An intelligent job scheduling system with multi-objective optimization and its scheduling method

Abstract

The present invention discloses an intelligent job scheduling system for multi-objective optimization and its scheduling method, belonging to the technical field of supercomputers. The scheduling method searches for the Pareto solution set in multi-objectives through the simulated annealing algorithm, combines the search of the simulated annealing algorithm with job scheduling, and seeks the optimal job execution order. The present invention combines the simulated annealing algorithm with the requirements of the job scheduling scenario, searches for the optimal scheduling method, makes job scheduling more intelligent, and thus improves the scheduling efficiency to a certain extent. Moreover, in a supercomputer with a shared burst buffer architecture, the present invention can effectively utilize the burst buffer requirements of jobs, and can break through the backfilling limitations existing in traditional scheduling, improving job scheduling efficiency and system resource utilization rate.

Description

Technical Field

[0001] The present invention belongs to the technical field of supercomputers, and particularly relates to an intelligent job scheduling system for multi-objective optimization and its scheduling method. Background Art

[0002] In a high-performance computing (HPC) system, the coordinated utilization of computing resources and I / O (input / output) resources has always been a key factor affecting system efficiency. However, with the rapid growth of computing power, the improvement speed of I / O performance is relatively slow, and the performance gap between the two has become increasingly significant. To alleviate this problem, the Burst Buffer technology is introduced as a high-performance intermediate storage layer. It provides a cache layer between computing nodes and storage nodes, significantly improving the efficiency of I / O operations, thereby alleviating the imbalance between computing and storage.

[0003] Although the Burst Buffer technology shows significant advantages in improving I / O performance, the finiteness of its resources brings new challenges. Specifically, how to effectively utilize Burst Buffer resources during the job scheduling process while avoiding resource conflicts and fragmentation has become one of the key problems in HPC scheduling optimization. Currently, the mainstream scheduling strategies (such as first-come-first-served, FCFS, and its variants combined with backfilling technology) do not fully consider the reservation and optimized management of Burst Buffer resources, resulting in inefficient resource utilization and a decline in job execution performance.

[0004] However, currently, in the HPC scheduling manager, traditional scheduling strategies are mainly relied on to manage jobs and resource allocation. Its limitation is that the backfilling mechanism usually only focuses on the idle periods of computing resources and lacks fine management of the requirements for other resources such as Burst Buffer. Summary of the Invention

[0005] The object of the present invention is to: in view of the above problems, provide an intelligent job scheduling system for multi-objective optimization and its scheduling method, aiming to improve the limitations of the existing backfilling mechanism and enhance job scheduling efficiency and system resource utilization.

[0006] The technical solution adopted by the present invention is as follows: An intelligent job scheduling method for multi-objective optimization. The scheduling method searches for the Pareto solution set in multiple objectives through the simulated annealing algorithm, combines the search of the simulated annealing algorithm with job scheduling, and seeks the optimal job execution order. The scheduling method includes the following steps: The simulated annealing algorithm uses a random permutation of jobs as the initial solution, takes the initial solution as the starting point of the simulated annealing algorithm, and sets the initial temperature, temperature decay rate, and maximum number of iterations to optimize job scheduling. In each iteration process, the simulated annealing algorithm generates a new solution by locally perturbing the current initial solution. This new solution is the neighborhood solution, and the objective function value of the neighborhood solution is calculated. It is judged whether to accept the new neighborhood solution, and it is judged whether the neighborhood solution can be accepted according to the objective function value of the neighborhood solution. The acceptable neighborhood solutions form the Pareto optimal solution set. When the termination condition of the simulated annealing algorithm is reached, the algorithm is terminated; and the optimal neighborhood solution is returned as the optimal job scheduling solution.

[0007] Further, the initial temperature is the starting temperature of simulated annealing. During the algorithm iteration process, the initial temperature gradually decreases, the range of the neighborhood solution also becomes smaller, and finally it tends to be stable. The temperature decay rate is a parameter that controls the speed of temperature change, and the temperature decay rate is less than 1. The maximum number of iterations is set according to specific usage requirements.

