CN112506657B - Resource management system facing micro service based on reinforcement learning - Google Patents

Resource management system facing micro service based on reinforcement learning Download PDF

Info

Publication number
CN112506657B
CN112506657B CN202011425578.3A CN202011425578A CN112506657B CN 112506657 B CN112506657 B CN 112506657B CN 202011425578 A CN202011425578 A CN 202011425578A CN 112506657 B CN112506657 B CN 112506657B
Authority
CN
China
Prior art keywords
micro
service
resource
resource allocation
microservice
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
CN202011425578.3A
Other languages
Chinese (zh)
Other versions
CN112506657A (en
Inventor
李超
侯小凤
刘嘉成
过敏意
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.)
Shanghai Jiaotong University
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 CN202011425578.3A priority Critical patent/CN112506657B/en
Publication of CN112506657A publication Critical patent/CN112506657A/en
Application granted granted Critical
Publication of CN112506657B publication Critical patent/CN112506657B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5083—Techniques for rebalancing the load in a distributed system
    • 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
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A reinforcement learning-based microservice-oriented resource management system, comprising: the system comprises a state conversion unit, a resource allocation decision unit and a management method evaluation unit, wherein the state conversion unit acquires current running state information of the data center, executes the action of the resource allocation decision unit and converts a resource demand relation between micro services into a resource bipartite graph, the resource allocation decision unit obtains performance characteristics of the micro service resources through bipartite graph neural network modeling and deduction, a feasible resource allocation scheme is obtained through a sensor network, the management method evaluation unit evaluates the advantages and disadvantages of the resource allocation scheme by adopting a standard production total value index reflecting the benefits of the data center resources, and the optimization of the resource allocation scheme is realized by comparing GNP of systems under the resource allocation decision units with different configurations. The method and the device can ensure the time delay requirement of the micro-service application on the one hand and maximally utilize the data center resources on the other hand under the condition of resource limitation, thereby improving the execution capacity and the overall performance of the data center.

