CN116089083A - Multi-target data center resource scheduling method - Google Patents
Multi-target data center resource scheduling method Download PDFInfo
- Publication number
- CN116089083A CN116089083A CN202310059054.4A CN202310059054A CN116089083A CN 116089083 A CN116089083 A CN 116089083A CN 202310059054 A CN202310059054 A CN 202310059054A CN 116089083 A CN116089083 A CN 116089083A
- Authority
- CN
- China
- Prior art keywords
- evaluation function
- task
- function value
- scheduling
- scheduling scheme
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000011156 evaluation Methods 0.000 claims abstract description 97
- 239000003016 pheromone Substances 0.000 claims abstract description 56
- 241000257303 Hymenoptera Species 0.000 claims abstract description 41
- 230000007704 transition Effects 0.000 claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims description 36
- 238000004422 calculation algorithm Methods 0.000 claims description 24
- 238000012545 processing Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000001010 compromised effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
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
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5072—Grid computing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A multi-target data center resource scheduling method aims at solving the problems that cloud computing resource scheduling execution time is long and task execution cost is high, and a multi-target cloud resource scheduling model based on Z-number is established; acquiring a scheduling task of cloud computing resources, and acquiring an allocation relation between a model task and a virtual machine by utilizing a heuristic rule based on a minimum idle strategy to serve as a current optimal evaluation function value; initializing ant colony according to the relation, randomly placing ants, and reassigning tasks to the virtual machines through heuristic rules and state transition rules based on double pheromone matrixes to generate a scheduling scheme; comparing the evaluation function value of the scheduling scheme with the current optimal evaluation function value, if the evaluation function value is superior to the current optimal evaluation function value, searching a neighborhood of the current optimal evaluation function value by using a local search strategy, updating the current optimal evaluation function value, generating a new scheduling scheme, and sending a task result to a user side; otherwise, the task result of the current scheduling scheme is sent to the user side. Belongs to the technical field of cloud computing.
Description
Technical Field
The invention relates to a resource scheduling method, in particular to a resource scheduling method of a multi-target data center, and belongs to the technical field of cloud computing.
Background
The cloud computing system integrates heterogeneous and cheap computing resources through a network, provides services to the outside in a unified way, greatly improves the performance and reliability of the system, reduces the use and maintenance cost, and enables users to pay for use of the services in the cloud computing system conveniently, cheaply, reliably and efficiently. In the cloud computing system, computing resources can be machines of different models distributed at different places, and the cloud computing system integrates the complicated machines through technical means, but the structure of the cloud computing system is very complex. In the process of resource scheduling, the allocated resources may be too few, so that the task completion time is too long, the user operation time delay is too high, even if the currently allocated computing resources cannot meet the user operation requirement, the allocated resources may be too many, so that part of the resources are in an idle state for a long time, the utilization rate of the resources is greatly reduced, and a lot of unnecessary expenses are brought to users or cloud service operators. These unbalanced states not only result in wasted computing resources, but also add to the cost, and the benefits of the cloud service operators are compromised while reducing customer satisfaction with the service. Even if the computing resources are in a proper range, different resource scheduling schemes can produce different results in consideration of the total completion time of the tasks, the processing time delay of the tasks and the load balancing performance of the cloud computing system, and the execution efficiency and the cost of the tasks can be greatly affected. Therefore, on the premise of meeting the ever-increasing and changing demands of users, how to perform reasonable scheduling of computing resources with high efficiency and low cost, and finding a balance point between the capability and the demands are key problems of cloud computing.
Disclosure of Invention
The invention aims to solve the problems of longer execution time and higher task execution cost of cloud computing resource scheduling, and further provides a multi-target data center resource scheduling method.
