unit of the present invention includes the steps of:
s1, inputting physical machine resources including the number of physical machines of the cloud data center, and recording the types of available resources (such as CPU, memory, bandwidth, storage and the like) of any physical machine, wherein H is the number of the available resourcesi,jIndicates the ith physical machine PiThe capacity of the j-type resource.
S2, inputting computing unit resources applied by a user, including the number of computing units, and recording the required resources of any computing unit as Ri,jDenotes the ith calculation Unit ViThe number of demands on the j-th class of resources. In the invention, only CPU (processor) and memory are concerned in the resources applied by the computing unit, and no disk is considered, because network-attached storage (NAS) is generally adopted in distributed computing, the storage can be used as a single module.
And S3, formulaic modeling. The purpose of the embodiment is to implement balanced and reasonable mapping of a group of computing units onto a physical machine of a cloud data center through a computing unit scheduling algorithm, so that the goals of minimizing electric energy consumption, minimizing resource waste and minimizing SLA (service level agreement) violation rate are achieved while the resource constraint of the physical machine is met. Before designing a scheduling algorithm, three optimization targets need to be described in a formula, which is specifically as follows:
s31 power consumption: the electric energy consumption refers to the electric energy consumption of a physical machine of a data center, wherein the influence of a CPU on the power consumption of the physical machine occupies most of the energy consumption, the invention only considers the influence of the CPU on the power consumption, and according to the known literature, when the physical machine changes from a zero load (0%) to a full load (100%), the power consumption of the physical machine and the CPU use form a linear relation, while the electric energy consumed by the physical machine at the zero load is 67% of the electric energy consumed by the physical machine at the full load, and the power consumption P of the physical machine is PiCan be calculated as:wherein, PmaxRepresenting the power consumption of the physical machine at full load, PidleRepresenting the power consumption when the physical machine is idle,the CPU utilization rate of the ith physical machine.
S32, resource utilization rate: the utilization rates of various resources (such as resources such as CPU, internal memory, storage space and the like) of each physical machine of the data center should be kept balanced as much as possible, so that resource waste caused by a barrel effect is avoided, the resource utilization rate is reduced, and the resource waste rate R of the physical machine is reducediExpressed as:
wherein,the ratio of the residual CPU resource to the total CPU resource of the ith physical machineThe value, i.e., the CPU remaining rate,representing the residual memory rate of the physical machine, the value range of epsilon is 0-0.001,for the CPU utilization of the physical machine,the memory utilization rate of the physical machine.
S33, SLA violation rate: in distributed computing, a cloud provider needs to sign an SLA with a user, where matters such as service quality and user payment are defined, the service quality allocated by a computing unit is generally considered by user application performance, and the cloud provider needs to allocate resources equivalent to requirements to the user to ensure the performance. The SLA violation rate is defined as a function of CPU utilization, and an evaluation function f of SLA violation rateiExpressed as:wherein,for the CPU utilization of the ith physical machine,is the maximum threshold for CPU utilization.
The optimization objective of the computing unit scheduling is expressed as: minimizing power consumptionMinimizing waste of resourcesMinimizing SLA breach rate
And S4, as shown in the figure 2, solving the multi-objective optimized computing unit scheduling problem by adopting an improved grouping genetic algorithm. The method comprises the following steps:
s41, gene coding. And adopting a block coding mode by combining the mapping characteristics between the computing unit and the physical machine in the computing unit scheduling problem. The physical machine is regarded as a gene of a chromosome in the genetic algorithm, and the calculation unit disposed in the physical machine is regarded as a value of the gene. For example, in the block coding scheme shown in fig. 3, it is assumed that 7 computing units are deployed on 3 physical machines, one physical machine represents the code of one chromosome, and one computing unit number represents the gene value on the corresponding chromosome. When setting chromosome coding, in addition to the constraint condition that needs to be satisfied for solving the problem, it must be satisfied that any one gene can only appear on one chromosome, all genes must be mapped on the chromosome, but chromosomes without genes can appear. For example, the code of the chromosome in FIG. 3 is ABC, and the code mode of the chromosome is set as A according to the above principle: 123, B: 45, C: 67. therefore, the problem of ambiguity of the packet information in the traditional coding mode can be solved, namely, when the packet information is crossed and changed, the packet information can be operated on the computing unit, and the server information is transparent. In the conventional binary coding method, each individual is represented by a string of binary numbers with fixed length, and the operation can only be performed by taking the individual as a unit when the intersection and mutation operations are performed, and the structural characteristics and specific information of the required problem cannot be reflected.