Description

Resource management system facing micro service based on reinforcement learning
Technical Field
The invention relates to a technology in the field of computer information processing, in particular to a micro-service-oriented resource management system based on reinforcement learning in a data center.
Background
Micro-services, as an emerging software architecture, emphasize splitting a traditional large-scale single application service into a number of small services with irregular characteristics and dynamic changes. Although microservice architectures have been commonly used, resource management systems in microservice conditions face two challenges: firstly, for a plurality of micro services, the irregular characteristics of the micro services can be identified, which comprises the steps of obtaining the multidimensional resource requirements of different micro services, judging the importance of the resource allocation results of different micro services on the performance of the whole application, and the like; secondly, in a sudden and dynamically changing execution environment, a resource allocation decision can be self-optimized, that is, whenever an execution state such as a user load changes, the resource system is required to be capable of automatic decision-making without manual intervention.
Disclosure of Invention
Aiming at the defects of resource waste and application performance reduction of the existing data center under the two situations, the invention provides a reinforcement learning-based micro-service-oriented resource management system, which adopts reinforcement learning to obtain the characteristics of irregular and dynamically-changed micro-services and formulates a micro-service characteristic-aware resource allocation mechanism, so that under the condition of resource limitation, on one hand, the time delay requirement of micro-service application can be ensured, on the other hand, the data center resources can be maximally utilized, and the execution capacity and the overall performance of the data center are further improved.
The invention is realized by the following technical scheme:
the invention relates to a resource management system facing micro service based on reinforcement learning, which comprises: the system comprises a state conversion unit, a resource allocation decision unit and a management method evaluation unit, wherein: the management method comprises the steps that a state conversion unit collects current running state information of a data center, executes actions of a resource allocation decision unit and converts a resource demand relation between micro services into a resource bipartite graph, the resource allocation decision unit obtains performance characteristics of micro service resources through bipartite graph neural network modeling and deduction, a feasible resource allocation scheme is obtained through a sensor network, a management method evaluation unit evaluates the advantages and disadvantages of the resource allocation scheme by adopting a Gross Normalized Product (GNP) index reflecting benefits of the data center resources, and optimization of the resource allocation scheme is achieved by comparing GNPs of systems under resource allocation decision units with different configurations.
The performance characteristics of the micro service resources are as follows: the variation of the execution time of the microservice under different resource quantities.
The running state information includes: the method comprises the steps of data center CPU, storage, disk resource use condition and residual information, demand information of different micro services for different resources, execution time of different micro services under different resource quantities and whole application response time.
The service condition and the residual information are collected by using the existing server resource management tool, and the information of the disk is preferably checked by using a Linux tool df.
The demand information is acquired by using a process resource management tool provided by a server system, and the demand information of each micro service on different resources is acquired by preferably using a cgroups mechanism provided by Linux.
The execution time and the response time of the whole application are collected by a distributed tracking tool, preferably using Zipkin.
The resource bipartite graph is constructed according to the acquired demand information of the resources and the call relation between the micro services, wherein each node represents one micro service, the attribute of each node is an array, and the demand information of the micro service on different resources is recorded; the edges of the graph represent the calling relationships between the microservices.
The calling relationship among the micro-services refers to that: a group of nodes of the generated micro-service resource bipartite graph are API micro-services and are inlets of application requests, and the requests are distributed to a plurality of micro-services for execution; another group of nodes are function microservices, which are logical functions that perform requests, i.e., one type of request calls multiple function microservices through one API microservice.
