CN111045820B - Container scheduling method based on time sequence prediction - Google Patents
Container scheduling method based on time sequence prediction Download PDFInfo
- Publication number
- CN111045820B CN111045820B CN201911218095.3A CN201911218095A CN111045820B CN 111045820 B CN111045820 B CN 111045820B CN 201911218095 A CN201911218095 A CN 201911218095A CN 111045820 B CN111045820 B CN 111045820B
- Authority
- CN
- China
- Prior art keywords
- application
- node
- priority
- resource
- container
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000012544 monitoring process Methods 0.000 claims abstract description 6
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 4
- 238000012423 maintenance Methods 0.000 abstract description 3
- 230000000737 periodic effect Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 230000002159 abnormal effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000013508 migration Methods 0.000 description 4
- 230000005012 migration Effects 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 3
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical group C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 241000282414 Homo sapiens Species 0.000 description 1
- 241000728173 Sarima Species 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
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]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
A container scheduling method based on timing prediction, comprising the steps of: s1: selecting periodicity according to characteristics of industry application; s2: the time sequence prediction module acquires various historical load data of an application acquired by a monitoring system from an index database, learns each historical load data in a complete period of the application by utilizing a time sequence prediction algorithm according to the historical load data of the application, and predicts a resource characteristic value of each time point in the complete period; s3: storing the resource characteristic values into a cache library for use by a dispatching expansion module; s4: a new container application is created, an optimal node is selected in a list of candidate nodes according to priority, and the new container is placed on the optimal node. The method can automatically identify the characteristics of the application, and the peak shifting arrangement of the applications in different peak periods is realized, so that peak clipping and valley filling are realized, and the resource utilization rate and the automatic operation and maintenance level of the whole cluster are improved.
Description
Technical Field
The invention belongs to the field of telecom operator service, and particularly relates to a container scheduling method based on time sequence prediction.
Background
A Business operation support system (BOSS system) of a telecom operator is a complex enterprise-level application, which is divided into a plurality of subsystems such as CRM, charging, and settlement, and includes hundreds of applications, and under a micro-service architecture, the applications are distributed and run in a container mode into a container cluster, and the container is placed on a suitable node (hereinafter, this process will also be referred to as scheduling), so that the maximum utilization rate of the whole cluster resources is the core function of the container cluster. In the micro-service architecture, the applications and containers are in a one-to-one correspondence (i.e., only one application is deployed in one container), so in the description of the present invention, the containers and applications will be used in a mixed manner, and each represents a container of an application, for example, the load characteristics of a container refer to the load characteristics of a container of an application.
In terms of scheduling with predicted resources, the prior art scheme has the following problems:
first, elastic scaling or live migration is implemented based on predictions, i.e., existing deployments are adjusted (by scaling or live migration) by predicting the amount of use of resources for a period of time in the future. Whether telescoping or migration is post-optimization, there is some cost of modification. Meanwhile, the container often carries micro-service applications, which are constantly created and destroyed (such as version upgrade), unlike virtual machine scenes, so that the optimization is more effective in the placement stage.
Secondly, the existing scheduling algorithm of the container cluster uses the reference factor of the container resource usage, but the factor has two problems, namely firstly, the factor is a single value, and the single value cannot accurately describe the resource usage, because the resource usage of the container is often periodic, and peaks and valleys exist at different moments. Secondly, the value is an manually input experience value, manual evaluation is needed, and the next system of the micro-service architecture consists of a large number of applications, and the manual mode is inaccurate and has huge workload.
Finally, the main prediction algorithm applied to the field at present is ARIMA, but the ARIMA algorithm is effective for short-term prediction, and cannot fit periodic fluctuation accurately, and the SARIMA algorithm further improved by the ARIMA algorithm can fit periodic characteristics, but has huge calculation amount, and has no practicability when the application amount of micro services is large.
Disclosure of Invention
In order to solve the technical problems, the invention provides a container scheduling method based on time sequence prediction, which can automatically identify the application characteristics, and the peak shifting and the valley filling of the applications in different peak periods are realized, so that the resource utilization rate and the automatic operation and maintenance level of the whole cluster are improved.
In order to achieve the technical purpose, the invention adopts the following technical scheme:
a container scheduling method based on timing prediction, comprising the steps of:
s1: selecting periodicity of the industrial application according to characteristics of the industrial application to determine resource usage of the application in a complete period;
s2: the time sequence prediction module acquires various historical load data of an application acquired by a monitoring system from an index database, learns each historical load data in a complete period of the application by utilizing a time sequence prediction algorithm according to the historical load data of the application, and predicts a resource characteristic value of each time point in the complete period;
s3: storing the characteristic values of the resource parts into a cache library for the dispatch expansion module to use;
s4: creating a new container application, selecting an optimal node in a candidate node list according to the priority, and placing the new container on the optimal node;
s5: steps s2-s3 are periodically performed at regular time to ensure that the characteristic values are valid in time.
Further, in step s2, during time sequence prediction, firstly, the collected historical load data is filtered by using a quartile algorithm, and then, a propset algorithm is used for fitting and predicting a resource characteristic value of a complete period.
Furthermore, the time sequence prediction module is realized by python language, periodically acquires historical load data of the application from the index database, and predicts the resource characteristic value of the application by using a prediction algorithm.
Further, in step s4, the method for determining the optimal node includes: s41: adding all the application and the resource characteristic value of the new application on one node according to the time point, taking the maximum value, calculating the resource idle rate of the node at the maximum value, normalizing the resource idle rate and taking the normalized resource idle rate as the priority of the node; s42: the final priority of the node is obtained through weighting calculation of the priority obtained in the step s41, the instance distribution balance priority, the resource idle rate priority, the resource use balance priority, the affinity priority and the local mirror image priority; s43: and calculating the final priority of all the nodes, and selecting the node with the highest priority as the placement node of the new container.
Further, in step s4, the method for obtaining the candidate node list includes: when a user builds a container application, the container management platform firstly executes a built-in screening strategy, screens out some nodes which do not meet the conditions, then calls a predicte interface in a scheduling expansion module, and the scheduling expansion module calculates whether the processing capacity of all nodes in the request parameters can meet the following calculation formula or not:
wherein C is the total amount of resources of the node,for this total number of scheduled and to-be-scheduled application containers on the node,/for this reason>For application->At->Prediction of resource usage at point +.>As a point number of the period of time,
if the value is positive, the node is put in a qualified list, namely a candidate node list, and if the value is negative, the node is put in a disqualified list; and simultaneously, caching the obtained calculated value, wherein the absolute value of the calculated value is taken as the value of the priority in the step s 4.
Further, the scheduling expansion module is written and realized by the Go language, and provides an http api interface for the container management platform to call.
The technical scheme of the invention is that the link of placing the container by selecting the computing node for the container cluster is optimized, the load of the container in future time is predicted according to the historical load data of the container by a time sequence prediction algorithm based on machine learning, the prediction result is stored as the characteristic of the container, the optimization is realized when the container is placed by the first deployment of the selecting node by utilizing the load characteristic, and the container is scheduled, for example, the application of different peak periods can be scheduled to the same computing node, so that peak clipping and valley filling are realized, and the resource utilization rate of the cluster is improved.
Compared with the prior art, the container scheduling method based on time sequence prediction has the following advantages:
firstly, the technical scheme of the invention utilizes time sequence prediction to schedule and optimize when the first container is placed, and has no change cost and risk, because even if the virtual machine supports the hot migration, the container supports the graceful start-stop, but the risk still exists when the service normally runs, the technical difficulty is higher, and therefore, the scheme has more practicability.
Secondly, compared with the existing container placement strategy, the technical scheme of the invention can automatically evaluate the resource characteristic value of the container through a time sequence prediction algorithm without manual evaluation, and under a large-scale micro-service architecture, the invention can save operation and maintenance manpower and can improve the evaluation accuracy through machine learning. And the method can automatically and periodically recalculate and evaluate according to the latest monitoring data, thereby ensuring the timeliness of updating.
And comparing with the existing scheme, the technical scheme of the invention changes the factor of the application resource demand from a single value into a value of each time point in a complete period, can realize peak shifting deployment of different peak applications, achieves the effect of automatic peak clipping and valley filling, and can improve the resource utilization rate of the whole cluster.
And thirdly, the technical scheme of the invention can realize accurate prediction of periodic data by using a quartile method and a prophet algorithm, does not bring excessive calculated amount, and has more practicability.
Finally, the technical scheme of the invention does not cover the original strategy of the system, but adds a new factor in the calculation of the priority, does not influence the existing functions of the system, is an expansion and enhancement, and has more practicability.
Drawings
FIG. 1 is a diagram showing an implementation of a container placement method according to an embodiment of the present invention;
FIG. 2 is a schematic of the daily loads for two container applications;
fig. 3 is a schematic diagram of daily load in each scenario of deployment of two applications by one node.
Detailed Description
The invention is further described below with reference to the drawings and detailed description.
A container scheduling method based on timing prediction, comprising the steps of:
s1: selecting periodicity of the industrial application according to characteristics of the industrial application to determine resource usage of the application in a complete period;
s2: the time sequence prediction module acquires various historical load data of an application acquired by a monitoring system from an index database, learns each historical load data in a complete period of the application by utilizing a time sequence prediction algorithm according to the historical load data of the application, and predicts a resource characteristic value of each time point in the complete period;
s3: storing the characteristic values of the resource parts into a cache library for the dispatch expansion module to use;
s4: creating a new container application, selecting an optimal node in a candidate node list according to the priority, and placing the new container on the optimal node;
s5: steps s2-s3 are periodically performed at regular time to ensure that the characteristic values are valid in time.
In step s2, during time sequence prediction, firstly, filtering an abnormal value of collected historical load data by using a quartile algorithm, and then, fitting and predicting a resource characteristic value of a complete period by using a prophet algorithm.
The time sequence prediction module is realized by python language, periodically acquires historical load data of the application from the index database, and predicts the resource characteristic value of the application by using a prediction algorithm.
In step s4, the method for determining the optimal node includes: s41: adding all the application and the resource characteristic value of the new application on one node according to the time point, taking the maximum value, calculating the resource idle rate of the node at the maximum value, normalizing the resource idle rate and taking the normalized resource idle rate as the priority of the node; s42: the final priority of the node is obtained through weighting calculation of the priority obtained in the step s41, the instance distribution balance priority, the resource idle rate priority, the resource use balance priority, the affinity priority and the local mirror image priority; s43: and calculating the final priority of all the nodes, and selecting the node with the highest priority as the placement node of the new container.
In step s4, the method for obtaining the candidate node list includes: when a user builds a container application, the container management platform firstly executes a built-in screening strategy, screens out some nodes which do not meet the conditions, then calls a predicte interface in a scheduling expansion module, and the scheduling expansion module calculates whether the processing capacity of all nodes in the request parameters can meet the following calculation formula or not:
wherein C is the total amount of resources of the node,for this total number of scheduled and to-be-scheduled application containers on the node,/for this reason>For application->At->Prediction of resource usage at point +.>As a point number of the period of time,
if the value is positive, the node is put in a qualified list, namely a candidate node list, and if the value is negative, the node is put in a disqualified list; and simultaneously, caching the obtained calculated value, wherein the absolute value of the calculated value is taken as the value of the priority in the step s 4.
The scheduling expansion module is written and realized by the Go language, and provides an HTTPAPI interface for the container management platform to call.
The implementation is described below in connection with specific examples:
as shown in fig. 1, the embodiment is implemented based on a Kubernetes platform, where Kubernetes is a main stream management platform in the field of container arrangement, and is implemented by expanding a scheduling algorithm native to Kubernetes.
First, the generation process of the application resource feature:
in the first step, the time sequence prediction module acquires various historical index data, such as CPU load data, of applications acquired by the monitoring system from an index database. Usually, two months of historical data are acquired, and the time is aligned forward, for example, the time is periodic, and the two months are taken forward from the last sunday;
secondly, filtering the abnormal value of the acquired data by using a quartile algorithm;
thirdly, predicting a complete period by utilizing a prophet algorithm to obtain an applied resource characteristic value;
and fourthly, storing the characteristic value into a Redis cache library for other modules to use.
The generation of application resource features may be defined as periodic execution of timed tasks to ensure that application feature values are timely and valid.
Second, the Kubernetes-based optimized container placement process:
in the first step, the Kube-Scheduler process configuration of Kubernetes is modified, a Scheduler extension configuration (Scheduler extension) is added, and the IP address and port are directed to the newly added Scheduler extension module. Preferably, various policies (such as an affinity policy, an equalization policy and the like) existing in Kubernetes can be multiplexed through a scheduling expansion mechanism, and a new scheduling policy is introduced. Preferably, the ignore parameter in the scheduler extension configuration is opened, so that when the external scheduling extension module is abnormal, the container scheduling is not abnormal, and only a mechanism inside the Kubernetes is used, so that the usability of the whole scheme is improved.
Secondly, when a user builds a container application, the Kubernetes will execute a built-in screening policy first, screen out some nodes which do not meet the conditions (such as unavailable nodes, port conflict, etc.), and then call the predicte interface expanded by the scheduler. The scheduler extension calculates whether the processing capacity of all nodes in the request parameters can be met or not, and the calculation formula is as follows:
wherein C is the total amount of resources of the node,for this total number of scheduled and to-be-scheduled application containers on the node,/for this reason>For application->At->Prediction of resource usage at point +.>The number of cycles, such as a cycle of weeks, is 168.
If this value is positive, the node is placed on the eligibility list, if the value is negative, the node is placed on the diseligibility list, and then returned to Kubernetes.
Thirdly, the Kubernetes continuously executes a built-in Priority strategy and calls a Priority interface expanded by the scheduler; the dispatcher expands the value of each node obtained in the last step, takes the absolute value and returns the absolute value as the priority; kubernetes weights all priorities obtained by the internal and external priority policies to obtain the final priority of each node.
Fourth, kubernetes selects a highest priority node, binds the container to this node, and completes the container placement process.
The peak clipping and valley filling effects realized by the technical scheme of the invention are now described with reference to fig. 2 and 3.
As shown in fig. 2, the left is the daily CPU usage (converted into the number of CPUs) of a certain voice billing application, and the right is the daily CPU usage of a certain account closing application, it is obvious that the daily voice call is peak in daytime, and there is a peak in the morning and evening, which is matched with the daily activity of human beings, and the account closing application on the right has batch tasks in the night, so there is a peak in the night.
Fig. 3 shows the superposition effect of two types of applications on a node, and it can be seen that the overall resource usage of the mixed deployment of the charging class and the reconciliation class applications is the lowest.
Assuming two nodes with the same configuration, node a runs a billing class application, node B runs a closing class application, at this time, the user adds a closing class application again, evaluates by using the peak resource usage according to the original placement algorithm, calculates the load calculation value of the two nodes as a 10 (6+4=10), and B as 8 (4+4=8), and then dispatches the load calculation value to B. But if the resource characteristic value mode is adopted, a is 7, b is 8, and the scheduling is carried out on a, which is a more optimal scheduling scheme, and the overall load is lower.