And S42, generating an initial population. And generating a certain number of computing unit deployment schemes by adopting a random algorithm, namely randomly mapping the computing units requested by the user to the physical machine, wherein the precondition is that the resource capacity of the physical machine is necessarily greater than the resource capacity requested by the computing units. Converting the mapping relation between the computing units and the physical machine into gene codes in the packet genetic algorithm according to the packet coding mode of S41, wherein each computing unit deployment scheme corresponds to one individual in the algorithm, and all the individuals corresponding to all the computing unit deployment schemes form an initial population of the packet genetic algorithm;
s43, calculating the fitness value. The fitness value of an individual is calculated according to a fitness function, the fitness function is a selection standard of the genetic algorithm with high or low quality, in order to overcome the uncertainty of the influence degree of a plurality of optimization targets on the problem in the actual problem, the plurality of optimization targets are integrated and quantitatively evaluated through a fuzzy logic theory, and a maximum and minimum satisfaction fuzzy logic model is adopted to fuzzify the three optimization targets. The method comprises the following specific steps:
and T431, solving the optimal solution of each single target, wherein in order to determine whether the final solution performs well in each optimized target, the optimal solution of each single target needs to be solved firstly.
First, the minimum value of the number of physical machines is designated as Lmin。
Wherein R isall-cpu，Rall-memRespectively representing the CPU resource sum and the memory resource sum applied by all computing units of the data center, Hcpu，HmemRespectively representing the CPU resource and the memory resource owned by a single physical machine. The maximum number of physical machines is denoted LmaxN, where N represents the number of computational units.
The minimum value of the power consumption is expressed as: w*＝Lmin·Pmax(ii) a The maximum value of the power consumption is expressed as: wmax＝Lmax·Pmax，PmaxRepresenting the power consumption when the physical machine is fully loaded.
The minimum value of resource waste is expressed as:wherein Hcpu-all，Hmem-allRespectively representing the sum of the CPU resources and the sum of the memory resources of all the physical machines of the data center. The maximum value of resource waste is expressed as:wherein R isi,cpu，Ri,memRespectively representing CPU resources and memory resources applied by the ith computing unit, Hi,cpu，Hi,memRespectively representing the CPU resource and the memory resource of the ith physical machine.
The upper limit and the lower limit of the SLA violation rate are based on an evaluation function f of the SLA violation rateiAnd (6) determining.
And T432. constructing a membership function. The membership function reflects the degree of goodness of the optimization objective, and a smaller membership indicates that the solution is more unacceptable, a value of 0 indicates that the solution is completely unacceptable, and a value of 1 indicates an ideal value, which is generally impossible.
The membership functions for the three optimization objectives are represented as:
wherein, mu (f)i(x) A membership value representing the ith optimization objective, x represents a deployment scenario of the computing unit, W x + delta1Is the maximum value of the power consumption, in the same way, R + delta2，F*+δ3Maximum value representing resource waste and maximum value of SLA violation, δiRepresenting the difference between the ideal value (minimum) and the maximum value, F is the minimum value of the SLA violation.
And T433, constructing a fitness function. Fitness function in genetic algorithm determines the direction of population evolution, thereforeWhether this function is suitable directly determines whether the final solution of the algorithm is the optimal solution. And determining a fitness function by adopting a maximum minimum satisfaction method, wherein the method ensures that all target functions have the highest possible membership, and the fitness function is expressed as follows: μ (x) ═ min { μ1(x),μ2(x),μ3(x)}，μi(x) Membership function values for three optimization objectives. The multi-objective optimization computation unit schedule translates into a single objective optimization problem that maximizes μ (x):
where s.t. represents a constraint.