The optimization of the resource allocation scheme refers to: and limiting the resources which can be occupied by each micro service according to the resource allocation decision obtained by comparison.
The resource allocation decision unit comprises: a bipartite neural network and a perceptron network, wherein: the input of the bipartite graph neural network is a set of all micro-service feature vectors, and the updated feature vector set derived by modeling of the associated micro-service is output to reflect the importance of the micro-service resource performance features to the performance of the whole application; the sensor network obtains a feasible resource allocation scheme Y (M) through a SoftMax function according to the updated feature vector set and the current data center operation stateL…ReLU(M1H'+b1)+bL)。
The micro-service feature vector refers to: feature vector h of microservice ii＝{s1,…,sn}, wherein: vector value sjThe required quantity of the j-th resource by the micro-service is shown, and the dimension n of the feature vector is the quantity of the resource types in the system.
The micro-service feature vector reflects the requirements of the micro-service on different resources, for example, there are 3 types of resources including CPU, memory and disk in the system, then the feature vector dimension of each micro-service is equal to 3, and the numerical value of the feature vector represents the quantity of the requirements on the three types of resources.
The set of all micro-service feature vectors is as follows: h ═ H1,…,hpP is the total amount of system microservices, i.e. the sum of the number of API microservices and the number of function microservices, hiE H is the feature vector of microservice i.
The graph neural network driven by the two-way attention mechanism comprises: anterior and posterior attention, wherein: the former attention reflects the feature of an API microservice being affected by the feature of the function microservice it calls, and the latter attention reflects the feature of a function microservice being affected by the feature of the API microservice calling it.
The modeling derivation refers to: due to the relevance of the micro-service features, the graph neural network driven by a two-component attention mechanism is adopted to calculate the updated feature vector, and the method specifically comprises the following steps:
for any API micro-service i, the set of the called function micro-services is AiAttention to the front
Figure BDA0002824687450000021
Figure BDA0002824687450000022
Wherein: j belongs to Ai，MαIs the feature vector h of the API microservice iiIs the vector join operation,. T is the vector invert operation.
Second, calculating the updated characteristic vector of API micro service i by using K-layer neural network
Figure BDA0002824687450000031
Figure BDA0002824687450000032
Wherein: sigma is a function of the non-linear transformation,
Figure BDA0002824687450000033
is a weight function of the transformation.
For any function micro-service j, the called function micro-service set is FjAttention from behind
Figure BDA0002824687450000034
Figure BDA0002824687450000035
Wherein: i belongs to Fj，MβIs the feature vector h of the function microservice jjIs the vector join operation,. T is the vector invert operation.
Fourthly, calculating the updated characteristic vector of the function microservice j by utilizing the K-layer neural network
Figure BDA0002824687450000036
Figure BDA0002824687450000037
Wherein: sigma is a function of the non-linear transformation,
Figure BDA0002824687450000038
is a weight function of the transformation.
The bipartite graph neural network outputs the updated feature vector set of all the microservices as follows: h '═ H'0,…,h'p}。
The standard production total value
Figure BDA0002824687450000039
Wherein: n is the number of data center microservice applications, NcIs NPiThe number of microservices > 1 and a represents the weight of the impact of this part of the application on the system GNP, N-NcIs NPiThe number of micro-services ≦ 1 and b represents the portion of the application pairWeights of influence of statistical GNP, NPiThe performance is normalized by the micro-service application i, and is the ratio of the actual response time to the required response time of the micro-service application i.
The GNP of the system under the resource allocation decision units with different configurations is compared as follows: and continuously adjusting configuration parameters of the bipartite graph neural network and the sensor network in the resource allocation decision unit by comparing GNPs (global navigation protocols) of the data center under different schemes, and further solving the optimal resource allocation scheme.
Technical effects