Claims (3)
1. A method of scheduling containers based on timing predictions, comprising the steps of:
s1: selecting periodicity of the industrial application according to characteristics of the industrial application to determine resource usage of the application in a complete period;
s2: the time sequence prediction module acquires various historical load data of an application acquired by a monitoring system from an index database, learns each historical load data in a complete period of the application by utilizing a time sequence prediction algorithm according to the historical load data of the application, and predicts a resource characteristic value of each time point in the complete period; the time sequence prediction module is realized by python language, periodically acquires historical load data of the application from an index database, and predicts a resource characteristic value of the application by using a prediction algorithm;
s3: storing the resource characteristic values into a cache library for use by a dispatching expansion module, wherein the dispatching expansion module is written and realized by a Go language, and provides an HTTP-API interface for a container management platform to call;
s4: creating a new container, selecting an optimal node in a candidate node list according to the priority, and placing the new container on the optimal node; the method for determining the optimal node specifically comprises the following steps: s41: adding all the application and the resource characteristic value of the new application on one node according to the time point, taking the maximum value, calculating the resource idle rate of the node at the maximum value, normalizing the resource idle rate and taking the normalized resource idle rate as the priority of the node;
s42: the final priority of the node is obtained through weighting calculation of the priority obtained in the step s41, the instance distribution balance priority, the resource idle rate priority, the resource use balance priority, the affinity priority and the local mirror image priority;
s43: calculating the final priority of all nodes, and selecting the node with the highest priority as the placement node of the new container;
s5: steps s2-s3 are periodically performed at regular time to ensure that the characteristic values are valid in time.
2. The method according to claim 1, wherein in step s2, the collected historical load data is filtered by using a quartile algorithm during time sequence prediction, and then a prophet algorithm is used to fit and predict a resource characteristic value of a complete cycle.
3. The method for scheduling containers based on time series prediction according to claim 1, wherein in step s4, the method for obtaining the candidate node list is as follows: when a user builds a container application, the container management platform firstly executes a built-in screening strategy, screens out some nodes which do not meet the conditions, then calls a predicte interface in a scheduling expansion module, and the scheduling expansion module calculates whether the processing capacity of all nodes in the request parameters can meet the following calculation formula or not:
,
wherein,c is the total amount of resources of the node, n is the total amount of scheduled and to-be-scheduled application containers on the node, P j i Predicting the resource use of the i at the j point, wherein m is the number of hours in a complete period; if the value is positive, the node is put in a qualified list, namely a candidate node list, and if the value is negative, the node is put in a disqualified list; and simultaneously, caching the obtained calculated value, wherein the absolute value of the calculated value is taken as the value of the priority in the step s 4.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911218095.3A CN111045820B (en) | 2019-12-03 | 2019-12-03 | Container scheduling method based on time sequence prediction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911218095.3A CN111045820B (en) | 2019-12-03 | 2019-12-03 | Container scheduling method based on time sequence prediction |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111045820A CN111045820A (en) | 2020-04-21 |
CN111045820B true CN111045820B (en) | 2024-02-09 |
Family
ID=70234486
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911218095.3A Active CN111045820B (en) | 2019-12-03 | 2019-12-03 | Container scheduling method based on time sequence prediction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111045820B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112087504A (en) * | 2020-08-31 | 2020-12-15 | 浪潮通用软件有限公司 | Dynamic load balancing method and device based on working load characteristics |