And S44, selecting operation. The genetic algorithm simulates the biological evolution principle and adopts selection operation to eliminate individuals, the probability that the individuals with high fitness are transmitted to filial generations is high, and the probability that the individuals with low fitness are transmitted to the filial generations is low. However, if only the individuals with high fitness are selected for inheritance, the individuals with high fitness will be trapped in a local optimal solution, so the roulette algorithm is adopted to select the inherited individuals.
First, fitness values μ (x) of all individuals in the population are calculated according to step S2i) I is more than or equal to 1 and less than or equal to m, m is the population scale, the sum of the population fitness is calculated, and then the individual xiHas a selection probability of p (x)i) The selection probability of the ith computing unit deployment scenario, p (x), is showni) Comprises the following steps:
the cumulative probability of the individual is calculated,
randomly generating a random number r of (0,1) with a roulette algorithm, and the inheritance of the individualThe probabilities are matched to determine whether each individual is inherited by a descendant. If q isi-1＜r≤qiThen the ith individual will be inherited to the next generation. The selection operation can avoid the defect of falling into a local optimal solution caused by selecting only the individuals with high fitness for heredity.
S45, performing cross operation, wherein the cross operation is to exchange partial genes of two chromosomes with a certain cross probability to form a new chromosome individual, and is a main step for generating the new individual by a genetic algorithm, aiming at expecting that excellent genes can be inherited into offspring, and the cross operation step is as follows:
and T451, adaptively calculating the cross probability. The cross probability has extremely important influence on the performance and efficiency of the whole algorithm and is also important for the convergence of the algorithm, the larger the cross probability is, the faster new individuals are generated, and the possibility of damaging the genetic pattern is higher; on the other hand, the smaller the cross probability, the slower the whole search process becomes, even the less the search is stopped. The method for self-adaptively determining the cross probability is adopted to determine the cross probability, the cross probability automatically changes along with the fitness, when the individual fitness of a population tends to be consistent or tends to be locally optimal, the cross probability is increased, when the population fitness is relatively dispersed, the solution is protected to enter the next generation, and the method for self-adaptively determining the cross probability ensures the convergence of a genetic algorithm while keeping the diversity of the population.
Cross probability PcExpressed as:
wherein f ismaxRepresenting the maximum fitness value in the population, favgRepresents the mean fitness value of all individuals in the population, fcGreater fitness value, k, of two individuals to be crossed1，k2And taking the value between (0,1), and adjusting in the operation process. Randomly performing two operations on the individuals in the population according to the population individuals generated in the step S44Two pairs are paired, one pair of individuals is selected in turn to generate a random number r, if Pc>And r, performing cross operation on the selected individuals, otherwise, directly entering S46.
And T452, determining the intersection point according to the evaluation value of the physical machine. After determining the crossover probability in the crossover operation, the crossover point needs to be selected. And judging which gene is crossed according to the waste degree of physical machine resources in the gene and the SLA violation rate. The reason why the power consumption of the physical machine is not selected as a judgment basis is that the SLA violation rate and the power consumption are closely related to the CPU utilization rate and have a close variation trend, and the power consumption of a single physical machine cannot reflect the quality of the deployment scheme on the physical machine. And calculating the resource waste degree and SLA violation rate of the physical machines according to the steps S32 and S33, wherein the weighted sum of the resource waste degree and the SLA violation rate is the evaluation value of the physical machines in the individual, and performing cross operation on the physical nodes with small evaluation value in ascending order.
T453, a crossover operation is performed. If the chromosome Y has duplicate computing units in a plurality of physical machines, the physical machines with duplicate computing units are deleted, and the deletion operation may cause some computing units not to be allocated to physical machines, and then the unassigned computing units need to be recoded into the physical machines. Similarly, the crossover point in chromosome Y is inserted into chromosome X. After the crossover operation two new offspring chromosomes will be generated.