The invention integrally solves the technical problems that the existing data center is difficult to implement resource management measures of micro-service granularity due to the fact that the micro-service quantity is large and the characteristics of the micro-service are dynamically changed on fine granularity, and further causes the defects of data center resource waste and application performance reduction under the micro-service. According to the method, the resource allocation mechanism of micro-service perception is formulated by modeling and deducing the performance characteristics of irregular and dynamically-changed micro-service resources, and the standard production total value is provided as an evaluation index, so that under the condition of resource limitation, the time delay requirement of micro-service application can be ensured, the data center resources can be utilized to the maximum extent, and the execution capacity and the overall performance of the data center are improved.
Compared with the prior art, the method and the device have the advantages that the resource performance characteristics of the micro-service are obtained by constructing the resource bipartite graph of the micro-service and implementing the bipartite attention mechanism through the bipartite graph neural network. And obtaining a feasible resource allocation scheme by adopting the sensor network based on the obtained new characteristics of the micro-service. Meanwhile, the invention defines a new evaluation method and index of the resource allocation scheme, and the data center is calculated to apply the normalized performance value to guide the automatic adjustment of the parameter configuration of the bipartite graph neural network and the sensor network, thereby solving the optimal resource allocation scheme.
Drawings
FIG. 1 is a schematic diagram of the system of the present invention;
FIG. 2 is a flow chart of an embodiment;
fig. 3, 4 and 5 are schematic diagrams of embodiments.
Detailed Description
As shown in fig. 1, the present embodiment relates to a system for reinforcement learning-based micro service-oriented resource management, which includes: the system comprises a state conversion unit, a resource allocation decision unit and a management method evaluation unit.
The state conversion unit comprises: the system comprises a resource information acquisition submodule, a micro-service information acquisition submodule, a resource bipartite graph generation submodule, a performance acquisition submodule and an optimal resource allocation scheme execution submodule, wherein: the resource information acquisition submodule provides an interface between the system and a data center resource manager and acquires different resource use conditions and residual information in the data center; the micro-service information acquisition sub-module provides interfaces of the system and the micro-service application management system, and acquires a topological structure of micro-service application and information of different micro-services forming the application on different resources; the resource bipartite graph generation submodule generates a resource bipartite graph representing the micro-service application; the performance acquisition submodule acquires performance data of the micro-service and the micro-service application under different resource allocation schemes; and the optimal resource allocation scheme execution sub-module calls a bottom hardware resource management interface, and limits resources occupied by each micro-service according to the final resource allocation scheme to realize optimal resource allocation.
The resource information acquisition submodule specifically uses the existing server resource management tool to view resource information, for example, the Linux tool df can be used to view information of a disk.
The micro-service information acquisition sub-module is specifically implemented by a process resource management tool provided by the server system, for example, by using a cgroups mechanism provided by Linux.
Each node in the resource bipartite graph represents a micro service, the attribute of the node is an array, and the requirement information of the micro service on different resources is recorded; the edges of the graph represent the calling relation among the micro services, namely, a group of nodes of the generated micro service resource bipartite graph are API micro services and are inlets of application requests, and the requests are distributed to a plurality of micro services for execution; the other group of nodes is function microservices, which are logical functions that execute requests; generally, one type of request calls multiple function microservices through one API microservice.
The performance acquisition submodule acquires time delay information as a performance index through a distributed tracking tool, such as Zipkin and the like.
The resource allocation decision unit comprises: a bipartite neural network sub-module and a perceptron network sub-module, wherein: the bipartite graph neural network submodule is used for deducing the performance characteristics of the micro service resources based on the resource bipartite graph representing the micro service application; the sensor network sub-module has the function of solving by combining the performance characteristics of the micro-service resources and the current operation state of the data center to obtain a feasible resource allocation scheme.