It comprises the following steps:
s1, establishing a Z-number-based multi-target cloud resource scheduling model;
s2, the user side sends a scheduling task of the cloud computing resource to the cloud side, the cloud side obtains an initial solution of a multi-target cloud resource scheduling model based on Z-number by utilizing a heuristic rule based on a minimum idle strategy according to the scheduling task, the initial solution is used as a current optimal evaluation function value, and the initial solution is an allocation relation of the task and the virtual machine;
s3, initializing ant colony of an ant colony algorithm according to an initial solution, randomly placing ants, reassigning tasks to the virtual machines through heuristic rules and state transition rules based on a double pheromone matrix, and generating a scheduling scheme;
s4, calculating an evaluation function value of the scheduling scheme, comparing the evaluation function value with the current optimal evaluation function value in the S2, searching a neighborhood of the current optimal evaluation function value by using a local search strategy if the evaluation function value is better than the current optimal evaluation function value, updating the current optimal evaluation function value, generating a new scheduling scheme, and sending a task result to a user side according to the scheduling scheme; otherwise, the task result of the current scheduling scheme is sent to the user side without updating the current scheduling scheme and the optimal evaluation function value;
s5, acquiring a scheduling task of the cloud computing resource sent by the user side, executing the S2-S4 to obtain a scheduling scheme corresponding to the scheduling task, and sending a task result to the user side according to the scheduling scheme.
Further, in S1, a Z-number-based multi-objective cloud resource scheduling model is provided:
wherein Time (P) represents the total execution Time of scheduling scheme P; cost (P) represents the total Cost of executing a task; load represents a Load balancing function.
Further, in S2, the user side sends the scheduling task of the cloud computing resource to the cloud end, and the cloud end obtains an initial solution of the multi-objective cloud resource scheduling model based on the Z-number by using a heuristic rule based on a minimum idle policy according to the scheduling task, and takes the initial solution as a current optimal evaluation function value, wherein the initial solution is an allocation relation between the task and the virtual machine, and the specific process is as follows:
and placing all the tasks into an assignable task set, when the task set is not empty, if an idle virtual machine exists, assigning the tasks to the idle virtual machine with the highest processing speed, otherwise, assigning the tasks to the virtual machine with the shortest execution time in the virtual machine currently running, and finally removing the assigned tasks from the task set until the task set is empty, thereby obtaining the assignment relation between the tasks and the virtual machines.
Further, in S3, according to the ant colony of the ant colony algorithm initialized by the initial solution, ants are randomly placed, tasks are reassigned to the virtual machines through heuristic rules and state transition rules based on the double pheromone matrix, and a scheduling scheme is generated, and the specific process is as follows:
initializing ant colony of an ant colony algorithm according to an initial solution, randomly placing ants, after each round of iteration of the ant colony algorithm is finished, distributing all tasks of each ant to corresponding virtual machines through heuristic rules and state transition rules based on a double pheromone matrix, generating a scheduling scheme P, calculating an evaluation function value of the scheduling scheme P, sorting the ants from high to low according to all the evaluation function values, and selecting partial ants with the highest evaluation function value as elite ants;
all ant roots are processedThe evaluation function value of the scheduling scheme generated according to the evaluation function value is in the pheromone matrix M 1 Leaving pheromone, and putting the evaluation function value of the scheduling scheme generated by elite ants in the pheromone matrix M 2 Leaving the pheromone behind, matrix M 1 And a pheromone matrix M 2 And taking the two pheromone matrixes as the next iteration until the upper limit of the iteration times is met, and obtaining the distribution relation between the final task and the virtual machine.
Further, the heuristic function in S3 is:
wherein, time is j and costj Respectively search for task t for ants j Up to the time of execution of the task and the required cost; etc ij and cstij Respectively tasks t j In virtual machine vm i Time and cost required for execution.
Further, the state transition rule based on the dual pheromone matrix in S3 is:
wherein ,
to ant k at time t, task t j Allocation to virtual machines vm i Probability of (2); τ ij(t) and εij (t) represents the path points (vm) of ants and elite ants at time t i ，t j ) The pheromone concentration on the sample; alpha is the pheromone factor of common ants; gamma is the pheromone factor of elite ants; beta is a pheromone volatilization factor; η (eta) ij (t) is a heuristic function factor; VM is a set of virtual machines; s is the s-th virtual machine; al k The selected set of tasks is allowed for the next step for the ant.
Further, the evaluation function value is:
F＝ω 1 ·logTime(P)+ω 2 ·logCost(P)-ω 3 ·(2+logLoad) (4)
wherein ,ω1 ，ω 2 ，ω 3 Weight factors of task execution time, virtual machine cost consumption and system load balance degree respectively, and omega 1 +ω 2 +ω 3 ＝1。
Further, in S4, a local search strategy is used to search the neighborhood of the current optimal evaluation function value, update the current optimal evaluation function value, and generate a new scheduling scheme, which specifically includes:
two virtual machines are selected randomly in the neighborhood of the current optimal evaluation function value, a task is selected randomly on each virtual machine, the selected two tasks are exchanged, a new scheduling scheme is obtained after each virtual machine exchanges the tasks, the evaluation function value of the new scheduling scheme is calculated, if the evaluation function value is superior to the corresponding evaluation function value before the task exchange, the new scheduling scheme is used as the final scheduling scheme, the evaluation function value of the new scheduling scheme is used as the current optimal evaluation function value, and otherwise, the new scheduling scheme and the evaluation function value thereof are abandoned.
The beneficial effects are that:
according to the method, the multi-target cloud resource scheduling model based on the Z-number is established, and the uncertainty value can be described more accurately by introducing the Z-number to represent the execution time of the cloud resource scheduling task. According to the scheduling task of the cloud computing resource, an initial solution of a multi-target cloud resource scheduling model based on Z-number is obtained by utilizing a heuristic rule based on a minimum idle strategy, the initial solution is an allocation relation between the task and the virtual machine, and the initial solution is used as a current optimal evaluation function value. Based on the ant colony algorithm, task is reassigned to the virtual machine by combining a heuristic rule and a state transition rule based on a double pheromone matrix, and a scheduling scheme is generated, the heuristic rule based on a minimum idle strategy is an optimization method aiming at resource scheduling characteristics in a cloud computing system, and a better initial solution can be rapidly calculated in an initial stage of the algorithm, so that the subsequent search of the ant colony can be better guided by initializing the pheromone matrix of the ant colony, and the quality of the solution in local search is improved. Comparing the evaluation function value of the scheduling scheme with the current optimal evaluation function value, if the evaluation function value is better than the current optimal evaluation function value, searching the neighborhood of the current optimal evaluation function value by using a local search strategy, updating the current optimal evaluation function value, and generating a new scheduling scheme; otherwise, the current optimal evaluation function value is not updated. Based on the method, compared with other inventions, the method has excellent performance in solving quality, convergence speed and system load, and improves the scheduling execution time of cloud computing resources and the task executing cost.
Drawings
FIG. 1 is a schematic flow chart of the present invention;
FIG. 2 is a graph of ladder ambiguity membership functions;
FIG. 3 is a schematic diagram of a partial search;
FIG. 4 is a graph of a deterministic model versus a fuzzy model;
FIG. 5 is a comparative load balancing graph;
Detailed Description
The first embodiment is as follows: the present embodiment is described with reference to fig. 1 to 5, which illustrate a multi-objective data center resource scheduling method according to the present embodiment, including the following steps:
s1, establishing a Z-number-based multi-target cloud resource scheduling model.
Z-number is a fuzzy approach proposed by Zadeh in 2011, primarily to represent human uncertain knowledge. The Z-number has the advantages that the Z-number can be used for representing the possibility and the credibility of the fuzzy variable, the credibility of the value is considered on the basis of the fuzzy variable, and the fuzzy variable can be described more accurately, so that the Z-number is suitable for being used as a mathematical tool for fuzzy optimization. Since the execution time of the task on the virtual machine is actually an uncertain fuzzy number, the invention indicates the execution time of the task which is uncertain in the cloud resource scheduling by the Z-number. The general meaning of Z-number is that the value of the variable is ambiguous and the statement describing this ambiguous value is ambiguous, e.g., for task t j The form after the description by using the Z-number is that the execution time is 100, the credibility is moderate or the execution is performedThe time may be 90 and the reliability is high, and the two descriptions are more effective, and the calculation needs to be performed by using corresponding fuzzy mathematical tools.
in the resource scheduling scheme P, the virtual machine vm i Execute task t j Time of need vmTime i And cost vmCost i Respectively is
wherein ,dij To take task t j Assigned to virtual machine vm i Executing on, a task can only be executed on one virtual machine, d ij E {0,1}, and
i＝1,2，…，n；e ij to represent task t j In virtual machine vm i Execution time on;
i＝1,2，…，m。
vmCost i ＝vmTime i ×rcu i (3)
wherein ,rcui Cost per unit time.
The total execution time required to execute the scheduling scheme P is
Time(P)＝max{vmTime 1 ,vmTime 2 ,…,vmTime m } (4)
The total cost of executing the task is
Defining a load balancing function of a system as
The multi-objective optimization problem of cloud computing resource scheduling (Z-number based multi-objective cloud resource scheduling model) can be expressed as:
min [ Time (P), cost (P), -Load ] (7) data normalization is required for the three data in equation (7) because of asymmetry in data size of task execution Time, required Cost and Load function value. The invention processes the three data by using logarithmic normalization, and finally evaluates the expression of the function value:
F＝ω 1 ·logTime(P)+ω 2 ·logCost(P)-ω 3 ·(2+logLoad) (8)
wherein ,ω1 ，ω 2 ，ω 3 Weight factors of task execution time, virtual machine cost consumption and system load balance degree respectively, and omega 1 +ω 2 +ω 3 =1. At this time, the objective of the cloud computing resource scheduling is to minimize the evaluation function value.
S2, the user side sends the scheduling task of the cloud computing resource to the cloud side, the cloud side obtains an initial solution of the multi-target cloud resource scheduling model based on the Z-number by utilizing a heuristic rule based on a minimum idle strategy according to the scheduling task, the initial solution is used as a current optimal evaluation function value, and the initial solution is an allocation relation of the task and the virtual machine. The specific process is as follows:
and (3) putting all the scheduling tasks into an assignable task set, when the task set is not empty, if an idle virtual machine exists, assigning the scheduling tasks to the idle virtual machine with the highest processing speed, otherwise, assigning the scheduling tasks to the virtual machine with the shortest execution time in the currently running virtual machine, and finally removing the assigned scheduling tasks from the task set until the task set is empty, thereby obtaining the assignment relation between the scheduling tasks and the virtual machines.
The general intelligent heuristic algorithm is difficult to search the optimal solution in a short time, so the invention provides a heuristic rule based on the minimum idle strategy. In the initialization stage of the algorithm, a better initial solution of the multi-target cloud resource scheduling model based on the Z-number is rapidly obtained, so that subsequent searching is guided, a large amount of algorithm searching time can be saved, interference of inferior solutions is avoided, and searching efficiency is improved. The solution obtained by using heuristic rules has better performance in terms of total execution time of tasks and load balancing, but the minimum idle strategy does not consider cost factors and global optimization, so the initial solution obtained is essentially a locally optimal solution.
S3, initializing ant colony of an ant colony algorithm according to the initial solution, randomly placing ants, reassigning tasks to the virtual machines through heuristic rules and state transition rules based on a double pheromone matrix, and generating a scheduling scheme.
In the course of ant finding paths in ant colony algorithm, if the random strategy is completely used, a large number of low-quality solutions will be generatedTherefore, the invention uses heuristic function and pheromone matrix to guide ant path search, and completes task allocation. Some path node (vm) i ，t j ) Representing task t j Allocation to virtual machines vm i The heuristic function is:
wherein, time is j and costj Respectively search for task t for ants j Up to the time of execution of the task and the required cost; etc ij and cstij Respectively tasks t j In virtual machine vm i Time and cost required for execution.
The pheromone matrix is an important factor that ant colony can search better evaluation function value, and the invention uses two pheromone matrices M 1 and M2 And guiding ant path searching to complete task distribution. After each iteration of the ant colony algorithm is finished, each ant distributes all tasks of the ant to the virtual machine to generate a scheduling scheme P, namely each ant generates a corresponding scheduling scheme. And (3) calculating an evaluation function value (formula (8)) of the scheduling scheme P, sorting ants according to the evaluation function value from high to low, and selecting a part (the user-defined number or percentage) of ants with the best evaluation function value as elite ants. The evaluation function values of the scheduling schemes generated by all ants are calculated in a pheromone matrix M 1 Leaving the pheromone but in the pheromone matrix M 2 In which only elite ants can leave pheromone, matrix the pheromone M 1 And a pheromone matrix M 2 And taking the two pheromone matrixes as the next iteration until the upper limit of the iteration times is met, and obtaining the distribution relation between the final task and the virtual machine.