S46, mutation operation, wherein in the basic genetic algorithm, mutation operation is to randomly select one or more genes to perform mutation in individual chromosomes so as to avoid falling into local optimal solution and keep population diversity. The selection of the variation probability is crucial to the performance and the convergence of the genetic algorithm, the individual structure is not easy to generate if the variation probability is too small, the genetic algorithm becomes a pure random search algorithm if the variation probability is too large, and the steps of the variation operation are as follows:
and T461, adaptively calculating variation probability. When the individual fitness value is larger than the average individualIn the fitness value, a smaller mutation probability is selected, and excellent genes are kept as much as possible, whereas a larger mutation probability is selected. Adaptive computation of mutation probability PmThe following were used:
wherein f ismaxRepresenting the maximum fitness value in the population, favgRepresents the mean fitness value of all individuals in the population, fmIndicates the fitness value, k, of the individual to be mutated3，k4Take values between (0, 1). And adjusting in the operation process. According to the population individuals obtained in the step S45, sequentially selecting individuals in the population, calculating the mutation probability according to a calculation formula of the self-adaptive mutation probability, generating a random number r, and judging whether the mutation probability P existsmIf the value is more than r, mutation operation is carried out on the selected individual, otherwise, S47 is directly carried out.
And T462, determining a variation point according to the evaluation value of the physical machine, and selecting the physical machine with the largest evaluation value as the variation point according to the evaluation value of the physical machine calculated by the T452.
And T463, performing mutation operation. The mutation mode is random mutation, namely one or more of the original computing units are changed on the determined physical machine, repeated computing units may appear in the whole individual after mutation, at the moment, the repeated computing units are deleted, and at the moment, if undeployed computing units appear, the computing units are recoded on the physical machine according to the optimal adaptive algorithm.
S47, generating child individuals, generating a new child population after the steps, and if the iteration times reach the maximum times, determining the individual with the minimum fitness value in the current population as the optimal solution of the algorithm, and ending the algorithm; if the maximum iteration number is not reached, the step S44 is executed in a loop.
S5, chromosome codes of the optimal solution individuals are converted into corresponding computing unit deployment schemes, the conversion mode takes the computing unit deployment scheme in FIG. 3 as an example, and the computing unit deployment schemes correspond to a matrix with 7 rows and 3 columns in an algorithm, and the following steps are shown:
the position of 1 represents that the computing unit i is deployed on the physical machine j, and the converted matrix is output, namely the computing unit deployment scheme is output.
Claims (6)
1. The scheduling method of the computing unit is characterized by comprising the following steps:
s1, inputting physical machine resources: determining the number of physical machines, the types of resources owned by the physical machines and the capacity of the resources;
s2, inputting computing unit resources required by a user;
s3, performing formulaic description on the optimization target through a data model;
s4, scheduling the computing unit through a grouping genetic algorithm: firstly, determining a gene coding mode, then generating a plurality of groups of computing unit deployment schemes by a random method, enabling each computing unit deployment scheme to correspond to one individual in a packet genetic algorithm, converting a mapping relation between a computing unit and a physical machine into a gene code in the packet genetic algorithm by using the packet coding mode, and enabling all the individuals corresponding to the computing unit deployment schemes to form an initial population in the packet genetic algorithm; calculating the fitness value of each individual in the initial population, selecting the individual inherited to the next generation according to the selection probability of each individual, performing cross operation and variation operation on the corresponding physical machine according to the cross probability and the variation probability of each individual, then generating a progeny population, if the current iteration number reaches the maximum number set by the algorithm, the individual with the minimum fitness value in the current progeny population is the optimal solution of the grouped genetic algorithm, and entering step S5; if the current iteration times do not reach the maximum times set by the algorithm, returning to the step of selecting the individuals to be inherited to the next generation for circular execution until the optimal solution of the grouping genetic algorithm is obtained, and then entering step S5;
and S5, outputting a calculation unit deployment scheme corresponding to the optimal solution.