The input of the bipartite graph neural network sub-module is a set of all micro-service feature vectors. The microservice feature vector reflects the microservice's demand for different resources. Therefore, the dimension of each micro-service feature vector is equal to the number of types of resources in the system, for example, three types of resources including a CPU, a memory and a disk are in the system, the dimension of each micro-service feature vector is equal to 3, and each dimension represents the required amount of one of the resources by the micro-service. The set of all micro-service feature vectors is H ═ H1,…,hpWhere p is the total number of system microservices, in other words, p is equal to the sum of the number of API microservices and the number of function microservices, hiE H is a demand vector representing the microservice i for n-dimensional resources.
The output of the bipartite graph neural network sub-module is the updated feature vector set. Each updated feature vector is derived from the feature modeling of its associated micro-service, reflecting the importance of the micro-service resource performance features to the performance of the entire application.
In the resource bipartite graph of the microservice, for any one API microservice, its new feature vector is determined by its own feature vector and the features of the function microservice it calls. For any function microservice, its new feature vector is determined by its own feature vector and the features of the API microservice that called it. In consideration of the relevance of the micro-service features, the embodiment adopts a graph neural network driven by a two-dimensional attention mechanism to calculate the updated feature vector.
The graph neural network driven by the two-way attention mechanism comprises front attention and rear attention, wherein: the former attention reflects the feature of an API microservice being affected by the feature of the function microservice it calls, and the latter attention reflects the feature of a function microservice being affected by the feature of the API microservice calling it.
The output of the bipartite graph neural network sub-module, namely the set of new feature vectors, is obtained by the following calculation process:
for any API micro-service i, the set of the called function micro-services is AiAttention to the front
Figure BDA0002824687450000051
Figure BDA0002824687450000052
Wherein: j belongs to Ai，MαIs the feature vector h of the API microservice iiIs the vector join operation,. T is the vector invert operation.
Second, calculating new characteristic vector of API micro service i by using K layer neural network
Figure BDA0002824687450000053
Figure BDA0002824687450000054
Wherein: sigma is a function of the non-linear transformation,
Figure BDA0002824687450000055
is a weight function of the transformation.
For any function micro-service j, the called function micro-service set is FjAttention from behind
Figure BDA0002824687450000056
Figure BDA0002824687450000057
Wherein: i belongs to Fj，MβIs the feature vector h of the function microservice jjIs the vector join operation,. T is the vector invert operation.
Fourthly, calculating new characteristic vector of function microservice j by utilizing K-layer neural network
Figure BDA0002824687450000058
Figure BDA0002824687450000059
Wherein: sigma is a function of the non-linear transformation,
Figure BDA00028246874500000510
is a weight function of the transformation.
The output of the bipartite graph neural network sub-module, namely the new feature vector set, is as follows: h '═ H'0,…,h'p}。
As shown in fig. 4, the present embodiment uses a bipartite graph neural network of one layer with one dimension of 3.
As shown in fig. 2, this embodiment adopts a system simulation form to verify the above system, and the setting of the width of the bipartite graph neural network being 32 and the depth being 1, and the width of the sensor network being 32 and the depth being 3 includes, in 6 different load situations as shown in fig. 3: (a) AlphaR represents the system of the embodiment; (b) FairSched allocates the same resource to each microservice; (c) CurSched allocates resources according to the current load state of each micro-service, such as queue length; (d) distributing resources by FullSched according to the historical load condition and the current load condition of the micro-service; (e) the PartProfile allocates resources according to the relation between the resources of the micro-service and the execution time; (f) and the FullProfile allocates resources according to the performance relation between the micro-service resources and the whole application.
As shown in FIG. 5, the AlphaR of the system can ensure that the standard production total value (GNP) of the system is about 17000 under different load conditions, which is about 0.48-8 times that of other schemes.
In conclusion, the invention can ensure 0.48-8 times of GNP, namely ensure higher resource benefit of the data center.
The foregoing embodiments may be modified in many different ways by those skilled in the art without departing from the spirit and scope of the invention, which is defined by the appended claims and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

Claims (7)

1. A reinforcement learning-based micro service-oriented resource management system, comprising: the system comprises a state conversion unit, a resource allocation decision unit and a management method evaluation unit, wherein: the management method comprises the following steps that a state conversion unit collects current running state information of a data center, executes actions of a resource allocation decision unit and converts a resource demand relation between micro services into a resource bipartite graph, the resource allocation decision unit obtains performance characteristics of micro service resources through bipartite graph neural network modeling and deduction, a feasible resource allocation scheme is obtained through a sensor network, a management method evaluation unit evaluates the advantages and disadvantages of the resource allocation scheme by adopting a standard production total value index reflecting the benefits of the data center resources, and the optimization of the resource allocation scheme is realized by comparing GNPs of systems under resource allocation decision units with different configurations;
the performance characteristics of the micro service resources are as follows: the change characteristics of the execution time of the micro-service under different resource quantities;
the running state information includes: the resource use condition and residual information of a CPU, a memory and a disk of the data center, the demand information of different micro services for different resources, the execution time of different micro services under different resource quantities and the response time of the whole application;
the service condition and the residual information are collected by utilizing the existing server resource management tool;
the demand information is collected by utilizing a process resource management tool provided by a server system;
the execution time and the whole application response time are collected through a distributed tracking tool;
the resource bipartite graph is constructed according to the acquired demand information of the resources and the call relation between the micro services, wherein each node represents one micro service, the attribute of each node is an array, and the demand information of the micro service on different resources is recorded; edges of the graph represent call relationships between microservices;
the calling relationship among the micro-services refers to that: a group of nodes of the generated micro-service resource bipartite graph are API micro-services and are inlets of application requests, and the requests are distributed to a plurality of micro-services for execution; the other group of nodes are function micro-services and are logic functions for executing requests, namely, one type of requests call a plurality of function micro-services through one API micro-service;
the optimization of the resource allocation scheme refers to: and limiting the resources which can be occupied by each micro service according to the resource allocation decision obtained by comparison.
2. The reinforcement learning-based micro-service oriented resource management system of claim 1, wherein the resource allocation decision unit comprises: a bipartite neural network and a perceptron network, wherein: the input of the bipartite graph neural network is a set of all micro-service feature vectors, and the updated feature vector set derived by modeling of the associated micro-service is output to reflect the importance of the micro-service resource performance features to the performance of the whole application; the sensor network obtains a feasible resource allocation scheme Y (M) through a SoftMax function according to the updated feature vector set and the current data center operation stateL...ReLU(M1H′+b1)+bL)；
The micro-service feature vector refers to: feature vector h of microservice ii＝{s1，...，sn}, wherein: characteristic value sjThe required quantity of the j-th resource by the micro-service is shown, and the dimension n of the feature vector is the quantity of the resource types in the system.
3. The reinforcement learning-based microservice-oriented resource management system of claim 2 wherein all microservice features are defined byThe set of vectors is: h ═ H1，...，hpP is the total amount of system microservices, i.e. the sum of the number of API microservices and the number of function microservices, hiE H is the feature vector of microservice i.
4. The reinforcement learning-based microservice-oriented resource management system of claim 1 or 2, wherein the bipartite graph neural network comprises: anterior and posterior attention, wherein: the former attention reflects the feature of an API microservice being affected by the feature of the function microservice it calls, and the latter attention reflects the feature of a function microservice being affected by the feature of the API microservice calling it.
5. The reinforcement learning-based microservice-oriented resource management system of claim 1 wherein the modeling derivation is: due to the relevance of the micro-service features, the graph neural network driven by a two-component attention mechanism is adopted to calculate the updated feature vector, and the method specifically comprises the following steps:
for any API micro-service i, the set of the called function micro-services is AiAttention to the front
Figure FDA0003506898280000021
Figure FDA0003506898280000022
Wherein: j belongs to Ai，MαIs the feature vector h of the API microservice iiIs the vector join operation.TIs a vector inversion operation;
second, calculating the updated characteristic vector of API micro service i by using K-layer neural network
Figure FDA0003506898280000023
Figure FDA0003506898280000024
Wherein: sigma being a non-linear transformationThe function of the function is that of the function,
Figure FDA0003506898280000025
is a weight function of the transformation;
for any function micro-service j, the called function micro-service set is FjAttention from behind
Figure FDA0003506898280000026
Figure FDA0003506898280000027
Wherein: i belongs to Fj，MβIs the feature vector h of the function microservice jjIs the vector join operation.TIs a vector inversion operation;
fourthly, calculating the updated characteristic vector of the function microservice j by utilizing the K-layer neural network
Figure FDA0003506898280000028
Figure FDA0003506898280000029
Wherein: sigma is a function of the non-linear transformation,
Figure FDA00035068982800000210
is a weight function of the transformation;
the bipartite graph neural network outputs the updated feature vector set of all the microservices as follows: h '═ H'0，...，h′p}。
6. The reinforcement learning-based microservice-oriented resource management system of claim 1, wherein the standard production total value
Figure FDA00035068982800000211
Wherein: n is the number of data center microservice applications, NcIs NPiThe number of microservices > 1 and a indicates thisPartial application of weights on influence of system GNP, N-NcIs NPiThe number of micro-services ≦ 1 and b represents the weight of the influence of this part of the application on the system GNP, NPiThe performance is normalized by the micro-service application i, and is the ratio of the actual response time to the required response time of the micro-service application i.
7. The reinforcement learning-based microservice-oriented resource management system of claim 1, wherein the GNPs for comparing systems with different configured resource allocation decision units are: and continuously adjusting configuration parameters of the bipartite graph neural network and the sensor network in the resource allocation decision unit by comparing GNPs (global navigation protocols) of the data center under different schemes, and further solving the optimal resource allocation scheme.
CN202011425578.3A 2020-12-09 2020-12-09 Resource management system facing micro service based on reinforcement learning Active CN112506657B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011425578.3A CN112506657B (en) 2020-12-09 2020-12-09 Resource management system facing micro service based on reinforcement learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011425578.3A CN112506657B (en) 2020-12-09 2020-12-09 Resource management system facing micro service based on reinforcement learning

Publications (2)

Publication Number Publication Date
CN112506657A CN112506657A (en) 2021-03-16
CN112506657B true CN112506657B (en) 2022-04-08

Family

ID=74971588

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011425578.3A Active CN112506657B (en) 2020-12-09 2020-12-09 Resource management system facing micro service based on reinforcement learning

Country Status (1)

Country Link
CN (1) CN112506657B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023048609A1 (en) * 2021-09-27 2023-03-30 Telefonaktiebolaget Lm Ericsson (Publ) Device and method for scaling microservices
CN114205419B (en) * 2021-12-14 2023-04-18 上海交通大学 Data center request scheduling system and method oriented to micro-service multi-dimensional disturbance characteristics
CN115037749B (en) * 2022-06-08 2023-07-28 山东省计算中心（国家超级计算济南中心） Large-scale micro-service intelligent multi-resource collaborative scheduling method and system

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109067619A (en) * 2018-09-25 2018-12-21 海南民航凯亚有限公司 A kind of elasticity volume dispatching method and processing terminal that micro services are administered
US10198250B1 (en) * 2018-01-04 2019-02-05 Accenture Global Solutions Limited Partitioning based migration of systems to container and microservice based platforms
CN109672709A (en) * 2017-10-17 2019-04-23 上海仪电（集团）有限公司中央研究院 A kind of mixed cloud service scheduling system and method
CN110362494A (en) * 2019-07-18 2019-10-22 腾讯科技（深圳）有限公司 Method, model training method and the relevant apparatus that micro services status information is shown
CN110825518A (en) * 2019-10-14 2020-02-21 上海交通大学 Nanosecond power resource distribution method and system for microservice
CN111859027A (en) * 2019-04-24 2020-10-30 华为技术有限公司 Graph calculation method and device

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109672709A (en) * 2017-10-17 2019-04-23 上海仪电（集团）有限公司中央研究院 A kind of mixed cloud service scheduling system and method
US10198250B1 (en) * 2018-01-04 2019-02-05 Accenture Global Solutions Limited Partitioning based migration of systems to container and microservice based platforms
CN109067619A (en) * 2018-09-25 2018-12-21 海南民航凯亚有限公司 A kind of elasticity volume dispatching method and processing terminal that micro services are administered
CN111859027A (en) * 2019-04-24 2020-10-30 华为技术有限公司 Graph calculation method and device
CN110362494A (en) * 2019-07-18 2019-10-22 腾讯科技（深圳）有限公司 Method, model training method and the relevant apparatus that micro services status information is shown
CN110825518A (en) * 2019-10-14 2020-02-21 上海交通大学 Nanosecond power resource distribution method and system for microservice

Also Published As

Publication number Publication date
CN112506657A (en) 2021-03-16

Similar Documents

Publication Publication Date Title
CN112506657B (en) Resource management system facing micro service based on reinforcement learning
US20220391771A1 (en) Method, apparatus, and computer device and storage medium for distributed training of machine learning model
US10558498B2 (en) Method for scheduling data flow task and apparatus
CN103631657B (en) A kind of method for scheduling task based on MapReduce
CN113037877B (en) Optimization method for time-space data and resource scheduling under cloud edge architecture
CN112118312B (en) Network burst load evacuation method facing edge server
Junaid et al. Modeling an optimized approach for load balancing in cloud
CN103701886A (en) Hierarchic scheduling method for service and resources in cloud computation environment
KR102247629B1 (en) Apparatus and method for optimizing cloud virtual machine resources based on machine-learning
CN110347504B (en) Many-core computing resource scheduling method and device
US10565085B2 (en) Two-stage distributed estimation system
US20120072456A1 (en) Adaptive resource allocation for multiple correlated sub-queries in streaming systems
WO2022252546A1 (en) Information adjusting method and device, and storage medium
CN111782359B (en) Distributed computing system task allocation method and related equipment
Aggarwal et al. Self adaptive fruit fly algorithm for multiple workflow scheduling in cloud computing environment
Zhao et al. Efficient task scheduling for Many Task Computing with resource attribute selection
TW202138999A (en) Data dividing method and processor for convolution operation
US20220300323A1 (en) Job Scheduling Method and Job Scheduling Apparatus
CN114185679A (en) Container resource scheduling method and device, computer equipment and storage medium
CN113051130A (en) Mobile cloud load prediction method and system of LSTM network combined with attention mechanism
Soula et al. Intelligent tasks allocation at the edge based on machine learning and bio-inspired algorithms
CN111414961A (en) Task parallel-based fine-grained distributed deep forest training method
CN104618480A (en) Cloud system source distributing method driven on basis of network link utilization rates
CN116915869A (en) Cloud edge cooperation-based time delay sensitive intelligent service quick response method
CN115499511B (en) Micro-service active scaling method based on space-time diagram neural network load prediction

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