CN109495541B - Cross-data-center-based cloud service workflow scheduling method - Google Patents

Cross-data-center-based cloud service workflow scheduling method Download PDF

Info

Publication number
CN109495541B
CN109495541B CN201811197185.4A CN201811197185A CN109495541B CN 109495541 B CN109495541 B CN 109495541B CN 201811197185 A CN201811197185 A CN 201811197185A CN 109495541 B CN109495541 B CN 109495541B
Authority
CN
China
Prior art keywords
node
nodes
path
deployment scheme
task
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
CN201811197185.4A
Other languages
Chinese (zh)
Other versions
CN109495541A (en
Inventor
曹健
姚艳
钱诗友
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.)
Jiangyin Zhuri Information Technology Co ltd
Original Assignee
Shanghai Jiaotong 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 Shanghai Jiaotong University filed Critical Shanghai Jiaotong University
Priority to CN201811197185.4A priority Critical patent/CN109495541B/en
Publication of CN109495541A publication Critical patent/CN109495541A/en
Application granted granted Critical
Publication of CN109495541B publication Critical patent/CN109495541B/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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a cross-data-center-based cloud service workflow scheduling method, which comprises the following steps: step 1, searching a candidate path; step 2, acquiring a pre-deployment scheme; step 3, determining a conflict node deployment scheme; and 4, step 4: and determining a final deployment scheme by combining the pre-deployment scheme and the conflict node deployment scheme. The scheduling method takes six basic structures of the workflow into consideration according to the characteristics of the cloud service workflow, carries out heuristic scheduling by excavating the structural characteristics of the workflow, and has lower time complexity and shorter workflow execution time.