2. A method for scheduling a computational unit as claimed in claim 1, characterized by: the computing unit resource described in step S2 includes the requested number of computing units and the resource type and resource capacity of each computing unit.
3. A method for scheduling a computational unit as claimed in claim 1, characterized by: in step S4, fitness values for each individual in the initial population are calculated by fuzzy logic theory in a manner that maximizes the minimum satisfaction.
4. A method for scheduling a computational unit as claimed in claim 1, characterized by: in step S4, the selection probability of each individual is calculated by a roulette algorithm.
5. A method for scheduling a computational unit as claimed in claim 1, characterized by: in step S4, the cross probability and the mutation probability of the individual are determined by an adaptive algorithm, the evaluation values of the physical machines in the individual are calculated, the evaluation values are arranged in an ascending order, and the physical machine with the largest evaluation value is selected to perform the cross operation and the mutation operation.
6. Method for scheduling a computing unit according to one of claims 1 to 5, characterized in that: the optimization objectives described in step S3 include minimizing power consumption of the physical machine, minimizing resource utilization, and minimizing SLA breach rates.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610875838.4A CN106484512B (en) | 2016-10-08 | 2016-10-08 | The dispatching method of computing unit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610875838.4A CN106484512B (en) | 2016-10-08 | 2016-10-08 | The dispatching method of computing unit |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106484512A CN106484512A (en) | 2017-03-08 |
| CN106484512B true CN106484512B (en) | 2019-07-09 |
Family
ID=58269148
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610875838.4A Active CN106484512B (en) | 2016-10-08 | 2016-10-08 | The dispatching method of computing unit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106484512B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108279968B (en) * | 2017-12-29 | 2021-05-11 | 中国联合网络通信集团有限公司 | Virtual machine resource scheduling method and device |
| CN108287666B (en) * | 2018-01-16 | 2021-01-26 | 中国人民公安大学 | Data storage method and device for cloud storage environment |
| CN108521446B (en) * | 2018-03-20 | 2021-02-23 | 深圳大学 | Scheduling method, device and equipment of cloud computing resources and storage medium |
| CN109857526A (en) * | 2018-12-27 | 2019-06-07 | 曙光信息产业(北京)有限公司 | A kind of scheduling system towards mixing computation frame |
| CN110058924B (en) * | 2019-04-23 | 2023-08-04 | 东华大学 | Multi-objective optimized container scheduling method |
| CN110308993B (en) * | 2019-06-27 | 2022-12-13 | 大连理工大学 | Cloud computing resource allocation method based on improved genetic algorithm |
| US11023813B2 (en) * | 2019-10-09 | 2021-06-01 | Nmetric, Llc | Genetic algorithm with deterministic logic |
| CN113132445B (en) * | 2020-01-10 | 2023-04-28 | 阿里巴巴集团控股有限公司 | Resource scheduling method, equipment, network system and storage medium |
| CN111274030B (en) * | 2020-01-16 | 2022-09-23 | 中国人民解放军国防科技大学 | Efficient multiprocessor system-on-chip design space mining method oriented to application features |
| CN113300982B (en) * | 2020-06-08 | 2022-08-23 | 阿里巴巴集团控股有限公司 | Resource allocation method, device, system and storage medium |
| CN116502473B (en) * | 2023-06-27 | 2024-01-12 | 中科航迈数控软件(深圳)有限公司 | Wire harness electromagnetic compatibility optimization method and device, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102055196A (en) * | 2010-12-20 | 2011-05-11 | 南京软核科技有限公司 | 10 kv-distribution network reactive power compensation optimization method in power system |
| CN103412792A (en) * | 2013-07-18 | 2013-11-27 | 成都国科海博计算机系统有限公司 | Dynamic task scheduling method and device under cloud computing platform environment |
| CN103744714A (en) * | 2011-12-31 | 2014-04-23 | 华茂云天科技(北京)有限公司 | Virtual machine management platform based on cloud computing |
| CN104657215A (en) * | 2013-11-19 | 2015-05-27 | 南京鼎盟科技有限公司 | Virtualization energy-saving system in Cloud computing |
