CN110830294A - Edge calculation task allocation method based on branch-and-bound method - Google Patents

Edge calculation task allocation method based on branch-and-bound method Download PDF

Info

Publication number
CN110830294A
CN110830294A CN201911063618.1A CN201911063618A CN110830294A CN 110830294 A CN110830294 A CN 110830294A CN 201911063618 A CN201911063618 A CN 201911063618A CN 110830294 A CN110830294 A CN 110830294A
Authority
CN
China
Prior art keywords
task
edge
tasks
solution
tempval
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.)
Granted
Application number
CN201911063618.1A
Other languages
Chinese (zh)
Other versions
CN110830294B (en
Inventor
裴廷睿
李梦瑶
田淑娟
邹娟
曹江莲
关屋大雄
崔荣埈
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.)
Xiangtan University
Original Assignee
Xiangtan University
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 Xiangtan University filed Critical Xiangtan University
Priority to CN201911063618.1A priority Critical patent/CN110830294B/en
Publication of CN110830294A publication Critical patent/CN110830294A/en
Application granted granted Critical
Publication of CN110830294B publication Critical patent/CN110830294B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0803—Configuration setting
    • H04L41/0823—Configuration setting characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0893—Assignment of logical groups to network elements
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/1097—Protocols in which an application is distributed across nodes in the network for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]

Abstract

The invention provides an edge calculation task allocation method based on a branch-and-bound method, and belongs to the field of edge calculation. The invention minimizes the total energy consumption of task allocation under the condition of considering the constraints of task completion time, DAG parallel system reliability requirement and the like. Firstly, relaxing the optimization problem, and solving a temporary solution by using an interior point method; and then taking discrete values for the decision variables of the first task in the temporary solution and meeting the constraint that one task can only be executed on one edge server, namely only one of the decision variables is taken as 1, and the others are all 0, traversing the value taking situation from the first edge server to the last edge server, continuously adopting an interior point method for the rest tasks to solve and calculate the energy consumption values of different solutions, taking the solution with the minimum energy consumption value, and repeating the step until the last task. The invention has the advantages of good convergence, small calculated amount and the like.

Description

Edge calculation task allocation method based on branch-and-bound method
Technical Field
The invention relates to the field of edge calculation, in particular to an edge calculation task distribution method based on a branch-and-bound method.
Background
With the further exuberance of new production requirements such as internet of things and industrial internet of things, a centralized data processing technology taking a cloud computing model as a core cannot efficiently process a large amount of data generated by edge devices. Edge computing is a new type of computing model that performs computing at the edge of a network, and the edge of edge computing refers to any computing and network resources between the path from a data source to a cloud computing center. For example, a smart home gateway may be considered an edge between an in-home electronic device and a cloud computing center. The basic principle of edge computing is to migrate the computing task to the edge device that produces the source data.
With the development of edge computing in the application of the internet of things, more and more terminal devices are added into an edge computing mode. Most of traditional cloud computing adopts a centralized management method, and transmission of mass data from the edge of a network to a cloud computing center causes long network delay and does not meet application programs with real-time requirements. Meanwhile, if the intermediate network is attacked when the data is transmitted to the cloud computing center, the problems of security and privacy disclosure of network edge data can be caused. Finally, with more and more user applications running in the cloud computing center, the requirements of the large-scale data center on energy consumption will be difficult to meet in the future.
In summary, in order to meet the service quality requirement of the user terminal, the edge computing platform is used to push the cloud service to the network edge, when the user terminal initiates a request, the task is decomposed into a group of parallel tasks, and the parallel tasks are processed by the edge server, so that the requirements of real-time performance, safety, low energy consumption and the like of data processing are ensured. In order to realize the edge computing processing of the user terminal, the tasks are described by using a DAG task graph, and under the condition that the task priority and the constraint condition are met, the parallel tasks are distributed to different edge servers for execution by a branch-and-bound method, so that the total energy consumption of task execution is reduced.
Disclosure of Invention
The invention provides an edge computing task allocation method based on a branch-and-bound method, which is mainly applied to the aspect of edge computing and has the main advantages of optimizing the total energy consumption of task execution by considering the distributed processing of parallel tasks.
1. An edge calculation task distribution method based on a branch-and-bound method at least comprises the following steps:
step one, arranging an edge calculation scene, wherein the scene is composed of a network model formed by a user terminal and a plurality of edge servers;
step two, describing the tasks initiated by the user terminal by using a DAG task graph G ═ T, P, wherein the vertex of the graph is described by using a set T ═ T1,t2,...,tmRepresents the tasks needed to be executed by the edge server, the number of the tasks in the task graph is m, and the tasks tiByDefinition of wherein pi、eiAndrespectively representing the number of CPU cycles required for executing the task, the transmission power of the task and the maximum tolerance time for completing the task; p ═ Pij|ti,tjE T represents a set of communication edges between tasks, PijRepresenting slave tasks tiTo task tjThe task has priority, and the subsequent task must start processing after all the predecessor tasks are finished;
step three, a network model formed by connecting a plurality of edge servers is described by a mesh network N ═ (A, D), wherein a vertex set A ═ a of the network model1,a2,...,anIndicates the edge server, the number of edge servers in the network model is n, the edge server ajFrom a to aj(vj,ej,ej,init,hj) Definition of wherein vj、ej、ej,initAnd hjRespectively representing the processing rate of the edge server, the energy consumption per unit time, the initial capacity and the constant failure rate per unit time, and the edge set D ═ DijMeans forCommunication distances between different edge servers;
step four, distributing the tasks in the DAG task graph in the step two to the network model formed by the interconnection of the edge servers in the step three, wherein the task t isiCompletion time of
Figure BDA0002257519380000023
Wherein lijRepresenting slave edge server ajTransmitted for performing task tiB represents the network bandwidth; task tiAt edge server ajTotal energy consumption of upper execution
Figure BDA0002257519380000024
Reliability of DAG task graph
Figure BDA0002257519380000025
In order to minimize the energy consumption of the edge server for executing all tasks and meet the constraint condition, a task allocation matrix X is obtained by a branch-and-bound methodm×nEach element x in the matrixijRepresenting a task tiWhether or not at edge server ajIs executed if task tiAt ajUpper execution, then xijIs 1, otherwise is 0.
2. The method of claim 1, wherein the task is executed in a sequence determined by traversing the set of tasks in the DAG task graph, wherein if a task is a predecessor of another task, the task is processed preferentially, and if multiple tasks have the same execution sequence, the tasks are processed randomly.
3. The method for task allocation based on branch-and-bound method in edge computing as claimed in claim 1, wherein the task allocation matrix X is obtained by branch-and-bound method according to the objectives and constraints mentioned in step fourm×nAt least comprises the following steps:
1) establishing an optimization problem model P1：
Figure BDA0002257519380000026
Constraint conditions are as follows:
Figure BDA0002257519380000031
R(G)≥Rreq(G) wherein
Figure BDA0002257519380000032
Meaning that a task can only be executed on one edge server, Rreq(G) Representing the reliability requirements of the task graph G;
2) initializing parameters, randomly initializing task allocation matrix Xm×nSolution set Solution fetch
Figure BDA0002257519380000033
Taking the optimal value optVal to be + ∞;
3) relaxation Xm×nBinary decision variable x in (1)ijE {0,1} is a real variable xij∈[0,1]Obtaining a relaxation optimization problem model P2；
4) Solving by adopting an interior point method to obtain temporary solution X'm×nAnd the current energy consumption value tempVal;
5) judging whether the tempVal is larger than the optVal or not and whether the current problem has a feasible solution or not, if the tempVal is larger than the optVal or the current problem has no feasible solution, pruning the current problem, and ending, otherwise, executing the step 6);
6) judging X'm×nElement x ofijWhether the discrete value is 0 or 1 is taken, if so, the Solution set Solution is updated, otherwise, the current problem is branched and X 'is set'm×nThe initial value i of the row number is 1;
7) prepared from X'm×nElement x of the ith rowijTake discrete values of 0 or 1 and satisfy
Figure BDA0002257519380000034
That is, only one element in the ith row takes a value of 1, and the rest are all 0, at this time, the element in the ith row can be obtainedUnder the condition of n different discrete values, the elements in the remaining m-i rows still take continuous values and are solved by adopting an interior point method to respectively obtain n solutions X 'of the element in the ith row under n different discrete conditions'm×nAnd resolving the tempVal corresponding to the n-th tempVal, comparing the n-th tempVal, and taking X 'corresponding to the minimum tempVal'm×nDissolving other n-1 pieces of the mixture into X'm×nPruning and taking the tempVal as a lower bound;
8) repeating the step 7) until i is m, updating the Solution set Solution, and enabling the X 'corresponding to the minimum tempVal in the Solution'm×nIs given to Xm×nX obtained at this timem×nI.e. the matrix is assigned to the task being solved.
Compared with the prior art, the invention has the advantages that:
(1) the invention minimizes the energy consumption for completing the edge computing task under the condition of comprehensively considering time and reliability, reflects the requirements in various aspects in the process of executing the edge computing network task, and the task allocation scheme obtained by the method can improve the comprehensive performance for completing the task;
(2) the method takes 1 from each decision variable according to only one value to carry out a branching strategy on the original problem, then adopts an interior point method to solve, and prunes the solution, compared with a traversal algorithm, the computational complexity is O (n)m) Reduction to o (mn);
(3) the method is a deterministic solution and has the advantages of good convergence, stable calculation result and the like.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a diagram of an edge computation scenario;
FIG. 3 is a task allocation process diagram;
fig. 4 is a schematic branching diagram.
Detailed Description
The present invention is described in further detail below with reference to the attached drawing figures.
Task collection: an application program running on a terminal device is represented by a DAG task graph G ═ (T, P), and each member is defined as follows:
T＝{t1,t2,...,tmthe representation of the task set that needs the edge server to execute, the task number in the task graph is m, in fig. 3, there are 6 tasks; p ═ Pij|ti,tjE T represents a set of communication edges between tasks, PijRepresenting slave tasks tiTo task tjIn FIG. 3, task t1To task t2One directed edge of is P12。
Edge network: the edge network formed by the edge server is represented by a mesh network N ═ (a, D), and each member is defined as follows:
A＝{a1,a2,...,anindicates an edge server set, where the number of edge servers is n, and in fig. 3, there are 4 edge servers in total; d ═ DijDenotes the communication distance between different edge servers, in fig. 3, edge server a1To the edge server a2Has a communication distance d12。
The task allocation of the edge calculation is generally divided into three steps, namely, firstly, the execution sequence of task nodes is determined, a task allocation matrix is initialized, then, an optimization problem model is established, and finally, the tasks are allocated to proper edge servers according to the proposed branch-and-bound method.
Step one, determining an execution sequence of tasks, and generating a task allocation matrix:
1) determining the execution order of tasks from the DAG task graph as { t }1}、{t2,t3,t4}、{t5}、{t6}；
2) The rows of the task allocation matrix correspond to tasks and need to be arranged from top to bottom according to the execution sequence of the tasks; if a task precedes another task in the DAG task graph, the row corresponding to the task is above; if the tasks have the same execution sequence, randomly arranging rows corresponding to the tasks;
3) the columns of the task assignment matrix correspond to edge servers, for each row, only 1 column takes 1 and the rest all take 0.
4) Through the steps, the initial task allocation matrix is obtained as follows:
Figure BDA0002257519380000051
step two, establishing an optimization problem model:
the whole optimization problem model can be expressed as the total energy consumption for the whole task distribution is minimized on the premise of meeting the total task completion time constraint and the task reliability constraint.
And step three, distributing the tasks to proper edge equipment according to the proposed branch and bound method.
1) Relaxation task assignment matrix X6×4Binary decision variable x in (1)ijE {0,1} is a real variable xij∈[0,1]The random initialization task allocation matrix is as follows:
Figure BDA0002257519380000052
solution extraction
Figure BDA0002257519380000053
Taking the optimal value optVal to be + ∞;
2) solving by adopting an interior point method to obtain a temporary solution
Figure BDA0002257519380000054
Prepared from X'6×4Maximum value of each row
Taking 1 and all the rest as 0 to obtain an approximate solution
Figure BDA0002257519380000055
And the current energy consumption value tempVal;
3) if the tempVal is larger than the optVal or the current problem has no feasible solution, pruning the current problem, and ending, otherwise, executing the step 4);
4) judging X'6×4Element x ofijWhether the discrete values are 0 and 1, if so, updating Solution, otherwise, branching the current problem and setting X'6×4The initial value i of the row number is 1;
5) prepared from X'6×4Element x of the ith rowijTake discrete values of 0 or 1 andin this case, the i-th element can take 4 different discrete values, case 1: get xi1＝1，xi2＝0，xi3＝0，xi40; case 2: get xi1＝0，xi2＝1，xi3＝0，xi40; case 3: get xi1＝0，xi2＝0，xi3＝1，xi40; case 4: get xi1＝0，xi2＝0，xi3＝0，xi41 is ═ 1; the elements in the remaining 6-i rows are still continuous values and are solved by adopting an interior point method to obtain 4 solutions X 'of the elements in the ith row under 4 different discrete conditions'6×4And resolving the corresponding tempVal, comparing the 4 tempVals, and taking X 'corresponding to the minimum tempVal'6×4And dissolving the other 3 of the three residues by X'6×4Pruning and taking the tempVal as a lower bound;
6) repeating the step 5) until i is 6, updating the Solution set Solution, and enabling the X 'corresponding to the minimum tempVal in the Solution'6×4Is given to X6×4X obtained at this time6×4I.e. the matrix is assigned to the task being solved.

Claims (3)

1. An edge calculation task distribution method based on a branch-and-bound method at least comprises the following steps:
step one, arranging an edge calculation scene, wherein the scene is composed of a network model formed by a user terminal and a plurality of edge servers;
step two, describing the tasks initiated by the user terminal by using a DAG task graph G ═ T, P, wherein the vertex of the graph is described by using a set T ═ T1,t2,...,tmRepresents the tasks needed to be executed by the edge server, the number of the tasks in the task graph is m, and the tasks tiBy
Figure FDA0002257519370000011
Definition of wherein pi、eiAnd
Figure FDA0002257519370000012
respectively representing the number of CPU cycles required for executing the task, the transmission power of the task and the maximum tolerance time for completing the task; p ═ Pij|ti,tjE T represents a set of communication edges between tasks, PijRepresenting slave tasks tiTo task tjThe task has priority, and the subsequent task must start processing after all the predecessor tasks are finished;
step three, a network model formed by connecting a plurality of edge servers is described by a mesh network N ═ (A, D), wherein a vertex set A ═ a of the network model1,a2,...,anIndicates the edge server, the number of edge servers in the network model is n, the edge server ajFrom a to aj(vj,ej,ej,init,hj) Definition of wherein vj、ej、ej,initAnd hjRespectively representing the processing rate of the edge server, the energy consumption per unit time, the initial capacity and the constant failure rate per unit time, and the edge set D ═ DijDenotes the communication distance between different edge servers;
step four, distributing the tasks in the DAG task graph in the step two to the network model formed by the interconnection of the edge servers in the step three, wherein the task t isiCompletion time of
Figure FDA0002257519370000013
Wherein lijRepresenting slave edge server ajTransmitted for performing task tiB represents the network bandwidth; task tiAt edge server ajTotal energy consumption of upper execution
Figure FDA0002257519370000014
Reliability of DAG task graph
Figure FDA0002257519370000015
In order to minimize the energy consumption of the edge server for executing all tasks and meet the constraint condition, a task allocation matrix X is obtained by a branch-and-bound methodm×nEach element x in the matrixijRepresenting a task tiWhether or not at edge server ajIs executed if task tiAt ajUpper execution, then xijIs 1, otherwise is 0.
2. The method of claim 1, wherein the task is executed in a sequence determined by traversing the set of tasks in the DAG task graph, wherein if a task is a predecessor of another task, the task is processed preferentially, and if multiple tasks have the same execution sequence, the tasks are processed randomly.
3. The method for task allocation based on branch-and-bound method in edge computing as claimed in claim 1, wherein the task allocation matrix X is obtained by branch-and-bound method according to the objectives and constraints mentioned in step fourm×nAt least comprises the following steps:
1) establishing an optimization problem model P1：
Figure FDA0002257519370000021
Constraint conditions are as follows:
Figure FDA0002257519370000022
xij∈{0,1}，
Figure FDA0002257519370000023
R(G)≥Rreq(G) wherein
Figure FDA0002257519370000024
xijE {0,1} indicates that a task can only be executed on one edge server, Rreq(G) Representing the reliability requirements of the task graph G;
2) initializing parameters, randomly initializing task allocation matrix Xm×nSolution set Solution fetch
Figure FDA0002257519370000025
Taking the optimal value optVal to be + ∞;
3) relaxation Xm×nBinary decision variable x in (1)ijE {0,1} is a real variable xij∈[0,1]Obtaining a relaxation optimization problem model P2；
4) Solving by adopting an interior point method to obtain temporary solution X'm×nAnd the current energy consumption value tempVal;
5) judging whether the tempVal is larger than the optVal or not and whether the current problem has a feasible solution or not, if the tempVal is larger than the optVal or the current problem has no feasible solution, pruning the current problem, and ending, otherwise, executing the step 6);
6) judging X'm×nElement x ofijWhether the discrete value is 0 or 1 is taken, if so, the Solution set Solution is updated, otherwise, the current problem is branched and X 'is set'm×nThe initial value i of the row number is 1;
7) prepared from X'm×nElement x of the ith rowijTake discrete values of 0 or 1 and satisfy
Figure FDA0002257519370000026
xijE {0,1}, j e {1, 2., n }, i.e. only one element in the ith row takes a value of 1, and the rest are 0, at this time, the element in the ith row can obtain n different discrete value conditions, the elements in the remaining m-i rows still take continuous values and are solved by adopting an interior point method, and n solutions X 'of the element in the ith row under n different discrete conditions are obtained respectively'm×nAnd resolving the tempVal corresponding to the n-th tempVal, comparing the n-th tempVal, and taking X 'corresponding to the minimum tempVal'm×nPruning other n-1 solutions X' and taking the tempVal as a lower bound;
8) repeating the step 7) until i is m, updating the Solution set Solution, and enabling the X 'corresponding to the minimum tempVal in the Solution'm×nIs given to Xm×nX obtained at this timem×nI.e. the matrix is assigned to the task being solved.
CN201911063618.1A 2019-11-01 2019-11-01 Edge calculation task allocation method based on branch-and-bound method Active CN110830294B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911063618.1A CN110830294B (en) 2019-11-01 2019-11-01 Edge calculation task allocation method based on branch-and-bound method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911063618.1A CN110830294B (en) 2019-11-01 2019-11-01 Edge calculation task allocation method based on branch-and-bound method

Publications (2)

Publication Number Publication Date
CN110830294A true CN110830294A (en) 2020-02-21
CN110830294B CN110830294B (en) 2022-05-10

Family

ID=69552422

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911063618.1A Active CN110830294B (en) 2019-11-01 2019-11-01 Edge calculation task allocation method based on branch-and-bound method

Country Status (1)

Country Link
CN (1) CN110830294B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111817903A (en) * 2020-09-02 2020-10-23 湖南双菱电子科技有限公司 Link fault analysis and alarm method for digital signal transmission processing equipment
CN113254178A (en) * 2021-06-01 2021-08-13 苏州浪潮智能科技有限公司 Task scheduling method and device, electronic equipment and readable storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060112049A1 (en) * 2004-09-29 2006-05-25 Sanjay Mehrotra Generalized branching methods for mixed integer programming
CN109240818A (en) * 2018-09-04 2019-01-18 中南大学 Task discharging method based on user experience in a kind of edge calculations network
CN109814951A (en) * 2019-01-22 2019-05-28 南京邮电大学 The combined optimization method of task unloading and resource allocation in mobile edge calculations network
CN110096318A (en) * 2019-05-08 2019-08-06 北京邮电大学 A kind of task discharging method and device based on mobile edge calculations

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060112049A1 (en) * 2004-09-29 2006-05-25 Sanjay Mehrotra Generalized branching methods for mixed integer programming
CN109240818A (en) * 2018-09-04 2019-01-18 中南大学 Task discharging method based on user experience in a kind of edge calculations network
CN109814951A (en) * 2019-01-22 2019-05-28 南京邮电大学 The combined optimization method of task unloading and resource allocation in mobile edge calculations network
CN110096318A (en) * 2019-05-08 2019-08-06 北京邮电大学 A kind of task discharging method and device based on mobile edge calculations

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
THAI T. VU ET AL.: "Offloading Energy Efficiency with Delay Constraint for Cooperative Mobile Edge Computing Networks", 《2018 IEEE GLOBAL COMMUNICATIONS CONFERENCE (GLOBECOM)》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111817903A (en) * 2020-09-02 2020-10-23 湖南双菱电子科技有限公司 Link fault analysis and alarm method for digital signal transmission processing equipment
CN111817903B (en) * 2020-09-02 2020-12-01 湖南双菱电子科技有限公司 Link fault analysis and alarm method for digital signal transmission processing equipment
CN113254178A (en) * 2021-06-01 2021-08-13 苏州浪潮智能科技有限公司 Task scheduling method and device, electronic equipment and readable storage medium
CN113254178B (en) * 2021-06-01 2021-10-29 苏州浪潮智能科技有限公司 Task scheduling method and device, electronic equipment and readable storage medium
US11934871B1 (en) 2021-06-01 2024-03-19 Inspur Suzhou Intelligent Technology Co., Ltd. Task scheduling method and apparatus, electronic device, and readable storage medium

Also Published As

Publication number Publication date
CN110830294B (en) 2022-05-10

Similar Documents

Publication Publication Date Title
CN107888669B (en) Deep learning neural network-based large-scale resource scheduling system and method
Zhao et al. Deepthings: Distributed adaptive deep learning inference on resource-constrained iot edge clusters
CN110533183B (en) Task placement method for heterogeneous network perception in pipeline distributed deep learning
CN114756383B (en) Distributed computing method, system, equipment and storage medium
WO2021088207A1 (en) Mixed deployment-based job scheduling method and apparatus for cloud computing cluster, server and storage device
CN105046327B (en) A kind of intelligent grid information system and method based on machine learning techniques
CN110830294B (en) Edge calculation task allocation method based on branch-and-bound method
CN111106999A (en) IP-optical network communication service joint distribution method and device
CN113472597B (en) Distributed convolutional neural network fine-grained parameter transmission scheduling method and device
CN105094970B (en) The method of more times scheduling models of task can be divided under a kind of solution distributed system
CN108270805A (en) For the resource allocation methods and device of data processing
CN115934333A (en) Historical data perception-based cloud computing resource scheduling method and system
Zhang et al. Dynamic DNN decomposition for lossless synergistic inference
AlOrbani et al. Load balancing and resource allocation in smart cities using reinforcement learning
Zhou et al. Big data and knowledge graph based fault diagnosis for electric power systems
Ghasemi et al. Energy-efficient mapping for a network of dnn models at the edge
CN104933110B (en) A kind of data prefetching method based on MapReduce
CN116367190A (en) Digital twin function virtualization method for 6G mobile network
Tuli et al. Optimizing the Performance of Fog Computing Environments Using AI and Co-Simulation
CN115345306A (en) Deep neural network scheduling method and scheduler
CN113157344B (en) DRL-based energy consumption perception task unloading method in mobile edge computing environment
CN104507150A (en) Method for clustering virtual resources in baseband pooling
Cao et al. Online cost-rejection rate scheduling for resource requests in hybrid clouds
Ma et al. Task scheduling considering multiple constraints in mobile edge computing
CN116089021B (en) Deep learning-oriented large-scale load mixed part scheduling method, device and medium

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