Description

Cross-data-center-based cloud service workflow scheduling method
Technical Field
The invention relates to the technical field of cloud resource scheduling, in particular to a cross-data-center-based cloud service workflow scheduling method. In particular to a cross-geographic-position-distribution-oriented cloud service workflow scheduling method.
Background
Cloud computing is a computing mode based on providing resources to users in the form of services through the Internet, following the development of technologies such as distributed computing, grid computing, and the like. Resource services provided by cloud computing can be divided into three layers, namely: infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). Numerous cloud providers such as Amazon, Google, and the like establish huge cloud data centers around the world, realize storage of massive computing tasks and massive information, provide cloud computing services for users, and have succeeded in commercial applications.
With the development of cloud computing technology, more and more applications are deployed in a cloud environment and provided as cloud services. Each cloud service realizes certain functions, and more complex tasks can be realized by coordinating a plurality of cloud services, so that the cloud service workflow is generated at the same time. The cloud service workflow means that each task node in the process is completed by calling cloud service, namely the cloud service workflow realizes complex things by cooperating with the cloud service.
Cloud services are generally globally distributed, and therefore, distributed execution is required for cloud service workflows. Different areas call the same cloud service, and response time of the cloud service is different, and generally, the closer the network distance is, the smaller the delay is, and the shorter the call time is. However, because the workflow task nodes also have a dependency relationship, the nearby deployment may cause adjacent workflow task nodes to be allocated to different data centers, thereby causing an increase in data transmission time therebetween. Therefore, how to execute the cloud service workflow in the cross-regional data center is an optimization problem to be solved urgently, so that the execution time of the process is as short as possible.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a cross-data-center-based cloud service workflow scheduling method.
The invention provides a cross-data-center-based cloud service workflow scheduling method, which comprises the following steps:
step 1, candidate path searching step: acquiring a candidate path set CandPaths capable of covering all task nodes;
step 2, acquiring a pre-deployment scheme: for each path in the candidate path set CandPaths, acquiring a deployment scheme of each node in the candidate path, and marking as pathBind to obtain a pre-deployment scheme;
step 3, determining a conflict node deployment scheme: traversing all task nodes, searching for conflict nodes, and obtaining a conflict node deployment scheme; the conflict nodes refer to branch nodes, and data centers distributed by the nodes are inconsistent under pre-deployment schemes of different paths;
and 4, step 4: and determining a final deployment scheme by combining the pre-deployment scheme and the conflict node deployment scheme.
Preferably, in the step 1:
in a graph model of a workflow, an adjacency list is used for storing the connectivity of graph nodes, an array T [1.. n ] stores task nodes in the workflow, each task node T [ i ] has two attributes children and parent, children records a child node set of the task node, and parent records a parent node set of the node;
the idea of path search is that firstly, a predicted critical path ExpPath from each node to a termination node and a predicted execution time ExpTime under the path are found, the sum of the average data transmission time of each node and child nodes thereof, the average task execution time of the node under different data centers and the ExpTime of the child nodes is taken as a weight, the ExpPath of the child node with the largest weight is taken as a subsequent path from the current node to the termination node, and the sum of the weight and the ExpTime of the child nodes is taken as the ExpTime of the current node; and obtaining a predicted critical path from the starting node to the terminating node, adding the path into CandPaths, taking the parent node with the maximum ExpTime in the node as an extension node of the ExpPath of the node for the node which is not traversed by the path in CandPaths, extending the node to the starting node, and adding the obtained new path into CandPaths.
Preferably, in the step 3:
marking conflicting nodes as TkThe candidate path is denoted as piCandidate path piIn the deployment scenario of (2) node TkDeployed in data centre dcpiIn the above, by the execution time of all candidate paths containing the node, dc per deployment scenariopiTo perform grouping; for the same data centre dcpiTaking the average value of the corresponding path execution time as the cost weight of the deployment scheme, and taking the dc corresponding to the deployment scheme with the maximum weightpiAs a node TkA deployed data center; when the weight is determined, different flow structures and different weight distribution schemes are adopted.
Preferably, the flow structure of the conflict node is an Add-Split, Add-Join, Or-Split Or-Join structure;
and ad-Split: the different branches under the Add-Split structure are in a parallel execution relationship, the different branches can be continuously executed downwards, the weight ratio of the paths under the different branches corresponding to the child nodes of the conflict node to the deployment schemes to be selected of the nodes is 1, namely the influence of the paths under the different branches on the deployment schemes selected by the nodes is equal;
Or-Split: the different branches under the Or-Split structure are in a selective execution relation, the Or-Split structure with a judgment condition can determine which branch to execute continuously according to the state of the current flow, and the weight occupation ratios of the different branches corresponding to the child nodes of the conflict node to the deployment schemes to be selected of the nodes are different; counting the execution probability of each branch according to the historical record, multiplying the execution time of paths under different branches by the probability of child nodes contained in the branch, namely the weighted execution time, for the deployment scheme of the current node, and using the execution time for calculating the cost weight of the deployment scheme;
and d-Join: for the ad-Join task node, the task node can start to execute only after all father nodes finish executing, the influence of paths under different branches corresponding to the father nodes of the task node on the cost weight of the node deployment scheme is equal, And weighting calculation is not needed;
Or-Join: for the Or-Join task node, as long as one father node completes execution, the task node starts execution, and assuming that the execution completion time of different father nodes is uniformly distributed, the influence of paths under different branches corresponding to the father node of the task node on the cost weight of the deployment scheme of the node is considered to be equal.
Preferably, in the step 4:
and combining the pre-deployment scheme and the conflict node deployment scheme, adjusting the rest nodes, and determining a final deployment scheme.
Compared with the prior art, the invention has the following beneficial effects:
the scheduling method takes six basic structures of the workflow into consideration according to the characteristics of the cloud service workflow, carries out heuristic scheduling by excavating the structural characteristics of the workflow, and has lower time complexity and shorter workflow execution time.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
fig. 1 is a flowchart of a cloud service workflow scheduling method according to the present invention.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.
The invention mainly aims at the scheduling of the cloud service workflow, and the task nodes in the cloud service workflow have the characteristic of global distribution, so that distributed execution is necessary. Therefore, the invention provides a heuristic scheduling method based on a dynamic critical path of pseudo polynomial time.
Firstly, the invention provides six basic structures of a workflow diagram: sequence (sequential structure), And-Join (And branching), And-Join (And joining), Or-Join (Or branching), Or-Join (Or joining), And Iteration (round-robin structure).
In the Sequence structure, all nodes are in a serial execution relationship, that is, a node can start executing only after its predecessor node is completed. In the Add structure such as Add-Split And Add-Join, all branches are in parallel relation And need to be executed simultaneously, while in the Or structure such as Or-Split And Or-Join, branch nodes are in conditional relation, And only branches with true conditions can be executed each time according to condition judgment. The Iteration structure indicates that a certain task node is executed iteratively one or more times.
A standard workflow diagram can be regarded as a combination of these six basic structures. Where the And-Join And the nd-Split occur in pairs, i.e., one And-Split necessarily corresponds to one And-Join. The same is true for Or-Split and Or-Join. For the Iteration structure, because the execution time of the Iteration structure is influenced by the Iteration number, the product of the expected value of the Iteration number and the expected execution time of a single task in the circular flow can be used as the expected execution time of the circular flow, so that the Iteration structure is regarded as a special task node, and therefore the Iteration structure is not considered in the subsequent workflow model definition and workflow scheduling problem.
Cloud Data Center (Data Center) dckDenotes data centers located in the k-th area, each data centerThere is a workflow system engine that instantiates a task node in the workflow model assigned to it and forwards received requests from other task nodes to associated nodes that send output data to its children after performing the corresponding task, which children may be on the same data center or in data centers in other areas. We also define the set of data centers as DC, with m representing the total number of data centers, and thus DCk∈DC(1≤k≤m)。
Task execution time CompTime (t)i) Denotes the ith task node tiThe average execution time of the corresponding task is recorded as the time of one task execution from the time when the task node receives the output data of all the father nodes of the task node to the time when the task execution is finished and the output data is generated. Since the task of the node is usually a web service, and there is a time difference between different data centers calling the same web service, the task execution time is also related to the data center where the task node is deployed.
Data transmission time CommTime (e)i,j) Denotes the ith task node tiThe output data transmitted after the end of execution is taken as tiJ-th task node t of child nodejThe total receiving time is proportional to the data size of the output data and inversely proportional to the network bandwidth among the task nodes. Meanwhile, we assume that when two task nodes are deployed to the same data center, the data transmission time is zero.
Completion time (Makespan) M, representing the total time for workflow execution to complete. Completion time of workflow equals termination node tendTime for completion of task execution minus start node tstartThe time to start executing the task.
Path, representing the starting node t of the slave workflowstartTo the terminating node tendIs selected. The traversed node of the path is t ∈ Tpath, and the traversed directed edge is e ∈ Epath, wherein Tpath represents the set of the traversed node of the path, and Epath represents the set of the traversed directed edge of the path. Road surfaceThe completion time costtime (path) of the path is the sum of the task execution time of the node t traversed by the path and the data transmission time of the directed edge e, and thus we can define the execution completion time of the path P as:
CostTime(path)＝sum(CompTime(ti))+sum(CommTime(ei,j))
where sum denotes the sum, ei,jRepresenting a task node tiTo tjHas a directed edge.
Critical Path (CP) representing the starting node t from the workflowstartTo the terminating node tendThe total time of the path is the sum of the task execution time of all nodes on the path and the data transmission time of all edges, and the critical path determines the length of the completion time in a workflow model.
According to the definition of the completion time (Makespan) M, the completion time M is equal to the execution completion time of the critical path CP, and is calculated as follows:
M＝CostTime(CP)
next, the present invention will be described in more detail.
Step 1: searching candidate paths: the shortest execution time of the workflow is determined by the execution completion time of the critical path, and for the workflow with the branch nodes, the critical path can be calculated only after the data centers distributed by all the nodes are determined. Therefore, before performing scheduling optimization on the workflow, we need to obtain a candidate path set CandPaths capable of covering all task nodes.
In the graph model of the workflow, an adjacency table is used for storing the connectivity of graph nodes, an array T [1.. n ] stores task nodes in the workflow, each task node T [ i ] has two attributes child and parent, child records are child node sets of the task node, and parent records are parent node sets of the node. The idea of path search is that firstly, a predicted critical path ExpPath from each node to a termination node and an expected execution time ExpIme under the path are found, the sum of the average data transmission time of each node and its child nodes, the average task execution time of the node under different data centers and the ExpIme of the child nodes is taken as a weight, the ExpPath of the child node with the largest weight is taken as a subsequent path from the current node to the termination node, and the sum of the weight and the ExpIme of the child nodes is taken as the ExpIme of the current node. After preprocessing, a predicted critical path from a starting node to a terminating node can be obtained, the path is added into CandPaths, meanwhile, for nodes which are not traversed by the path in CandPaths, a parent node with the maximum ExpTime in the nodes is used as an extension node of the ExpPath, the parent node is extended upwards to the starting node according to the rule, and the obtained new path is added into CandPaths.
The time complexity of the algorithm is related to the number of nodes and the number of edges of the graph, wherein n is the number of workflow task nodes, and m represents the number of directed edges. The time complexity of the preprocessing is O (mn), the path generation algorithm needs to trace back to the starting node, and the worst time complexity is O (mn), so the total time complexity of the algorithm is O (mn).
Step 2: acquiring a pre-deployment scheme: for each path in the candidate path set, an optimal deployment scheme of each node in the candidate path is obtained through an optimal solution algorithm (such as dynamic planning, branch definition, genetic algorithm, and the like), and is marked as pathBind.
And step 3: determining a conflict node deployment scheme: and traversing all task nodes, and searching for a conflict node (namely a branch node, wherein the data centers distributed by the node are inconsistent under the pre-deployment scheme of different paths). We label the conflicting node as TkThe candidate path is denoted as piCandidate path piIn the deployment scenario of (2) node TkDeployed in data centre dcpiIn the above, we follow a deployment scenario dc by the execution time of all candidate paths containing the nodepiTo perform the grouping. For the same data centre dcpiTaking the average value of the corresponding path execution time as the cost weight (cost weight) of the scheme, and taking the dc corresponding to the scheme with the maximum weight as the cost weight (cost weight)piAs a node TkA deployed data center. When the weight is determined, different flow structures and different weight distribution schemes are adopted.And the possible flow structures of the conflict nodes are And-Split, And-Join, Or-Split And Or-Join.
(1) And ad-Split: different branches under the Add-Split structure are in a parallel execution relationship, And different branches can continue to execute downwards, so when a conflict node is Add-Split, the weight ratio of paths under different branches corresponding to child nodes to a candidate scheme of the node is 1, that is, the influence of paths under different branches on the selection scheme of the node is equal.
(2) Or-Split: the different branches under the Or-Split structure are in a selective execution relationship, and the Or-Split structure with a judgment condition can determine which branch to continue to execute downwards according to the state of the current flow, so that when the conflict node is Or-Split, the weight ratio of the different branches corresponding to the child nodes to the candidate schemes of the nodes should be different. The execution probability of each branch can be counted according to the history, and for the deployment scheme of the current node, it is expected that the influence weight of the path corresponding to the higher branch execution probability on the final selection scheme is also larger, so for the paths under different branches, the execution time of the paths under different branches is multiplied by the probability of the child nodes included in the paths, namely the weighted execution time, and the time is used for calculating the cost weight of the deployment scheme.
(3) And d-Join: for the ad-Join task node, the task node needs to wait until all the father nodes of the ad-Join task node finish executing, the task node can start executing, the influence of paths under different branches corresponding to the father nodes of the task node on the cost weight of the node deployment scheme is equal, And weighting calculation is not needed at this time.
(4) Or-Join: for an Or-Join task node, as long as one parent node finishes execution, the task node starts execution, however, the execution completion time of the parent node cannot be effectively predicted, and the influence of the workflow structure is large, so that the execution completion time of different parent nodes is assumed to be uniformly distributed, and the influence of paths under different branches corresponding to the parent node of the task node on the cost weight of the node deployment scheme is still considered to be equal in the algorithm.
And 4, step 4: and determining a final deployment scheme by combining the pre-allocation scheme and the conflict node deployment scheme.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.

Claims (4)

1. A cross-data center based cloud service workflow scheduling method is characterized by comprising the following steps:
step 1, candidate path searching step: acquiring a candidate path set CandPaths capable of covering all task nodes;
step 2, acquiring a pre-deployment scheme: for each path in the candidate path set CandPaths, acquiring a deployment scheme of each node in the candidate path, and marking as pathBind to obtain a pre-deployment scheme;
step 3, determining a conflict node deployment scheme: traversing all task nodes, searching for conflict nodes, and obtaining a conflict node deployment scheme; the conflict nodes refer to branch nodes, and data centers distributed by the nodes are inconsistent under pre-deployment schemes of different paths;
and 4, step 4: determining a final deployment scheme by combining the pre-deployment scheme and the conflict node deployment scheme;
in the step 1:
in a graph model of a workflow, an adjacency list is used for storing the connectivity of graph nodes, an array T [1.. n ] stores task nodes in the workflow, each task node T [ i ] has two attributes children and parent, children records a child node set of the task node, and parent records a parent node set of the node;
the idea of path search is that firstly, a predicted critical path ExpPath from each node to a termination node and a predicted execution time ExpTime under the path are found, the sum of the average data transmission time of each node and child nodes thereof, the average task execution time of the node under different data centers and the ExpTime of the child nodes is taken as a weight, the ExpPath of the child node with the largest weight is taken as a subsequent path from the current node to the termination node, and the sum of the weight and the ExpTime of the child nodes is taken as the ExpTime of the current node; and obtaining a predicted critical path from the starting node to the terminating node, adding the path into CandPaths, taking the parent node with the maximum ExpTime in the node as an extension node of the ExpPath of the node for the node which is not traversed by the path in CandPaths, extending the node to the starting node, and adding the obtained new path into CandPaths.
2. The cross-data center based cloud service workflow scheduling method according to claim 1, wherein in the step 3:
marking conflicting nodes as TkThe candidate path is denoted as piCandidate path piIn the deployment scenario of (2) node TkDeployed in data centre dcpiIn the above, by the execution time of all candidate paths containing the node, dc per deployment scenariopiTo perform grouping; for the same data centre dcpiTaking the average value of the corresponding path execution time as the cost weight of the deployment scheme, and taking the dc corresponding to the deployment scheme with the maximum weightpiAs a node TkA deployed data center; when the weight is determined, different flow structures and different weight distribution schemes are adopted.
3. The cross-data center based cloud service workflow scheduling method of claim 2, wherein the flow structure of the conflict node is an nd-Split, an nd-Join, Or-Split, Or-Join structure;
and ad-Split: the different branches under the Add-Split structure are in a parallel execution relationship, the different branches can be continuously executed downwards, the weight ratio of the paths under the different branches corresponding to the child nodes of the conflict node to the deployment schemes to be selected of the nodes is 1, namely the influence of the paths under the different branches on the deployment schemes selected by the nodes is equal;
Or-Split: the different branches under the Or-Split structure are in a selective execution relation, the Or-Split structure with a judgment condition can determine which branch to execute continuously according to the state of the current flow, and the weight occupation ratios of the different branches corresponding to the child nodes of the conflict node to the deployment schemes to be selected of the nodes are different; counting the execution probability of each branch according to the historical record, multiplying the execution time of paths under different branches by the probability of child nodes contained in the branch, namely the weighted execution time, for the deployment scheme of the current node, and using the execution time for calculating the cost weight of the deployment scheme;
and d-Join: for the ad-Join task node, the task node can start to execute only after all father nodes finish executing, the influence of paths under different branches corresponding to the father nodes of the task node on the cost weight of the node deployment scheme is equal, And weighting calculation is not needed;
Or-Join: for the Or-Join task node, as long as one father node completes execution, the task node starts execution, and assuming that the execution completion time of different father nodes is uniformly distributed, the influence of paths under different branches corresponding to the father node of the task node on the cost weight of the deployment scheme of the node is considered to be equal.
4. The cross-data center based cloud service workflow scheduling method according to claim 1, wherein in the step 4:
and combining the pre-deployment scheme and the conflict node deployment scheme, adjusting the rest nodes, and determining a final deployment scheme.
CN201811197185.4A 2018-10-15 2018-10-15 Cross-data-center-based cloud service workflow scheduling method Active CN109495541B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811197185.4A CN109495541B (en) 2018-10-15 2018-10-15 Cross-data-center-based cloud service workflow scheduling method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811197185.4A CN109495541B (en) 2018-10-15 2018-10-15 Cross-data-center-based cloud service workflow scheduling method

Publications (2)

Publication Number Publication Date
CN109495541A CN109495541A (en) 2019-03-19
CN109495541B true CN109495541B (en) 2021-04-02

Family

ID=65689737

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811197185.4A Active CN109495541B (en) 2018-10-15 2018-10-15 Cross-data-center-based cloud service workflow scheduling method

Country Status (1)

Country Link
CN (1) CN109495541B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104035864A (en) * 2013-12-17 2014-09-10 北京邮电大学 Path generation method based on contradiction fragment mode
CN106100719A (en) * 2016-06-06 2016-11-09 西安电子科技大学 Moonlet network efficient resource dispatching method based on earth observation task
CN107637024A (en) * 2015-07-30 2018-01-26 华为技术有限公司 The virtual network insertion that connectivity perceives

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8531969B2 (en) * 2011-01-21 2013-09-10 Ciena Corporation Path computation systems and methods for heterogeneous multi-domain networks
CN104468406B (en) * 2013-09-13 2019-10-22 中兴通讯股份有限公司 Across host node method and device for business processing
CN103686776B (en) * 2013-12-18 2017-04-19 中国人民解放军重庆通信学院 Method for controlling stability of delay-related MANETs
CN104239141B (en) * 2014-09-05 2017-07-28 北京邮电大学 Optimizing and scheduling task method based on workflow critical path in data center
US10572442B2 (en) * 2014-11-26 2020-02-25 Microsoft Technology Licensing, Llc Systems and methods for providing distributed tree traversal using hardware-based processing

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104035864A (en) * 2013-12-17 2014-09-10 北京邮电大学 Path generation method based on contradiction fragment mode
CN107637024A (en) * 2015-07-30 2018-01-26 华为技术有限公司 The virtual network insertion that connectivity perceives
CN106100719A (en) * 2016-06-06 2016-11-09 西安电子科技大学 Moonlet network efficient resource dispatching method based on earth observation task

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Traffic-Aware Virtual Machine Migration in Topology-Adaptive DCN;Yong Cui等;《IEEE/ACM TRANSACTIONS ON NETWORKING》;20171231;第25卷(第6期);全文 *

Also Published As

Publication number Publication date
CN109495541A (en) 2019-03-19

Similar Documents

Publication Publication Date Title
US10884795B2 (en) Dynamic accelerator scheduling and grouping for deep learning jobs in a computing cluster
CN110348571A (en) A kind of neural network model training method, device, chip and system
CN104636204A (en) Task scheduling method and device
WO2022139865A1 (en) User context migration based on computation graph in artificial intelligence application executing in edge computing environment
CN110689262A (en) Space-based information system task scheduling method and device and electronic equipment
CN112148492A (en) Service deployment and resource allocation method considering multi-user mobility
CN103914556A (en) Large-scale graph data processing method
CN115134371A (en) Scheduling method, system, equipment and medium containing edge network computing resources
CN111510334B (en) Particle swarm algorithm-based VNF online scheduling method
CN110958666B (en) Network slice resource mapping method based on reinforcement learning
CN111049900B (en) Internet of things flow calculation scheduling method and device and electronic equipment
CN114429195B (en) Performance optimization method and device for training mixed expert model
CN117834560A (en) Resource allocation method based on computing power network and related equipment
CN113489787B (en) Method and device for collaborative migration of mobile edge computing service and data
CN113741999B (en) Dependency-oriented task unloading method and device based on mobile edge calculation
CN115309544A (en) Cluster resource planning method, device and apparatus
Feljan et al. Task allocation optimization for multicore embedded systems
CN109495541B (en) Cross-data-center-based cloud service workflow scheduling method
Pham-Nguyen et al. Dynamic resource provisioning on fog landscapes
Kang et al. Scheduling multiple divisible loads in a multi-cloud system
CN106570572B (en) Travel time calculation method and device based on MapReduce
CN116737370A (en) Multi-resource scheduling method, system, storage medium and terminal
CN112256983B (en) Navigation information processing method and device, electronic equipment and storage medium
Zhang Multihop Transmission‐Oriented Dynamic Workflow Scheduling in Vehicular Cloud
CN111836274A (en) Service processing method and device

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
TR01 Transfer of patent right

Effective date of registration: 20220802

Address after: 214400 room 317-108, building 12, No. 2 Binjiang West Road, Jiangyin City, Wuxi City, Jiangsu Province

Patentee after: JIANGYIN ZHURI INFORMATION TECHNOLOGY Co.,Ltd.

Address before: 200240 No. 800, Dongchuan Road, Shanghai, Minhang District

Patentee before: SHANGHAI JIAO TONG University

TR01 Transfer of patent right