CN112187894B (en) * | 2020-09-17 | 2022-06-10 | 杭州谐云科技有限公司 | Container dynamic scheduling method based on load correlation prediction |
CN112350898A (en) * | 2020-11-10 | 2021-02-09 | 安徽继远检验检测技术有限公司 | Micro-service application full-link performance real-time monitoring system and detection method thereof |
CN113064696A (en) * | 2021-03-25 | 2021-07-02 | 网易(杭州)网络有限公司 | Cluster system capacity expansion method, device and medium |
US11842214B2 (en) | 2021-03-31 | 2023-12-12 | International Business Machines Corporation | Full-dimensional scheduling and scaling for microservice applications |
CN114840313B (en) * | 2022-07-04 | 2022-09-30 | 北京邮电大学 | Dispatching method and device for container assembly |
CN115543577B (en) * | 2022-08-08 | 2023-08-04 | 广东技术师范大学 | Covariate-based Kubernetes resource scheduling optimization method, storage medium and device |
CN116614517B (en) * | 2023-04-26 | 2023-09-29 | 江苏博云科技股份有限公司 | Container mirror image preheating and distributing method for edge computing scene |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108009016A (en) * | 2016-10-31 | 2018-05-08 | 华为技术有限公司 | A kind of balancing resource load control method and colony dispatching device |
US10484301B1 (en) * | 2016-09-30 | 2019-11-19 | Nutanix, Inc. | Dynamic resource distribution using periodicity-aware predictive modeling |
CN110474966A (en) * | 2019-07-22 | 2019-11-19 | 腾讯科技(深圳)有限公司 | Handle the method and relevant device of cloud platform resource fragmentation |
-
2019
- 2019-12-03 CN CN201911218095.3A patent/CN111045820B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10484301B1 (en) * | 2016-09-30 | 2019-11-19 | Nutanix, Inc. | Dynamic resource distribution using periodicity-aware predictive modeling |
CN108009016A (en) * | 2016-10-31 | 2018-05-08 | 华为技术有限公司 | A kind of balancing resource load control method and colony dispatching device |
CN110474966A (en) * | 2019-07-22 | 2019-11-19 | 腾讯科技(深圳)有限公司 | Handle the method and relevant device of cloud platform resource fragmentation |
Also Published As
Publication number | Publication date |
---|---|
CN111045820A (en) | 2020-04-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111045820B (en) | Container scheduling method based on time sequence prediction | |
CN111414233A (en) | Online model reasoning system | |
US9112782B2 (en) | Reactive auto-scaling of capacity | |
US11579933B2 (en) | Method for establishing system resource prediction and resource management model through multi-layer correlations | |
CN105487930A (en) | Task optimization scheduling method based on Hadoop | |
CN109672709B (en) | Hybrid cloud service scheduling system and method | |
CN112685153A (en) | Micro-service scheduling method and device and electronic equipment | |
CN110751358A (en) | Scheduling method for airport ground service personnel, electronic equipment and storage medium | |
CN103092683A (en) | Scheduling used for analyzing data and based on elicitation method | |
CN113806018B (en) | Kubernetes cluster resource mixed scheduling method based on neural network and distributed cache | |
CN113110914A (en) | Internet of things platform construction method based on micro-service architecture | |
CN115297112A (en) | Dynamic resource quota and scheduling component based on Kubernetes | |
CN115543577B (en) | Covariate-based Kubernetes resource scheduling optimization method, storage medium and device | |
CN106407007B (en) | Cloud resource configuration optimization method for elastic analysis process | |
CN114205317B (en) | SDN and NFV-based service function chain SFC resource allocation method and electronic equipment | |
CN111143070A (en) | Resource scheduling method and device | |
CN110096339A (en) | A kind of scalable appearance configuration recommendation system and method realized based on system load | |
ElMekkawy et al. | Real-time scheduling with deadlock avoidance in flexible manufacturing systems | |
CN109936471A (en) | A kind of resource allocation methods and device of more clusters | |
CN115562841B (en) | Cloud video service self-adaptive resource scheduling system and method | |
Milocco et al. | Evaluating the upper bound of energy cost saving by proactive data center management | |
CN114466014B (en) | Service scheduling method and device, electronic equipment and storage medium | |
CN114971428A (en) | Multi-source work order data-based engineer busy pre-estimation method and system | |
CN115658319A (en) | Resource scheduling method, system, device and storage medium | |
CN113839795B (en) | Peak eliminating processing method and device for CDN node, electronic equipment and storage 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 |