| CN105302632A (en) * | 2015-11-19 | 2016-02-03 | 国家电网公司 | Cloud computing working load dynamic integration method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013040539A1 (en) * | 2011-09-16 | 2013-03-21 | Siemens Corporation | Method and system for energy control management |
-
2016
- 2016-10-08 CN CN201610875838.4A patent/CN106484512B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102055196A (en) * | 2010-12-20 | 2011-05-11 | 南京软核科技有限公司 | 10 kv-distribution network reactive power compensation optimization method in power system |
| CN103744714A (en) * | 2011-12-31 | 2014-04-23 | 华茂云天科技(北京)有限公司 | Virtual machine management platform based on cloud computing |
| CN103412792A (en) * | 2013-07-18 | 2013-11-27 | 成都国科海博计算机系统有限公司 | Dynamic task scheduling method and device under cloud computing platform environment |
| CN104657215A (en) * | 2013-11-19 | 2015-05-27 | 南京鼎盟科技有限公司 | Virtualization energy-saving system in Cloud computing |
| CN105302632A (en) * | 2015-11-19 | 2016-02-03 | 国家电网公司 | Cloud computing working load dynamic integration method |
Non-Patent Citations (2)
| Title |
|---|
| 云环境下调度问题的研究与实现;秦烁;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160215(第02期);I139-10 * |
| 基于遗传算法的分组算法;李峰 等;《长春工业大学学报（自然科学版）》;20130430;第34卷(第2期);214-217 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106484512A (en) | 2017-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106484512B (en) | The dispatching method of computing unit | |
| CN107172166B (en) | Cloud and mist computing system for industrial intelligent service | |
| CN108573326B (en) | Video server site selection method based on genetic ant colony hybrid algorithm | |
| CN109840154B (en) | Task dependency-based computing migration method in mobile cloud environment | |
| CN102063339B (en) | Resource load balancing method and equipment based on cloud computing system | |
| CN117539726B (en) | Energy efficiency optimization method and system for green intelligent computing center | |
| CN107995039A (en) | The resource self study of facing cloud software service and self-adapting distribution method | |
| CN109447264B (en) | Virtual machine placement genetic optimization method based on VHAM-R model in cloud computing environment | |
| CN109413710B (en) | Clustering method and device of wireless sensor network based on genetic algorithm optimization | |
| CN113285832B (en) | NSGA-II-based power multi-mode network resource optimization allocation method | |
| CN110008023B (en) | Cloud computing system budget constraint random task scheduling method based on genetic algorithm | |
| CN105550033A (en) | Genetic-tabu hybrid algorithm based resource scheduling policy method in private cloud environment | |
| CN106230827B (en) | A kind of multiple target service combining method based on cost-effectiveness optimization | |
| CN111176784B (en) | Virtual machine integration method based on extreme learning machine and ant colony system | |
| CN109582985A (en) | A kind of NoC mapping method of improved genetic Annealing | |
| CN110163546A (en) | A method of optimal replenishment quantity is determined based on big data | |
| CN112270398A (en) | Cluster behavior learning method based on gene programming | |
| CN116366453A (en) | Self-adaptive dynamic deployment method for heterogeneous network element service demand characterization and virtual network element | |
| CN112036651A (en) | Electricity price prediction method based on quantum immune optimization BP neural network algorithm | |
| Fedorchenko et al. | Modified genetic algorithm to determine the location of the distribution power supply networks in the city | |
| Korejo et al. | Multi-population methods with adaptive mutation for multi-modal optimization problems | |
| CN109102203A (en) | A kind of Target Assignment optimization method based on more string chromosomal inheritance algorithms | |
| Ouyang et al. | Amended harmony search algorithm with perturbation strategy for large-scale system reliability problems | |
| CN112380006A (en) | Data center resource allocation method and device | |
| Yu | [Retracted] Research on Optimization Strategy of Task Scheduling Software Based on Genetic Algorithm in Cloud Computing Environment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |