CN114860435B - Big data job scheduling method based on task selection process reinforcement learning - Google Patents

Big data job scheduling method based on task selection process reinforcement learning Download PDF

Info

Publication number
CN114860435B
CN114860435B CN202210449623.1A CN202210449623A CN114860435B CN 114860435 B CN114860435 B CN 114860435B CN 202210449623 A CN202210449623 A CN 202210449623A CN 114860435 B CN114860435 B CN 114860435B
Authority
CN
China
Prior art keywords
task
job
node
scheduled
scheduling
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
Application number
CN202210449623.1A
Other languages
Chinese (zh)
Other versions
CN114860435A (en
Inventor
夏莹杰
武建伟
陈天祥
刘瑞峰
张雷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taizhou Jiema Technology Co ltd
Research Institute of Zhejiang University Taizhou
Original Assignee
Taizhou Jiema Technology Co ltd
Research Institute of Zhejiang University Taizhou
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Taizhou Jiema Technology Co ltd, Research Institute of Zhejiang University Taizhou filed Critical Taizhou Jiema Technology Co ltd
Priority to CN202210449623.1A priority Critical patent/CN114860435B/en
Publication of CN114860435A publication Critical patent/CN114860435A/en
Application granted granted Critical
Publication of CN114860435B publication Critical patent/CN114860435B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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]
    • G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
    • G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
    • G06F9/5038—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the execution order of a plurality of tasks, e.g. taking priority or time dependency constraints into consideration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a big data job scheduling method based on reinforcement learning in a task selection process, which comprises the following steps: s1, carrying out vectorization processing on the jobs in the cluster; s2, for the selected task node, the agent determines how many task instances of the selected task node are scheduled; s3, selecting a task executor type with the best matched resource from the selected task nodes by the resource matching module, and then selecting an idle task executor from the task executor types to prepare to execute a currently scheduled task instance; and S4, deploying and executing the scheduled task instance on the big data task execution platform. The method has strong adaptability and universality for big data operations of different types.

Description

Big data job scheduling method based on task selection process reinforcement learning
Technical Field
The invention relates to a big data job scheduling method based on reinforcement learning in a task selection process, and belongs to the field of artificial intelligence.
Background
The big data application is the utilization of the data value, namely effective information is mined from mass data through data analysis, and decision support is provided for users. How to process these data efficiently becomes critical, and scheduling between tasks in data processing has a great significance to overall performance and resource utilization.
In the workflow, the dependency relationship between task nodes becomes more complex as the number of nodes increases. However, the traditional heuristic algorithm cannot well process complex dependency relationships among task nodes, and only can process the complex dependency relationships through artificial feature engineering construction feature values, so that the artificial influence is great. The graph convolution neural network can autonomously learn task node characteristics and deeply excavate relations in a topological structure, so that human interference is avoided. Therefore, the graph convolution neural network is used for task dependency processing in the workflow scheduling problem, so that the algorithm scheduling effect is improved, and the hot spot direction of the study is achieved.
At present, the existing method based on the graph convolution neural network can only schedule and execute large data jobs of a specific type (running on a Hadoop computing cluster), cannot execute dynamic parameter adjustment on scheduled tasks to optimize a scheduling process, and has great limitation.
Disclosure of Invention
The invention aims to solve the defects in the background art, and is used for completing the dispatching of big data operation in a dynamic workflow environment and improving the overall completion efficiency of the operation.
In order to achieve the above object, the present invention provides a big data job scheduling method based on reinforcement learning in a task selection process, comprising the steps of: s1, carrying out vectorization processing on the jobs in the cluster: if the operation set is not completed, inputting the embedded vector into the reinforcement learning module for decision making; after the full-connection layer processing is completed, the decision network for scheduling carries out a differential value reprocessing process on each task node; finally, the agent of the reinforcement learning module selects the next scheduled task node by setting a softmax layer; s2, for the selected task node, the agent determines how many task instances of the selected task node are scheduled; the corresponding decision network reduces the probability that the job which cannot be scheduled is selected after the full connection layer is processed and at the softmax layer, and takes the number of task executors with the highest probability of the selected task nodes as the number of task instances of specific scheduling; s3, selecting a task executor type with the best matched resource from the selected task nodes by the resource matching module, and then selecting an idle task executor from the task executor types to prepare to execute a currently scheduled task instance; s4, deploying and executing the scheduled task instance on the big data task execution platform: after the execution of the job subset of the current batch is finished, quantifying the rewarding value of the action according to the whole dispatching time of the round of dispatching; if the decision of the agent improves the scheduling effect, the agent will obtain a positive prize value and increase the probability of selecting the decision afterwards; otherwise, the agent will obtain a reverse prize value and reduce the probability of later selecting the decision.
Further, the step S1 includes: s11, converting the graph by using a graph convolution neural network, wherein the processed embedded vector is divided into three classes of node level, operation level and global level; and S12, after the embedded vector is processed by the last full connection layer of the decision network, judging whether each task node can be scheduled. If the scheduling is impossible, reducing vector values corresponding to the task nodes; otherwise, the state is maintained unchanged; and S13, finally, the decision network sets a softmax layer for further processing, and selects the next scheduled task node.
Further, the step S2 includes: s21, judging whether the job of the selected task node is selectable, if the job of the task node cannot be selected, reducing the value corresponding to the job after the decision network full-connection layer processing; otherwise, the state is maintained unchanged; s22, the decision network reduces the probability that the job which cannot be scheduled is selected through further processing of a softmax layer; s23, according to the job where the selected task node is located, calculating a corresponding row vector in the matrix, and selecting the number of task executors with highest probability in the row vector as the number of task instances of specific scheduling.
Further, the step S3 includes: s31, when the agent determines the next scheduled task node, the resource matching module calculates the matching value of the task node and the task executor category, and selects the task executor category corresponding to the maximum matching value; the S32 resource matching module selects an idle task executor from the selected task executor categories to prepare to execute the currently scheduled task instance.
Further, the step S4 includes: s41, after the agent determines the next scheduled task node and the corresponding task node instance, a task executor selected by the resource matching module executes the given task instance in the big data job cluster; s42, after the current scheduling period is finished, using the whole scheduling time of the job as an independent variable, and obtaining a corresponding rewarding value by using a rewarding function; the agent utilizes the feedback of the forward or reverse prize value to learn constantly to explore a series of optimal actions to maximize the cumulative prize expectation.
Compared with the prior art, the invention has the beneficial effects that:
the adaptability is stronger, and the method has universality for big data jobs of different types, such as MapReduce, spark and the like.
Drawings
FIG. 1 is a decision network block diagram of task node scheduling in accordance with one embodiment of the present invention;
FIG. 2 is a block diagram of a resource matching module decision making according to one embodiment of the invention.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1 and fig. 2, in one embodiment of the present invention, step 1, vectorizing a job in a cluster. If the operation set is not completed, the embedded vector is input into the reinforcement learning module to make a decision. And (3) carrying out a scheduling decision network, and carrying out a differential value reprocessing process on each task node after the full connection layer processing is completed. Finally, the agent of the reinforcement learning module selects the next scheduled task node by setting a softmax layer.
The graph is transformed by using a graph convolution neural network, and the processed embedded vectors are divided into three classes of node level, operation level and global level.
The task node level includes: task node v i Execution time dur of each task instance in (3) i The method comprises the steps of carrying out a first treatment on the surface of the Task node v i Is the total task instance number inst of (2) i The method comprises the steps of carrying out a first treatment on the surface of the Task nodev i The remaining task instance number rest of (2) i The method comprises the steps of carrying out a first treatment on the surface of the Task node v i Priority rank of a network i The method comprises the steps of carrying out a first treatment on the surface of the Task node v i The required core for CPU core number per task instance i The method comprises the steps of carrying out a first treatment on the surface of the Task node v i The memory size requirements mem for each task instance of (a) i . The job level includes: the total number of task executors used_num_exec that the current job n already occupies n The current global remaining idle task executor number gl_rest_num_exec. The global level is calculated by calculating embedded vectors of a plurality of job levels using the same calculation formula on the basis of the job level.
The decision network for selecting the next scheduled task node consists of 4 full connection layers and 3 leak ReLU activation functions. The 4 fully connected layers contain 32, 16, 4, 1 neurons, respectively.
The values maintained for the task nodes, including the remaining instance number rest_num_inst, the remaining execution time rest_exec_time, will vary as the agent continues to perform scheduling operations. Both are represented using two-dimensional vectors as follows:
(rest_num_inst,rest_exec_time)
the rest_num_inst in the formula is obtained by subtracting the number of task instances which are already executed and completed by each task node maintained by the scheduling algorithm from the total number of the task nodes, and the rest_exec_time is obtained by multiplying the rest_num_inst by the execution time dur required by the task instance.
After the embedded vector is processed by the last full connection layer of the decision network, whether each task node can be scheduled or not is judged. If the scheduling is impossible, the value corresponding to the node is subtracted by 1e in the matrix after the full connection layer processing 10 . If the task node can be scheduled, the state is maintained.
The decision network calculates a matrix dimension [ total_num_task_nodes,1] in which each task node in each job has a corresponding value. This value represents the probability that the task node is selected for scheduling.
Finally, the decision network sets a softmax layer for further processing, and selects the next scheduled task node.
In one specific example, assuming 10 task nodes, the final calculated result matrix of the decision network is as follows:
at the 8 th task node (with probability of 0.5723) with the highest corresponding probability is selected.
Fig. 1 is a block diagram of a decision network for task node scheduling in accordance with one embodiment of the present invention.
Step 2, further, for the selected task node, the agent determines how many task instances to schedule. And the corresponding decision network reduces the probability that the job which cannot be scheduled is selected after the full connection layer is processed and at the softmax layer, and takes the number of task executors with the highest probability of the selected task nodes as the number of task instances of specific scheduling.
And judging whether the job where the selected task node is located is selectable. If the job where the task node is located cannot be selected, reducing the value corresponding to the job after the decision network full-connection layer processing; otherwise, remain unchanged.
The information that the agent maintains for the job includes the total number of task executors used by the job, used_num_exec, and the total number of idle task executors gl_rest_num_exec, and these information are stored using two-dimensional vectors. The specific contents are as follows:
(used_num_exec,gl_rest_num_exec)
the state information of the task executors needs to contain the current residual quantity of each task executor, and the vector is used for representing the state information as follows:
(free_num_exec_type 1 ,free_num_exec_type 2 ,...,free_num_exec_type k ) The calculation formula of the global remaining idle task executor number gl_rest_num_exec of the job is as follows:
as shown in the above formula, the global remaining task executor number is the sum of the remaining idle task executors number of all dimensions.
The decision network for performing the job selection judgment is also composed of 4 full connection layers and 3 leak ReLU activation functions. The fully connected layers contain 32, 16, 4, 1 neurons, respectively. If the job of the task node cannot be selected, the value corresponding to the job is subtracted by 1e in the matrix after the full link layer processing 10 If the job can be selected, the job remains as it is. The reason that a job cannot be selected is either that the job has been executed to completion or that the global remaining task executor number gl_rest_num_exec is 0.
The final decision network reduces the probability that an unscheduled job is selected through further processing by the softmax layer.
The dimension of the calculation result of the decision network after shape reconstruction becomes [ total_num_jobs ], total_num_exec ]. In the matrix, each job has a corresponding row vector, the dimension in the row vector represents how many task executors can be allocated to the job, and the corresponding value represents the probability of allocating the number of task executors. Because the agent finally needs to determine the number of task instances specifically scheduled in the currently selected task node, the scheduling algorithm needs to calculate the corresponding row vector in the matrix according to the job in which the selected task node is located, and select the number of task executors with the highest probability in the row vector as the number of task instances specifically scheduled.
In one embodiment, assuming that the task node selected in step 1 is x, the task executor row vector of the corresponding job is
(4,3,2,5,1)
The corresponding probability line vector is
(0.35,0.12,0.27,0.15,0.11)
The task executor number with the highest corresponding probability is 4 (the probability is 0.35), namely, 4 task executors are selected to execute the job corresponding to the task node x.
And 3, selecting the task executor type with the best matching resource from the selected task nodes by the resource matching module. An idle task executor is then selected from the task executor classes to prepare to execute the currently scheduled task instance.
When the agent determines that the next scheduled task node is good, the resource matching module calculates matching values of the task node and task executor types, and selects the task executor type corresponding to the maximum matching value.
The demand of the task instance for the resources is divided into two parts, namely a CPU and a memory. The scheduling algorithm needs to make sure whether the whole job is focused on preferentially satisfying the CPU or preferentially satisfying the memory. If the CPU resources are more intense, the CPU should be preferentially satisfied, otherwise, the memory should be preferentially satisfied.
The emphasis is quantified by using a weight coefficient alpha, wherein alpha is the weight which preferentially satisfies the CPU, and 1-alpha is the weight which preferentially satisfies the memory. The calculation formula is as follows:
the method C (&) in the formula is used for calculating the requirement of task nodes on the CPU core number or the CPU core number configuration of the task executor, V represents all task node sets in all jobs, E represents the types of all task executors, and N (E) represents the number of E-type task executors. The M (& gt) method in the formula is used for calculating the requirement of the task node on the memory size or the memory configuration of the task executor.
When the agent determines that the next scheduled task node v is good, the resource matching module will calculate the task nodeMatching value score to task executor class e v,e ：
The resource matching module selects the task executor category corresponding to the maximum matching value for the task node v, namely:
further, the resource matching module selects an idle task preparation executor from the selected task executor categories to execute the currently scheduled task instance.
FIG. 2 is a block diagram illustrating a decision of a resource matching module according to an embodiment of the present invention.
And step 4, deploying and executing the scheduled task instance on the big data task execution platform. And after the execution of the job subset of the current batch is finished, quantifying the rewarding value of the action according to the whole dispatching time of the round of dispatching. If the decision of the agent improves the scheduling effect, the agent will obtain a positive prize value and increase the probability of selecting the decision afterwards; otherwise, the agent will obtain a reverse prize value and reduce the probability of later selecting the decision.
When the agent determines that the next scheduled task node and the corresponding task node instance are good, the task executor selected by the resource matching module will execute the given task instance in the big data job cluster.
And after the current scheduling period is finished, the whole scheduling time of the job is taken as an independent variable, and a corresponding rewarding value is obtained by using a rewarding function. The agent utilizes the feedback of the forward or reverse prize value to learn constantly to explore a series of optimal actions to maximize the cumulative prize expectation.
Once all the set actions have been performed, the environment updates the global time point to gtime. The previous global time point is recorded as prev_gtime, and the initial global time point is recorded as 0. The bonus function is designed as follows:
r＝-(gtime-prev_gtime)*r_s
r_s in the formula is a hyper-parameter.
Finally, it should be noted that: the above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.

Claims (3)

1. The big data job scheduling method based on the task selection process reinforcement learning is characterized by comprising the following steps:
s1, carrying out vectorization processing on the jobs in the cluster: if the operation set is not completed, inputting the embedded vector into the reinforcement learning module for decision making; after the full-connection layer processing is completed, the decision network for scheduling carries out a differential value reprocessing process on each task node; finally, the agent of the reinforcement learning module selects the next scheduled task node by setting a softmax layer;
s2, for the selected task node, the agent determines how many task instances of the selected task node are scheduled; the corresponding decision network reduces the probability that the job which cannot be scheduled is selected after the full connection layer is processed and at the softmax layer, and takes the number of task executors with the highest probability of the selected task nodes as the number of task instances of specific scheduling;
s3, selecting a task executor type with the best matched resource from the selected task nodes by the resource matching module, and then selecting an idle task executor from the task executor types to prepare to execute a currently scheduled task instance;
s4, deploying and executing the scheduled task instance on the big data task execution platform: after the execution of the job subset of the current batch is finished, quantifying the rewarding value of the action according to the whole dispatching time of the round of dispatching; if the decision of the agent improves the scheduling effect, the agent will obtain a positive prize value and increase the probability of selecting the decision afterwards; otherwise, the agent will obtain a reverse prize value and reduce the probability of later selecting the decision;
the step S1 includes:
s11, converting the graph by using a graph convolution neural network, wherein the processed embedded vector is divided into three classes of node level, operation level and global level;
and S12, after the embedded vector is processed by the last full connection layer of the decision network, judging whether each task node can be scheduled. If the scheduling is impossible, reducing vector values corresponding to the task nodes; otherwise, the state is maintained unchanged;
s13, finally, the decision network sets a softmax layer for further processing, and selects a next scheduled task node;
the step S4 includes:
s41, after the agent determines the next scheduled task node and the corresponding task node instance, a task executor selected by the resource matching module executes the given task instance in the big data job cluster;
s42, after the current scheduling period is finished, using the whole scheduling time of the job as an independent variable, and obtaining a corresponding rewarding value by using a rewarding function; the agent utilizes the feedback of the forward or reverse prize value to learn constantly to explore a series of optimal actions to maximize the cumulative prize expectation.
2. The big data job scheduling method based on the task selection process reinforcement learning according to claim 1, wherein the step S2 includes:
s21, judging whether the job of the selected task node is selectable, if the job of the task node cannot be selected, reducing the value corresponding to the job after the decision network full-connection layer processing; otherwise, the state is maintained unchanged;
s22, the decision network reduces the probability that the job which cannot be scheduled is selected through further processing of a softmax layer;
s23, according to the job where the selected task node is located, calculating a corresponding row vector in the matrix, and selecting the number of task executors with highest probability in the row vector as the number of task instances of specific scheduling.
3. The big data job scheduling method based on the task selection process reinforcement learning according to claim 2, wherein the step S3 includes:
s31, when the agent determines the next scheduled task node, the resource matching module calculates the matching value of the task node and the task executor category, and selects the task executor category corresponding to the maximum matching value;
s32, selecting idle task execution from the selected task executor types by the resource matching module
The walker prepares to execute the currently scheduled task instance.
CN202210449623.1A 2022-04-24 2022-04-24 Big data job scheduling method based on task selection process reinforcement learning Active CN114860435B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210449623.1A CN114860435B (en) 2022-04-24 2022-04-24 Big data job scheduling method based on task selection process reinforcement learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210449623.1A CN114860435B (en) 2022-04-24 2022-04-24 Big data job scheduling method based on task selection process reinforcement learning

Publications (2)

Publication Number Publication Date
CN114860435A CN114860435A (en) 2022-08-05
CN114860435B true CN114860435B (en) 2024-04-05

Family

ID=82633031

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210449623.1A Active CN114860435B (en) 2022-04-24 2022-04-24 Big data job scheduling method based on task selection process reinforcement learning

Country Status (1)

Country Link
CN (1) CN114860435B (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114281528A (en) * 2021-12-10 2022-04-05 重庆邮电大学 Energy-saving scheduling method and system based on deep reinforcement learning and heterogeneous Spark cluster

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102020204351A1 (en) * 2020-04-03 2021-10-07 Robert Bosch Gesellschaft mit beschränkter Haftung DEVICE AND METHOD FOR PLANNING A MULTIPLE ORDERS FOR A VARIETY OF MACHINERY

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114281528A (en) * 2021-12-10 2022-04-05 重庆邮电大学 Energy-saving scheduling method and system based on deep reinforcement learning and heterogeneous Spark cluster

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
亢中苗 ; 汪莹 ; 张珮明 ; 陶志强 ; 李家樑 ; .云计算环境下基于强化学习的虚拟机资源调度.自动化与仪器仪表.2020,(10),全文. *

Also Published As

Publication number Publication date
CN114860435A (en) 2022-08-05

Similar Documents

Publication Publication Date Title
Zhu et al. An efficient evolutionary grey wolf optimizer for multi-objective flexible job shop scheduling problem with hierarchical job precedence constraints
CN108880663B (en) Space-ground integrated network resource allocation method based on improved genetic algorithm
Shen et al. Mathematical modeling and multi-objective evolutionary algorithms applied to dynamic flexible job shop scheduling problems
CN114186749B (en) Flexible workshop scheduling method and model based on reinforcement learning and genetic algorithm
CN113792924A (en) Single-piece job shop scheduling method based on Deep reinforcement learning of Deep Q-network
Xiao et al. A cooperative coevolution hyper-heuristic framework for workflow scheduling problem
CN104077634B (en) active-reactive type dynamic project scheduling method based on multi-objective optimization
CN110969362A (en) Multi-target task scheduling method and system under cloud computing system
CN117271101B (en) Operator fusion method and device, electronic equipment and storage medium
Zhang et al. An integer-coded differential evolution algorithm for simple assembly line balancing problem of type 2
CN111047272A (en) Project scheduling method and device for multi-language collaborative development
CN111813500A (en) Multi-target cloud workflow scheduling method and device
CN115168027A (en) Calculation power resource measurement method based on deep reinforcement learning
CN111831355A (en) Weight precision configuration method, device, equipment and storage medium
CN113157421A (en) Distributed cluster resource scheduling method based on user operation process
CN116700176A (en) Distributed blocking flow shop scheduling optimization system based on reinforcement learning
CN115543626A (en) Power defect image simulation method adopting heterogeneous computing resource load balancing scheduling
CN111831359A (en) Weight precision configuration method, device, equipment and storage medium
CN115827189A (en) Large-scale intelligent cluster task scheduling optimization method
CN111831354A (en) Data precision configuration method, device, chip array, equipment and medium
Nascimento et al. A reinforcement learning scheduling strategy for parallel cloud-based workflows
Ishankhodjayev et al. Development of information support for decision-making in intelligent energy systems
CN116938323B (en) Satellite transponder resource allocation method based on reinforcement learning
CN114860435B (en) Big data job scheduling method based on task selection process reinforcement learning
CN110362378A (en) A kind of method for scheduling task and equipment

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
GR01 Patent grant
GR01 Patent grant