[0008] Further, during the iteration process, the local perturbation method of the simulated annealing algorithm is: randomly select two jobs in the initial solution, exchange their positions, and form a new scheduling order as the neighborhood solution.

[0009] Further, after the neighborhood solution is generated by the simulated annealing algorithm, the objective function value of the neighborhood solution is calculated. Its objective function value includes three indicators. Whether the current scheduling order is the optimal job scheduling order is judged through each indicator of the objective function value. The indicators are as follows:

[0010] Total waiting time, which is the sum of the waiting times of all jobs in the queue;

[0011] Total slowdown ratio ,

[0012] ;

[0013] Among them, is the job waiting time, is the job running time, is the total slowdown ratio;

[0014] Total completion time, which is the maximum time when all jobs are completed minus the start time of the earliest job.

[0015] Further, the method for judging whether to accept the new neighborhood solution is:

[0016] Calculate the total difference between the objective function values of the new neighborhood solution and the previous neighborhood solution as , if , that is, the objective function value of the new neighborhood solution is better than that of the previous neighborhood solution, then accept the new neighborhood solution; if , that is, the objective function value of the new neighborhood solution is worse than that of the previous neighborhood solution, then selectively accept the new neighborhood solution according to the calculation of the probability formula;

[0017] The probability formula is as follows:

[0018] ;

[0019] Among them, is the difference between the objective function values of the new neighborhood solution and the previous neighborhood solution, is the current temperature;

[0020] The probability formula for selectively accepting the new neighborhood solution includes the following judgment: if the random number , then accept the new neighborhood solution, otherwise do not accept the new neighborhood solution;

[0021] Among them, the random number randomly generates a random number between 0 and 1, which is used to judge whether to accept a solution with a worse objective function value.

[0022] Furthermore, there is a set of neighborhood solutions in the Pareto optimal solution set, and the Pareto optimal solution set is constructed by judging the dominance relationship between the objective function values;

[0023] It should be noted that the dominance relationship here means that in the job scheduling scenario, each neighborhood solution corresponds to a set of objective function values, and these objective function values may represent different indicators of job scheduling, such as the waiting time and completion time of jobs. For any two neighborhood solutions, if one neighborhood solution is not worse than another neighborhood solution in all the indicators corresponding to the objective functions, and is better than another neighborhood solution in at least one indicator of the objective function, it is said that this neighborhood solution "dominates" another neighborhood solution.

[0024] The determination of its dominance relationship includes the following:

[0025] One-way comparison, when neighborhood solution A is greater than or equal to neighborhood solution B in all objectives, and is greater than neighborhood solution B in at least one objective, then neighborhood solution A dominates neighborhood solution B, where A has dominance and B does not have dominance;

[0026] Bidirectional comparison: The new neighborhood solution needs to be compared bidirectionally with all existing neighborhood solutions in the Pareto optimal solution set. If the new neighborhood solution is not dominated by any existing neighborhood solutions in the solution set, the new neighborhood solution is added to the Pareto optimal solution set, and the existing neighborhood solutions dominated by the new neighborhood solution in the Pareto optimal solution set are deleted; if the new neighborhood solution is directly dominated by any existing neighborhood solution, the new neighborhood solution is discarded.

[0027] When the termination condition of the simulated annealing algorithm is reached, the algorithm terminates, forming a Pareto optimal solution set, and all neighborhood solutions in the Pareto optimal solution set are non-dominant.

[0028] Furthermore, there are two termination conditions for the simulated annealing algorithm. One is that when the maximum number of iterations is reached, the algorithm terminates; the other is that when the temperature drops to the set threshold, the algorithm terminates; when the simulated annealing algorithm meets any one of the termination conditions, the algorithm terminates and returns the Pareto optimal solution set.

[0029] Furthermore, after several iterations and temperature decays, the simulated annealing algorithm will finally return a set of non-dominated Pareto optimal solution sets, which is denoted as the optimal job scheduling solution.

[0030] Even further, a multi-objective optimization intelligent job scheduling system, the system includes a waiting queue system and a multi-objective optimization system; the waiting queue system submits jobs by users and sends the jobs into the waiting queue for queuing, and the waiting queue automatically queues according to the order in which the jobs enter the system; the multi-objective optimization system optimizes the jobs within the time window in the mode of the time window, and generates the optimal job scheduling solution.

[0031] Furthermore, the multi-objective optimization system extracts the jobs in the waiting queue system for scheduling optimization, and the waiting queue system receives the optimal job scheduling solution generated by the multi-objective optimization system for processing and scheduling.

[0032] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are:

[0033] The present invention combines the simulated annealing algorithm with the requirements of the job scheduling scenario to search for the optimal scheduling method, making the job scheduling more intelligent, thereby improving the scheduling efficiency to a certain extent.

[0034] Moreover, in the supercomputer with a shared burst buffer architecture, the present invention can effectively utilize the burst buffer requirements of jobs and can break through the backfilling limitations existing in traditional scheduling, improving the job scheduling efficiency and system resource utilization rate. Description of the Drawings

[0035] Figure 1 is the flowchart of the method of the present invention;

[0036] Figure 2 is the schematic diagram of the system framework of the present invention. Detailed implementation manners

[0037] The present invention will be described in detail below with reference to the accompanying drawings.

[0038] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0039] As Figure 1 shown, a multi-objective optimization intelligent job scheduling method, the scheduling method searches for the Pareto solution set in multiple objectives through the simulated annealing algorithm, combines the search of the simulated annealing algorithm with job scheduling, and seeks the optimal job execution order; the scheduling method includes the following steps:

[0040] The simulated annealing algorithm uses a random permutation of jobs as the initial solution, takes the initial solution as the starting point of the simulated annealing algorithm, and sets the initial temperature, temperature decay rate, and maximum number of iterations to optimize job scheduling;

[0041] The initial temperature is the starting temperature of simulated annealing. At the beginning, a relatively high value is set for the initial temperature. During the algorithm iteration process, the initial temperature will gradually decrease, and the range of the neighborhood solution will also become smaller and finally tend to be stable;

[0042] The temperature decay rate is a parameter that controls the temperature change speed, and the temperature decay rate is less than 1. Generally, a value less than 1 is used, such as 0.95, indicating that the temperature will gradually decrease after each round of iteration;

[0043] The maximum number of iterations is set according to specific usage requirements.

[0044] During each iteration process, the simulated annealing algorithm generates a new solution by locally perturbing the current initial solution. This new solution is the neighborhood solution, and the objective function value of the neighborhood solution is calculated;

[0045] The local perturbation method of the simulated annealing algorithm is:

[0046] Randomly select two jobs in the initial solution, swap their positions, and form a new scheduling order as the neighborhood solution.

[0047] After the neighborhood solution is generated by the simulated annealing algorithm, the objective function value of the neighborhood solution is calculated. The objective function value includes three indicators. Whether the current scheduling order is the optimal job scheduling order is judged through the indicators of the objective function value. The indicators are as follows:

[0048] The total waiting time is the sum of the waiting times of all jobs in the queue;

[0049] The total reduction ratio ,

[0050] ;

[0051] Among them, is the job waiting time, is the job running time, is the total reduction ratio;

[0052] The total completion time is the maximum time when all jobs are completed minus the start time of the earliest job.

[0053] Judge whether to accept a new neighborhood solution, and judge whether the new neighborhood solution can be accepted according to the objective function value of the neighborhood solution;

[0054] The method for judging whether to accept a new neighborhood solution is as follows:

[0055] Calculate that the difference between the sum of the objective function values of the new neighborhood solution and the previous neighborhood solution is , if , that is, the objective function value of the new neighborhood solution is better than that of the previous neighborhood solution, then accept the new neighborhood solution; if , that is, the objective function value of the new neighborhood solution is worse than that of the previous neighborhood solution, then selectively accept the new neighborhood solution according to the calculation of the probability formula;

[0056] The probability formula is as follows:

[0057] ;

[0058] Among them, is the difference between the objective function values of the new neighborhood solution and the previous neighborhood solution, is the current temperature;

[0059] The probability formula selectively accepting the new neighborhood solution includes the following judgment: if the random number , then accept the new neighborhood solution, otherwise do not accept the new neighborhood solution;

[0060] Among them, the random number Randomly generate a random number between 0 and 1, which is used to judge whether to accept a solution with a worse objective function value.

[0061] Form a Pareto optimal solution set with the acceptable neighborhood solutions. When the termination condition of the simulated annealing algorithm is reached, terminate the algorithm; and return the optimal neighborhood solution as the optimal job scheduling solution.

[0062] There is a set of neighborhood solutions in the Pareto optimal solution set, and the Pareto optimal solution set is constructed by judging the dominance relationship between the objective function values;

[0063] The determination of its dominance relationship includes the following:

[0064] One-way comparison: When neighborhood solution A is greater than or equal to neighborhood solution B in all objectives and greater than neighborhood solution B in at least one objective, then neighborhood solution A dominates neighborhood solution B. Among them, A has dominance and B does not have dominance;

[0065] Two-way comparison: The new neighborhood solution needs to be compared with all existing neighborhood solutions in the Pareto optimal solution set in a two-way manner. If the new neighborhood solution is not dominated by any existing neighborhood solution in the solution set, the new neighborhood solution is added to the Pareto optimal solution set, and the existing neighborhood solutions dominated by the new neighborhood solution in the Pareto optimal solution set are deleted; If the new neighborhood solution is directly dominated by any existing neighborhood solution, then the new neighborhood solution is discarded;

[0066] When the termination condition of the simulated annealing algorithm is reached, the algorithm is terminated to form a Pareto optimal solution set, and all neighborhood solutions in the Pareto optimal solution set do not have dominance.

[0067] There are two termination conditions for the simulated annealing algorithm. One is that when the maximum number of iterations is reached, the algorithm terminates; the other is that when the temperature drops to the set threshold, the algorithm terminates; When the simulated annealing algorithm meets any one of the termination conditions, the algorithm terminates and returns the Pareto optimal solution set.

[0068] After several iterations and temperature decays, the simulated annealing algorithm will finally return a set of non-dominated Pareto optimal solution sets, and this Pareto optimal solution set is recorded as the optimal job scheduling solution.

[0069] This method introduces the resource demand characteristics of the Burst buffer in the job scheduling process, effectively allocates and uses the limited burst buffer resources, and uses the simulated annealing algorithm to find the optimal job execution order. Combined with the improved scheduling method, it can be more intelligent and quickly obtain a suitable optimal solution.

[0070] Example 1

[0071] As Figure 1 shown, an embodiment of the present invention is a multi-objective optimization intelligent job scheduling method. The scheduling method includes three scheduling plans, which are equivalent to the execution order of jobs:

[0072] Plan 1: Adopt the first-come-first-served plus backfilling method, where each backfilling needs to ignore the negative impact that the backfilled job will delay the subsequent first job, that is, each backfilling directly selects the first most suitable job from the subsequent waiting jobs for backfilling;

[0073] Plan 2: Adopt the method of shortest job first with backfilling. The backfilling mechanism still adopts the method of Plan 1. Note that by default, jobs with short running times are arranged at the front of the entire queue.

[0074] Plan 3: Adopt the mechanism of custom job order with backfilling. The backfilling mechanism adopts the method of Plan 1. The queue order is determined by the custom job order, and the jobs at the front have the qualification for priority backfilling.

[0075] Based on Plan 3, this embodiment combines the simulated annealing algorithm to search for the Pareto solution set in multi-objectives, combines the search of the simulated annealing algorithm with job scheduling, and seeks the optimal job execution order.

[0076] The scheduling method includes the following steps:

[0077] 1. Initial stage;

[0078] Arrange the jobs in a random order as the initial solution. Each job has a job_id, and the execution time, the number of required nodes, and the size of the burst buffer of each job are known. This initial solution is used as the starting point of the algorithm.

[0079] Initial temperature: The starting temperature of simulated annealing, usually set to a relatively high value.

[0080] Temperature decay rate: A parameter that controls the rate of temperature decrease. Generally, a value less than 1 is adopted, such as 0.95, indicating that the temperature will gradually decrease after each iteration.

[0081] Maximum number of iterations: To avoid the algorithm running for too long, a maximum number of iterations is usually set to ensure that the algorithm terminates within a reasonable time.

[0082] 2. Generation of neighborhood solutions;

[0083] In each iteration process, the simulated annealing algorithm generates a new solution by making a local perturbation to the current solution. The specific perturbation method is: randomly select two jobs in the current solution and swap their positions to form a new scheduling order. This new order is called the "neighborhood solution".

[0084] 3. Calculation of the objective function;

[0085] Whenever a new neighborhood solution is generated, the objective function value of this solution is calculated. Among them, the objective function includes three main indicators:

[0086] (1) Total waiting time: The sum of the waiting times of all jobs in the queue.

[0087] (2) Total completion time: The maximum time when all jobs are completed minus the start time of the earliest job.

[0088] (3) Total slowdown ratio: ;

[0089] Among them, is the job waiting time, is the job running time, is the total slowdown ratio;

[0090] These objective function values can reflect the quality of the scheduling scheme, and the performance of the current scheduling order is judged by calculating the objective values of each solution.

[0091] 4. Criteria for receiving new solutions;

[0092] The key part of simulated annealing is how to decide whether to accept a new neighborhood solution. The judgment method is as follows: Calculate the difference in the total objective function values between the new neighborhood solution and the previous neighborhood solution as If , that is, the objective function value of the new neighborhood solution is better than that of the previous neighborhood solution, then accept the new neighborhood solution; if , that is, the objective function value of the new neighborhood solution is worse than that of the previous neighborhood solution, then selectively accept the new neighborhood solution according to the calculation of the probability formula;

[0093] The probability formula is as follows:

[0094] ;

[0095] Among them, is the difference in the objective function values between the new neighborhood solution and the previous neighborhood solution, is the current temperature;

[0096] The probability formula selectively accepting the new neighborhood solution includes the following judgment: If the random number , then accept the new neighborhood solution, otherwise do not accept the new neighborhood solution;

[0097] Among them, the random number randomly generates a random number between 0 and 1, which is used to judge whether to accept a solution with a worse objective function value. This random number is generated by a pseudo-random number generator built into the algorithm, and its generation logic follows the uniform distribution characteristic, that is, the probability of each value appearing is equal.

[0098] 5. Pareto optimization;

[0099] In multi-objective optimization problems, there is often no single optimal solution, but rather a set of optimal solutions, called the Pareto optimal solution set. Therefore, the present invention constructs the Pareto optimal solution set by judging the dominance relationship between objective values. Specifically, if a solution is not inferior to the current solution in all objectives and is superior to the current solution in at least one objective, the current solution is considered a dominant solution. Through such a dominance relationship judgment, the algorithm can gradually construct a Pareto optimal solution set and maintain the non-dominated solutions in the solution set.

[0100] 6. Temperature Decay and Iteration;

[0101] After each iteration, the temperature decreases according to the set temperature decay rate, usually set to 0.95. As the temperature gradually decreases, the range of the algorithm to explore new solutions becomes smaller, and finally it tends to be stable and converges to a better solution.

[0102] 7. Termination Conditions;

[0103] There are usually two termination conditions for the simulated annealing algorithm: one is to reach the maximum number of iterations, and the other is that the temperature drops to a certain threshold. The algorithm stops when either of these conditions is met and returns the optimal scheduling solution.

[0104] 8. Output of the Final Solution;

[0105] After several iterations and temperature decays, the simulated annealing algorithm will finally return a set of non-dominated solutions (Pareto optimal solution set). These solutions represent the optimal scheduling order under different weight configurations in the multi-objective optimization problem. According to specific requirements, a suitable solution can be selected as the final scheduling order.

[0106] 9. Comprehensively consider the optimization indicators corresponding to the three plans. First, standardize each objective to unify the dimension, and then use the triple weight parameters ( , , ), which correspond to the weight parameters of each optimization indicator respectively, to find the set of optimization indicators with the minimum weighted sum. The corresponding job execution order is the optimal one, where the weight parameters are determined according to specific usage requirements.

[0107] In this embodiment, the resource requirement characteristics of the Burst buffer are introduced in the job scheduling process to effectively allocate and utilize the limited Burst buffer resources. At the same time, the simulated annealing algorithm is used to find the optimal job execution order, and combined with the improved scheduling method, the original job scheduling method becomes more intelligent and can quickly obtain a suitable optimal solution.

[0108] Embodiment 2

[0109] As Figure 2As shown in the figure, another embodiment of the present invention is a multi-objective optimization intelligent job scheduling system, which includes a waiting queue system and a multi-objective optimization system;

[0110] The waiting queue system receives jobs submitted by users and sends them into the waiting queue for queuing. The waiting queue automatically queues according to the order in which jobs enter the system;

[0111] The multi-objective optimization system intelligently schedules and optimizes the jobs within the time window in accordance with the time window mode to generate the optimal job scheduling solution.

[0112] The multi-objective optimization system extracts the jobs in the waiting queue system for scheduling optimization, and the waiting queue system receives the optimal job scheduling solution generated by the multi-objective optimization system for processing and scheduling.

[0113] Through the interaction of these two systems, they jointly perform intelligent job scheduling with multi-objective optimization, and can effectively utilize the Burst buffer requirements of jobs in a supercomputer with a Burstbuffer architecture, breaking the backfilling limitations existing in traditional scheduling, and improving job scheduling efficiency and system resource utilization rate.

[0114] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent job scheduling method for multi-objective optimization, characterized in that, The described scheduling method searches for the Pareto solution set in multiple objectives through the simulated annealing algorithm, combines the search of the simulated annealing algorithm with job scheduling, and seeks the optimal job execution order; the scheduling method includes the following steps: The simulated annealing algorithm takes a randomly arranged job as the initial solution, uses the initial solution as the starting point of the simulated annealing algorithm, and sets the initial temperature, temperature decay rate, and maximum number of iterations to optimize job scheduling; In each iteration process, the simulated annealing algorithm generates a new solution by locally perturbing the current initial solution. This new solution is the neighborhood solution, and the objective function value of the neighborhood solution is calculated; Judge whether to accept the new neighborhood solution, and judge whether the neighborhood solution can be accepted according to the objective function value of the neighborhood solution; Form the Pareto optimal solution set with the acceptable neighborhood solutions. When the termination condition of the simulated annealing algorithm is reached, terminate the algorithm; and return the optimal neighborhood solution as the optimal job scheduling solution; During the iteration process, the local perturbation method of the simulated annealing algorithm is as follows: Randomly select two jobs in the initial solution, swap their positions, and form a new scheduling order as the neighborhood solution; After the neighborhood solution is generated by the simulated annealing algorithm, calculate the objective function value of the neighborhood solution. Its objective function value includes three indicators, and judge whether the current scheduling order is the optimal job scheduling order through each indicator of the objective function value. The indicators are as follows: Total waiting time, which is the sum of the waiting times of all jobs in the queue; Overall reduction ratio , ； Among them, is the job waiting time, is the job running time, is the total reduction ratio; Total completion time, which is the maximum time when all jobs are completed minus the start time of the earliest job; The method for judging whether to accept the new neighborhood solution is as follows: Calculate the total difference in the objective function values between the new neighborhood solution and the previous neighborhood solution as , if , that is, the objective function value of the new neighborhood solution is better than that of the previous neighborhood solution, then accept the new neighborhood solution; if , that is, the objective function value of the new neighborhood solution is worse than that of the previous neighborhood solution, then selectively accept the new neighborhood solution according to the calculation of the probability formula; The probability formula is as follows: ； Among them, is the difference between the objective function values of the new neighborhood solution and the previous neighborhood solution, is the current temperature; The probability formula selectively accepts a new neighborhood solution, including the following judgment: if the random number , then accept the new neighborhood solution; otherwise, do not accept the new neighborhood solution. Among them, the random number Randomly generate a random number between 0 and 1, which is used to determine whether to accept a solution with a worse objective function value.

2. The intelligent job scheduling method for multi-objective optimization according to claim 1, characterized in that, The initial temperature is the starting temperature of the simulated annealing. During the algorithm iteration process, the initial temperature will gradually decrease, and the range of the neighborhood solution will also become smaller and finally tend to be stable; the temperature decay rate is a parameter that controls the temperature change speed, and the temperature decay rate is less than 1; the maximum number of iterations is set according to specific usage requirements.

3. The intelligent job scheduling method for multi-objective optimization according to claim 1, characterized in that, There is a set of neighborhood solutions in the Pareto optimal solution set, and the Pareto optimal solution set is constructed by judging the dominance relationship between the objective function values; The determination of its dominance relationship includes the following: One-way comparison. When neighborhood solution A is greater than or equal to neighborhood solution B in all objectives and greater than neighborhood solution B in at least one objective, then neighborhood solution A dominates neighborhood solution B. Among them, A has dominance and B does not have dominance; Two-way comparison. The new neighborhood solution needs to be compared with all existing neighborhood solutions in the Pareto optimal solution set in both directions. If the new neighborhood solution is not dominated by any existing neighborhood solution in the solution set, add the new neighborhood solution to the Pareto optimal solution set and delete the existing neighborhood solutions in the Pareto optimal solution set that are dominated by the new neighborhood solution; if the new neighborhood solution is directly dominated by any existing neighborhood solution, discard the new neighborhood solution; When the termination condition of the simulated annealing algorithm is reached, terminate the algorithm, form the Pareto optimal solution set, and all neighborhood solutions in the Pareto optimal solution set do not have dominance.

4. An intelligent job scheduling method for multi-objective optimization according to claim 1, characterized in that, There are two termination conditions for the simulated annealing algorithm. One is that when the maximum number of iterations is reached, the algorithm terminates; the other is that when the temperature drops to the set threshold, the algorithm terminates. When the simulated annealing algorithm meets any one of the termination conditions, the algorithm terminates and returns the Pareto optimal solution set.

5. An intelligent job scheduling method for multi-objective optimization according to claim 4, characterized in that After several iterations and temperature decays, the simulated annealing algorithm will finally return a set of non-dominated Pareto optimal solution sets, which is denoted as the optimal job scheduling solution.

6. An intelligent job scheduling system for multi-objective optimization, which uses a multi-objective optimization intelligent job scheduling method according to any one of claims 1-5, characterized in that, The system includes a waiting queue system and a multi-objective optimization system; The waiting queue system submits jobs by users and sends the jobs into the waiting queue for queuing. The waiting queue automatically queues according to the order in which the jobs enter the system; The multi-objective optimization system intelligently schedules and optimizes the jobs within the time window in the mode of the time window to generate the optimal job scheduling solution.

7. An intelligent job scheduling system for multi-objective optimization according to claim 6, characterized in that, The multi-objective optimization system extracts the jobs in the waiting queue system for scheduling optimization, and the waiting queue system receives the optimal job scheduling solution generated by the multi-objective optimization system for processing and scheduling.

Images