Pheromone matrix M 1 The pheromone update formula of (2) is
Pheromone matrix M 2 The pheromone update formula of (2) is
Wherein E is elite ant set, h is elite ant individual
Wherein ρ [ epsilon ] (0, 1) is the pheromone volatilization coefficient τ ij(t) and εij (t) represents the path points (vm) of ant k and elite ant h at time t i ，t j ) The concentration of the pheromone in the water,
representing the current iteration at the waypoint (vm i ，t j ) The pheromone increment left by ant k, +.>
The pheromone increment left for elite ant h, f (P k) and f(Ph ) The execution time of the ant k and elite h generation schemes, respectively. Q is a pheromone constant.
State transition rule based on double pheromone matrix is
wherein ,
to ant k at time t, task t j Allocation to virtual machines vm i Probability of (2); τ ij(t) and εij (t) represents the path points (vm) of ants and elite ants at time t i ，t j ) The pheromone concentration on the sample; alpha is the pheromone factor of common ants; gamma is the pheromone factor of elite ants; beta is a pheromone volatilization factor; η (eta) ij (t) is a heuristic function factor; VM is a set of virtual machines; s is the s-th virtual machine; al k The selected set of tasks is allowed for the next step for the ant.
S4, calculating an evaluation function value of each scheduling scheme, comparing the evaluation function value with the current optimal evaluation function value in the S2, searching a neighborhood of the current optimal evaluation function value by using a local search strategy if the evaluation function value is better than the current optimal evaluation function value, updating the current optimal evaluation function value, thereby improving the quality of the evaluation function value, generating a new scheduling scheme, and transmitting a task result to a user side according to the scheduling scheme; otherwise, the task result of the current scheduling scheme is sent to the user side without updating the current scheduling scheme and the optimal evaluation function value. The specific process is as follows:
two virtual machines are selected randomly in the neighborhood of the current optimal evaluation function value, a task is selected randomly on each virtual machine, the selected two tasks are exchanged, and a new scheduling scheme is obtained after each virtual machine exchanges the tasks, as shown in fig. 3. After the exchange operation is executed once, calculating an evaluation function value of the new scheduling scheme, if the evaluation function value is higher than the corresponding evaluation function value before the task exchange, the quality of the evaluation function value is improved, taking the new scheduling scheme as a final scheduling scheme, taking the evaluation function value of the new scheduling scheme as the current optimal evaluation function value, and otherwise, discarding the new scheduling scheme and the evaluation function value thereof. Because the local search takes a lot of time, only the current optimal evaluation function value is searched, and the better the current optimal evaluation function value is, the better the evaluation function value obtained by the local search is.
S5, converting the scheduling scheme obtained in the step S4 and the evaluation function value into data for simulation.
Performing simulation experiments on a cloud computing resource scheduling simulation algorithm platform cloudsim, and randomly generating task data and virtual machine data in intervals [30000, 130000] and intervals [300, 1300], wherein the task data comprise task execution time and cost, the virtual machine data comprise virtual machine execution speed and unit cost, the type of a data file is json format, wherein the mip and cost are virtual machine execution speed and unit cost respectively, and the etc and cst are task execution time matrix and cost matrix respectively. And calculating an ETC matrix, and converting the task execution time determined in the ETC matrix into a Z-number form to serve as a data set of the experiment. The main parameters of the HACO algorithm comprise alpha, gamma, beta, rho, Q, total ants, elite ants and ant colony iteration upper limit, wherein the total number of ants is 100, the elite ants are 10, and the ant colony iteration upper limit is 300 given Q=1; the parameters alpha, gamma, beta and rho are selected to be proper values in the respective ranges by an orthogonal experiment method, so that the parameter value ranges are alpha epsilon {1,2,3,4,5}, gamma epsilon {1,2,3,4,5}, beta epsilon {1,2,3,4,5}, rho epsilon {0.5,0.6, 0.7,0.8,0.9}.
And the cloud computing system generates a large number of tasks after receiving the demands of the user, and reasonably distributes the tasks to different computing nodes for execution. In a certain environment, it is generally assumed that, in a cloud computing system, the time taken for a virtual computing node to execute a task is a certain value, and at this time, a cloud computing resource scheduling problem can be described as:
task set t= { T 1 ,t 2 ,…,t n Reasonable scheduling of n tasks in the virtual resource node set vm= { VM 1 ,vm 2 ,…,vm m On m virtual machines (m)<n) to maximize the execution efficiency of the task as much as possible. The corresponding relation matrix of the task and the virtual machine is represented by D, wherein m acts as a virtual machine queue, and n acts as a task queue.
In equation (15), a complete correspondence matrix D is a scheduling scheme P. The execution time matrix ETC may be constructed from the correspondence matrix D.
etc ij ＝tr ij +run ij (17)
In formula (16), etc ij Representing task t j In virtual machine vm i Execution time on the same. In formula (17), tr ij For task t j Transfer to virtual machine vm i Time run spent on ij At t j At vm i Time it takes to perform the calculation.
In scheduling scheme P, vm i The time taken to perform the allocation task is:
then determining that the cloud computing resource scheduling problem model with the scheduling target of minimizing the task completion time is under the environment as follows
wherein ,
Time(P)＝max{vmTime 1 ,vmTime 2 ,…,vmTime m } (20)
equation (19) is an objective function and equation (20) is the final task completion time for a scheduling scheme.
S6, testing a model.
Under different task scales, the evaluation function values of the determining model and the fuzzy model are calculated by using HACO, and the comparison result of the evaluation function values of the cloud computing resource scheduling problem under the determining condition and the cloud computing resource scheduling problem under the uncertain condition is shown in fig. 4, so that the evaluation function value of the cloud computing resource scheduling model under the uncertain environment is higher, which means that the time and the cost for resource scheduling can be higher in the uncertain system.
S7, acquiring a scheduling task of the cloud computing resource sent by the user side, executing the S2-S4 to obtain a scheduling scheme corresponding to the scheduling task, and sending a task result to the user side according to the scheduling scheme.
Examples
The impact of each of the improved strategies of the algorithm of the present invention on performance was analyzed and the HACO of the present invention was compared to the other three algorithms as shown in table 1.
Table 1 comparison of optimization strategies
The HACO was compared with GA, ACO, CACO and CACO-SVR, and at different task scales, solved using five algorithms, respectively, with load balancing performance of the five algorithms at 50task and 500task, as shown in fig. 5. It can be seen that GA has higher load balancing performance compared with ACO algorithm due to its algorithm characteristics, and the load balancing performance of ACO, CACO and CACO-SVR increases in turn, while the load balancing performance of HACO is optimal. And as the task scale increases, the load balancing capability of ACO and CACO is obviously reduced, GA and CACO-SVR are basically kept unchanged, HACO always keeps extremely high load balancing, and as the task scale increases, the load balancing is higher.
In conclusion, the method provided by the invention can effectively improve the solving efficiency of the multi-target data center resource problem model when solving the multi-target data center resource problem model.
Claims (8)
1. A multi-target data center resource scheduling method is characterized in that: it comprises the following steps:
s1, establishing a Z-number-based multi-target cloud resource scheduling model;
s2, the user side sends a scheduling task of the cloud computing resource to the cloud side, the cloud side obtains an initial solution of a multi-target cloud resource scheduling model based on Z-number by utilizing a heuristic rule based on a minimum idle strategy according to the scheduling task, the initial solution is used as a current optimal evaluation function value, and the initial solution is an allocation relation of the task and the virtual machine;
s3, initializing ant colony of an ant colony algorithm according to an initial solution, randomly placing ants, reassigning tasks to the virtual machines through heuristic rules and state transition rules based on a double pheromone matrix, and generating a scheduling scheme;
s4, calculating an evaluation function value of the scheduling scheme, comparing the evaluation function value with the current optimal evaluation function value in the S2, searching a neighborhood of the current optimal evaluation function value by using a local search strategy if the evaluation function value is better than the current optimal evaluation function value, updating the current optimal evaluation function value, generating a new scheduling scheme, and sending a task result to a user side according to the scheduling scheme; otherwise, the task result of the current scheduling scheme is sent to the user side without updating the current scheduling scheme and the optimal evaluation function value;
s5, acquiring a scheduling task of the cloud computing resource sent by the user side, executing the S2-S4 to obtain a scheduling scheme corresponding to the scheduling task, and sending a task result to the user side according to the scheduling scheme.
2. A multi-objective data center resource scheduling method according to claim 1, wherein: z-number-based multi-target cloud resource scheduling model in S1:
min[Time(P),Cost(P),-Load](1)
wherein Time (P) represents the total execution Time of scheduling scheme P; cost (P) represents the total Cost of executing a task; load represents a Load balancing function.
3. A multi-objective data center resource scheduling method according to claim 2, wherein: s2, the user side sends a scheduling task of the cloud computing resource to the cloud side, the cloud side obtains an initial solution of a multi-target cloud resource scheduling model based on Z-number by utilizing heuristic rules based on a minimum idle strategy according to the scheduling task, the initial solution is used as a current optimal evaluation function value, and the initial solution is an allocation relation of the task and the virtual machine, and specifically comprises the following steps:
and (3) putting all the scheduling tasks into an assignable task set, when the task set is not empty, if an idle virtual machine exists, assigning the scheduling tasks to the idle virtual machine with the highest processing speed, otherwise, assigning the scheduling tasks to the virtual machine with the shortest execution time in the currently running virtual machine, and finally removing the assigned scheduling tasks from the task set until the task set is empty, thereby obtaining the assignment relation between the scheduling tasks and the virtual machines.
4. A multi-objective data center resource scheduling method according to claim 3, wherein: and S3, initializing an ant colony of an ant colony algorithm according to an initial solution, randomly placing ants, reassigning tasks to the virtual machine through heuristic rules and state transition rules based on a double pheromone matrix, and generating a scheduling scheme, wherein the specific process is as follows:
initializing ant colony of an ant colony algorithm according to an initial solution, randomly placing ants, after each round of iteration of the ant colony algorithm is finished, distributing all tasks of each ant to corresponding virtual machines through heuristic rules and state transition rules based on a double pheromone matrix, generating a scheduling scheme P, calculating an evaluation function value of the scheduling scheme P, sorting the ants from high to low according to all the evaluation function values, and selecting partial ants with the highest evaluation function value as elite ants;
the evaluation function values of the scheduling schemes generated by all ants are arranged in a pheromone matrix M 1 Leaving pheromone, and putting the evaluation function value of the scheduling scheme generated by elite ants in the pheromone matrix M 2 Leaving the pheromone behind, matrix M 1 And a pheromone matrix M 2 And taking the two pheromone matrixes as the next iteration until the upper limit of the iteration times is met, and obtaining the distribution relation between the final task and the virtual machine.
5. The multi-objective data center resource scheduling method according to claim 4, wherein: the heuristic function in S3 is:
wherein, time is j and costj Respectively search for task t for ants j Up to the time of execution of the task and the required cost; etc ij and cstij Respectively tasks t j In virtual machine vm i Time and cost required for execution.
6. The multi-objective data center resource scheduling method according to claim 5, wherein: the state transition rule based on the double pheromone matrix in the S3 is as follows:
wherein ,
to ant k at time t, task t j Allocation to virtual machines vm i Probability of (2); τ ij(t) and εij (t) represents the path points (vm) of ants and elite ants at time t i ，t j ) The pheromone concentration on the sample; alpha is the pheromone factor of common ants; gamma is the pheromone factor of elite ants; beta is a pheromone volatilization factor; η (eta) ij (t) is a heuristic function factor; VM is a set of virtual machines; s is the s-th virtual machine; al k The selected set of tasks is allowed for the next step for the ant.
7. The multi-objective data center resource scheduling method according to claim 6, wherein: the evaluation function value is:
F＝ω 1 ·logTime(P)+ω 2 ·logCost(P)-ω 3 ·(2+logLoad) (4)
wherein ,ω1 ，ω 2 ，ω 3 Weight factors of task execution time, virtual machine cost consumption and system load balance degree respectively, and omega 1 +ω 2 +ω 3 ＝1。
8. The multi-objective data center resource scheduling method as claimed in claim 7, wherein: s4, searching a neighborhood of the current optimal evaluation function value by using a local search strategy, updating the current optimal evaluation function value, and generating a new scheduling scheme, wherein the specific process is as follows:
two virtual machines are selected randomly in the neighborhood of the current optimal evaluation function value, a task is selected randomly on each virtual machine, the selected two tasks are exchanged, a new scheduling scheme is obtained after each virtual machine exchanges the tasks, the evaluation function value of the new scheduling scheme is calculated, if the evaluation function value is superior to the corresponding evaluation function value before the task exchange, the new scheduling scheme is used as the final scheduling scheme, the evaluation function value of the new scheduling scheme is used as the current optimal evaluation function value, and otherwise, the new scheduling scheme and the evaluation function value thereof are abandoned.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310059054.4A CN116089083A (en) | 2023-01-17 | 2023-01-17 | Multi-target data center resource scheduling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310059054.4A CN116089083A (en) | 2023-01-17 | 2023-01-17 | Multi-target data center resource scheduling method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116089083A true CN116089083A (en) | 2023-05-09 |
Family
ID=86198841
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310059054.4A Pending CN116089083A (en) | 2023-01-17 | 2023-01-17 | Multi-target data center resource scheduling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116089083A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117319505A (en) * | 2023-11-30 | 2023-12-29 | 天勰力(山东)卫星技术有限公司 | Satellite task order-robbing system facing software-defined satellite shared network |
-
2023
- 2023-01-17 CN CN202310059054.4A patent/CN116089083A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117319505A (en) * | 2023-11-30 | 2023-12-29 | 天勰力(山东)卫星技术有限公司 | Satellite task order-robbing system facing software-defined satellite shared network |
| CN117319505B (en) * | 2023-11-30 | 2024-02-06 | 天勰力(山东)卫星技术有限公司 | Satellite task order-robbing system facing software-defined satellite shared network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106875090B (en) | Dynamic task-oriented multi-robot distributed task allocation forming method | |
| CN107172166B (en) | Cloud and mist computing system for industrial intelligent service | |
| Shen et al. | Mathematical modeling and multi-objective evolutionary algorithms applied to dynamic flexible job shop scheduling problems | |
| CN110321217B (en) | Multi-target cloud resource scheduling method, device, equipment and storage medium | |
| CN106371924B (en) | A kind of method for scheduling task minimizing MapReduce cluster energy consumption | |
| Xiao et al. | A cooperative coevolution hyper-heuristic framework for workflow scheduling problem | |
| CN106126323A (en) | Real-time task scheduling method based on cloud platform | |
| CN112685138B (en) | Multi-workflow scheduling method based on multi-population hybrid intelligent optimization in cloud environment | |
| Wu et al. | Supporting distributed application workflows in heterogeneous computing environments | |
| CN115480876A (en) | Cloud computing task scheduling method and system based on ant colony algorithm optimization | |
| CN116089083A (en) | Multi-target data center resource scheduling method | |
| CN114978908A (en) | Computing power network node evaluation and operation method and device | |
| CN116302389A (en) | Task scheduling method based on improved ant colony algorithm | |
| CN116050540A (en) | Self-adaptive federal edge learning method based on joint bi-dimensional user scheduling | |
| Kalra et al. | An intelligent water drops-based approach for workflow scheduling with balanced resource utilisation in cloud computing | |
| CN113139639A (en) | MOMBI-based smart city application-oriented multi-target calculation migration method and device | |
| CN111913800B (en) | Resource allocation method for optimizing cost of micro-service in cloud based on L-ACO | |
| CN117032902A (en) | Cloud task scheduling method for improving discrete particle swarm algorithm based on load | |
| CN106874215B (en) | Serialized storage optimization method based on Spark operator | |
| CN112698911B (en) | Cloud job scheduling method based on deep reinforcement learning | |
| Bensalem et al. | Towards optimal serverless function scaling in edge computing network | |
| CN114980216A (en) | Dependent task unloading system and method based on mobile edge calculation | |
| Javadi-Moghaddam et al. | Resource allocation in cloud computing using advanced imperialist competitive algorithm. | |
| CN110297704B (en) | Particle swarm optimization method and system integrating reverse learning and heuristic perception | |
| Xiao et al. | Learning task allocation for multiple flows in multi-agent systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |