CN111327655A - Multi-tenant container resource quota prediction method and device and electronic equipment - Google Patents
Multi-tenant container resource quota prediction method and device and electronic equipment Download PDFInfo
- Publication number
- CN111327655A CN111327655A CN201811534570.3A CN201811534570A CN111327655A CN 111327655 A CN111327655 A CN 111327655A CN 201811534570 A CN201811534570 A CN 201811534570A CN 111327655 A CN111327655 A CN 111327655A
- Authority
- CN
- China
- Prior art keywords
- container
- resource usage
- time period
- resource
- predicted
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000000605 extraction Methods 0.000 claims description 10
- 238000013468 resource allocation Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000003860 storage Methods 0.000 claims description 6
- 238000013439 planning Methods 0.000 abstract description 12
- 238000004422 calculation algorithm Methods 0.000 description 23
- 238000012549 training Methods 0.000 description 18
- 238000012360 testing method Methods 0.000 description 14
- 230000008569 process Effects 0.000 description 13
- 238000012544 monitoring process Methods 0.000 description 12
- 238000012545 processing Methods 0.000 description 6
- 238000010606 normalization Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000007781 pre-processing Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 238000013500 data storage Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 101100272279 Beauveria bassiana Beas gene Proteins 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000011056 performance test Methods 0.000 description 2
- 238000001303 quality assessment method Methods 0.000 description 2
- 230000008439 repair process Effects 0.000 description 2
- 239000002904 solvent Substances 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 235000003642 hunger Nutrition 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000002688 persistence Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000037351 starvation Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
- H04L67/1097—Protocols in which an application is distributed across nodes in the network for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a device for predicting resource quota of a multi-tenant container and electronic equipment, wherein the method comprises the following steps: finding a first preset number of historical time sequences with the similarity of the first time sequence of the container to be detected and the similarity of the first time sequence of the container to be detected; determining a predicted value of the resource usage of the container to be detected in a time period to be predicted according to the resource usage of the container to be detected in a first time period and the first resource usage and the second resource usage in a first preset number of historical time sequences, wherein the first resource usage is the resource usage of the container to be detected in a specified historical time period, and the second resource usage is the resource usage of the container to be detected in a next time period of the specified historical time period. The technical scheme provided by the embodiment of the invention can automatically and accurately predict the future resource demand of the container, thereby reasonably planning the resource quota of the multi-tenant container.
Description
Technical Field
The invention relates to the technical field of container cloud operation and maintenance, in particular to a method and a device for predicting resource quota of a multi-tenant container and electronic equipment.
Background
Containerization is a new generation of cloud computing technology, and with the rapid development and wide adoption of container technology, cloud native application and container technology become popular choices in the IT field. The Kubernetes cloud platform gradually enters an enterprise production environment, and how to ensure application performance in the cloud environment and efficiently use resources as much as possible is very important for operation and maintenance personnel of the Kubernetes cloud platform and cloud application owners. The multi-tenant container resource quota plan is management that an administrator needs to set a more refined resource quota and resource limit for a container of a tenant in an environment of shared use of resources. When each tenant deploys an application on its resource partition, each tenant can only use the allocated resource amount, otherwise resource competition among containers is caused, and even a phenomenon that some container occupies a large amount of resources to cause starvation of other containers occurs. Therefore, resources are limited in the using process, so that containers are isolated from each other, and different tenants do not interfere with each other.
The container resource quota is the computing resources required to allocate the container runtime. Two types of computing resources are typically managed: a CPU (Central Processing Unit/Processor) and a memory. The configuration mode of the computing resource comprises two modes: one is resource Requests (Requests), which represent the amount of fully guaranteed resources to which a container is expected to be allocated, the value of Requests being provided to the kubernets scheduler in order to optimize the scheduling of containers based on resource Requests; the other is resource limitation, and Limits is an upper limit of available resources of the container, and the value of the upper limit influences a solution strategy when resource competition occurs on the node. Therefore, the essence of the multi-tenant container resource quota plan is to predict the amount of computing resources, i.e., Requests, Limits, required by the container runtime.
The traditional container resource quota planning method is based on a mode of combining administrator experience, performance test and monitoring, firstly, a set of offline test environment is built, then an online monitoring system is utilized to find the maximum flow in the near future and evaluate the flow in a future period of time by combining the experience, QA (QUALITY assessment) is enabled to perform performance test online, the resource usage amount of a container is found through monitoring, and the usage condition of the container resource is obtained through a gradual cycle test approach method. However, the above resource quota planning methods all adopt "experience + gradual attempt", so that the resource quota planning cannot be quantized to be "visible", the trend of the usage amount of the container resource is judged in a manual mode, and then the container resource quota is adjusted correspondingly, so that the method is complex, sometimes even wrong decisions are made, and when a new project is online or the capacity is expanded on the original basis, the container resource quota still needs to be re-evaluated. In addition, the containers are relatively loose on resource boundaries, which brings flexibility and uncertainty, and therefore, an administrator is required to set more elaborate management of resource quotas and resource limits for the containers.
Therefore, it is urgently needed to provide a scientific and effective method for supporting multi-tenant container resource quota planning.
Disclosure of Invention
The embodiment of the invention provides a method and a device for predicting resource quota of a multi-tenant container, electronic equipment and a storage medium, and aims to solve the problems that in the prior art, the method for judging the trend of the usage amount of the container resource by means of manual mode is complex in process and low in accuracy.
In a first aspect, an embodiment of the present invention provides a method for predicting resource quotas of a multi-tenant container, including:
a similar sequence extraction step, wherein a first preset number of historical time sequences with the similarity of a first time sequence of a container to be detected, which is ranked at the top, are found from a historical database, the first time sequence is a sequence in which the resource usage of the container to be detected in a second preset number of time periods before a first time period is ranked according to the time sequence, the first time period is a time period before a time period to be predicted, and the historical time sequence is a sequence in which the resource usage of the container to be detected in the second preset number of time periods before a specified historical time period is ranked according to the time sequence;
and a predicting step, namely determining a predicted value of the resource usage of the container to be predicted in a time period to be predicted according to the resource usage of the container to be detected in a first time period and the first resource usage and the second resource usage in a first preset number of historical time sequences, wherein the first resource usage is the resource usage of the container to be detected in a specified historical time period, and the second resource usage is the resource usage of the container to be detected in a next time period of the specified historical time period.
In a second aspect, an embodiment of the present invention provides a device for predicting resource quotas of multi-tenant containers, including:
the system comprises a similarity sequence extraction module, a similarity sequence extraction module and a comparison module, wherein the similarity sequence extraction module is used for finding a first preset number of historical time sequences which are ranked at the top with the similarity of a first time sequence of a container to be detected, the first time sequence is a sequence formed by arranging the resource usage of the container to be detected in a second preset number of time periods before a first time period according to the time sequence, the first time period is a time period before a time period to be predicted, and the historical time sequence is a sequence formed by arranging the resource usage of the container to be detected in the second preset number of time periods before a designated historical time period according to the time sequence;
the first prediction module is used for determining a predicted value of the resource usage of the container to be predicted in the time period to be predicted according to the resource usage of the container to be predicted in the first time period and the first resource usage and the second resource usage in the historical time sequences of the first preset number, wherein the first resource usage is the resource usage of the container to be predicted in the appointed historical time period, and the second resource usage is the resource usage of the container to be predicted in the next time period of the appointed historical time period.
In a third aspect, an embodiment of the present invention provides an electronic device, including a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver is configured to receive and transmit data under the control of the processor, and the processor implements any of the above method steps when executing the program.
In a fourth aspect, an embodiment of the invention provides a computer-readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the steps of any of the methods described above.
According to the technical scheme provided by the embodiment of the invention, the container resource quota is not simply guessed according to the service type and experience, and the change of the container resource consumption is accurately fitted by utilizing a KNN prediction model and combining the characteristics of container self-similarity, time correlation and the like, so that the future resource demand of the container can be scientifically and accurately predicted, and the multi-tenant container resource quota is reasonably planned. The method of the embodiment of the invention not only reduces the time of data analysis in the process of planning the resource quota of the multi-tenant container, but also improves the accuracy of the prediction result, thereby more effectively managing the resources of the container cloud platform.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flowchart of a resource quota predicting method for a multi-tenant container according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of a resource quota predicting method for a multi-tenant container according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a resource quota predicting apparatus for a multi-tenant container according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a resource quota predicting apparatus for a multi-tenant container according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention.
For convenience of understanding, terms referred to in the embodiments of the present invention are explained below:
the KNN (K-nearest neighbor) classification algorithm, also called K nearest neighbor algorithm, is one of the simplest methods in data mining classification technology. The core idea of the KNN algorithm is that if most of k nearest neighbor samples of a sample in the feature space belong to a certain class, the sample also belongs to the class and has the characteristics of the sample on the class. The method only determines the category of the sample to be classified according to the category of the nearest sample or samples in the determination of classification decision. The KNN algorithm can be used for classification and regression, is commonly used for prediction analysis, time series models and finding causal relationships among variables, and is widely applied to hot fields such as text classification, pattern recognition, cluster analysis, multi-classification and the like.
Prometheus is a set of open source monitoring, alarm, time series database combinations, originally developed by soundlog corporation. In a conventional kubernets container management system, promemeeus is usually used for monitoring.
Any number of elements in the drawings are by way of example and not by way of limitation, and any nomenclature is used solely for differentiation and not by way of limitation.
In a specific practical process, the existing resource quota planning method adopts a mode of 'experience + gradual trial' for prediction, so that the resource quota planning cannot be quantized to be 'visible', the trend of the usage amount of the container resources is judged in a manual mode, and then the container resource quota is correspondingly adjusted, so that the method is complex, wrong decisions are easily made due to the fact that the container resource quota is simply predicted according to the service type and the experience, and the container resource quota still needs to be re-evaluated when a new project is online or the capacity is expanded on the original basis. Therefore, the existing container resource quota planning method depends on manual experience too much, so that the process of judging the usage trend of the container resources is complicated, and the accuracy is low.
The inventor discovers from the characteristics of load and memory use that the changes of the container application load and the memory have strong correlation with time, namely the computing resources such as the container load can be regarded as a time series and have high similarity with the trend of the computing resources, so that the prediction of the container load and the memory use can be performed according to the prediction mode of the time series. Therefore, the inventor of the invention provides an initial prediction algorithm based on similarity and a KNN prediction algorithm based on time series based on the characteristics of self-similarity, time correlation, randomness, volatility and the like of container application, and intelligently calculates the calculation resources required by the container in different operation stages. Based on similarity prediction algorithm, similarity among containers is utilized, an operated container most similar to a user pre-operated container is found out by matching historical data of the operated container, a resource usage peak value of the operated container is taken as an initial resource allocation amount, resource waste is avoided, and normal use of the pre-operated container in an initial operation stage is guaranteed. After the container runs for a period of time, enough historical data of the container can be acquired, the historical resource usage amount of the container is analyzed and predicted based on a KNN prediction algorithm of a time sequence, the prediction of future container resource consumption is driven by the observation data of an actual container, and the container resource quota is not simply guessed according to the service type and experience, so that a more accurate and scientific predicted value is provided for the container resource quota.
It should be noted that the resources referred to in this embodiment are computing resources, such as CPU resources and memory resources.
Having described the general principles of the invention, various non-limiting embodiments of the invention are described in detail below.
First, the process of creating and training the prediction model will be described.
The method comprises the following steps: and (5) constructing a theoretical model.
A difficulty with multi-tenant container resource quota planning is modeling. At present, a prediction algorithm aiming at the use of container resources does not exist, and most of the traditional prediction algorithms are aimed at network traffic. From the characteristics of load and memory usage, the change of the container application load and the memory has strong correlation with time, that is, the computing resources such as the container load can be regarded as a time series and have high similarity with the trend of the computing resources, so that the prediction of the container load and the like can be performed according to the prediction mode of the time series. Therefore, an initial resource prediction algorithm based on the similarity and a KNN prediction algorithm based on the time sequence are provided based on the characteristics of self-similarity, time correlation, randomness, volatility and the like of the container application.
The container resource prediction process is divided into two stages, prediction is carried out according to an initial resource prediction algorithm based on similarity when a container is initially created, and prediction is carried out according to a KNN model based on a time sequence after the container is operated for a period of time. The data taken from the database will vary from stage to stage. Thus, prediction is divided into two phases based on historical data: an initial resource forecasting phase and a runtime resource demand forecasting phase. Therefore, the prediction models employed in the present embodiment include a similarity prediction model and a KNN prediction model.
The similarity prediction model is obtained by an initial resource prediction algorithm based on similarity, and the method usually adopted is to calculate the distance between samples. In the embodiment, a similarity prediction algorithm is adopted at the stage of predicting the initial resources of the container to be operated, a container most similar to a user pre-operation container is found out by matching with historical data of the existing containers in a historical database, and the peak value of the resource usage amount of the container is taken as the resource allocation amount of the pre-operation container during initial operation.
The KNN prediction model is determined based on a time series KNN prediction algorithm, and the running condition of the container to be tested is analyzed by analyzing the current time point and historical data before the current time point.
Step two: and (4) preprocessing data.
Data preprocessing is the most important step in machine learning: the method mainly comprises the processing procedures of data acquisition, data storage, data cleaning, data normalization and the like.
(1) And (6) data acquisition.
The important step in machine learning is sample data acquisition, and the model can be trained only after the sample data exists, so that the model is tested. The quality of the sample data determines the quality of the model, and further influences the later verification, so that the acquisition of the sample data plays a crucial role in the generation of the model. The main work of data acquisition is to collect data, and the data is from monitoring and is comprehensively considered from the aspects of instantaneity, overhead, data consistency, expandability, data characteristics and the like. The conventional cloud service system is monitored by a Prometheus monitoring system, and the Prometheus monitoring system pulls historical data of a container from a server of an operating container, so that the embodiment relies on the Prometheus monitoring system, the historical data of the container is pulled from the Prometheus monitoring system and stored in a historical database, upstream monitoring data can be more flexibly acquired, data storage is more controllable, an acquisition end cannot sense the existence of the monitoring system, and the data acquisition system is independent of the service system of the operating container, so that stable and efficient operation of cloud application and a cloud platform is ensured.
(2) And (4) storing data.
After data acquisition, in order to retain original data, the original data is transmitted to a data center for persistence without any processing, and is analyzed by a subsequent program. The collected historical data includes: container ID, CPU resource usage, memory resource usage, sampling interval, timestamp, container name, mirror image, start command, port, running status, etc. Each item of data collected must be identifiable and stored in a uniform format for subsequent processing by a program.
(3) And (6) data cleaning.
In the process of data acquisition and storage, the reliability of data cannot be guaranteed, and the information stored in the historical database necessarily contains some noises, such as data loss, abnormality and the like, so that the data at two ends of a sample are usually removed and the noises are removed. Because the embodiment adopts the time series prediction algorithm, it is necessary to ensure that the sample data has continuity, and if some data is lost, a certain measure may be taken to repair the data, for example, the data repair method may be to fill up the lost data with the same data as the previous data.
(4) And (6) normalizing the data.
Each container application service includes a plurality of attributes, for example, attributes such as a container mirror image, a runtime command, a service port, and a running state, and data dimensions of each attribute are different, so that it is necessary to perform normalization processing on an acquired attribute value of a container to prepare for comparison of similarity between subsequent containers.
And preprocessing various historical data of the operated container and storing the preprocessed historical data into a historical database. The data in the historical database can be used as sample data in model training and historical data used in prediction.
Step three: and initializing parameters.
Before the model training, some model parameters need to be initialized, and when the parameters are set, a large amount of experimental verification needs to be performed firstly, and then some proper parameters are selected for training. By analyzing the algorithm, it can be known that the time series KNN algorithm is mainly used for predicting the future resource usage amount based on the historical resource usage amount of the container, and the embodiment performs a comparison experiment by acquiring the historical data of the container to be tested from the historical database.
In a comparative experiment of the resource prediction algorithm, different service type containers were tested. Taking tomcat container as an example, the resource usage amount of the previous 150 hours can be selected for model training, wherein the resource usage amount of the previous 120 hours is used as a training set, the resource usage amount of the next 30 hours is used as a test set, the time series is divided into time subsequences by 10 minutes as a period, three parameters in the KNN prediction model are respectively set as k-5, q-1, and p-3 (k-5 represents finding 5 nearest neighbor sequences, q-1 represents using 1 time subsequences before the current time as a comparison sequence, and p-3 represents using 3 time slice data as a prediction period).
In the initialization of parameters of the KNN prediction model, the most important content is the content and the attributes related to the resource prediction and related to the container in the Kubernetes container cloud platform are represented numerically. Assume that the set of container applications in the Kubernetes cloud platform is V ═ V1,v2,v3,...,vn}TWherein v isiIndicating the ith container application service. Each container application service can be represented as
Wherein
Representing container application services viEach container application service contains n attributes, for example, attributes such as container mirror, runtime command, service port, running state, etc., and the value of the attribute a of the containerkCarrying out normalization operation to obtain normalized attribute vector
The resource usage of the container application service at runtime is described as
Wherein the content of the first and second substances,
representing the amount of resource usage of the CPU,
representing the resource usage of the memory. Since the representation of the resource usage amounts are similar, taking the resource usage amount of the CPU as an example here, the resource usage amount of the CPU from the start of service to the current time t of each container application can be represented as
t denotes the index of the time series, for example t-1 stands for the first cycle in the time series. In this embodiment, the historical resource usage amount of a container in the Kubernetes cloud platform is divided into time periods according to a fixed time period to form a time period sequence, and a time period sequence is formed according to the time period sequence
To predict future resource negativityCarrier
Wherein
Denotes the predicted value at t +1 th cycle, and p denotes the number of steps to predict.
Step four: and (5) training a model.
The model training is to adjust the model parameters according to the training result of each sample data, so that the model is finally trained towards the target direction, and the performance of the model is gradually improved. The selection of the hyperparameter k in the KNN prediction model has a large influence on the prediction result, the selection of the k value is large, overfitting occurs, and the selection of the k value is small, and under-fitting occurs. Therefore, the KNN prediction model needs to be trained using the historical data of each operated container in the historical database to determine the hyper-parameter k in the KNN prediction model. And when the model error meets a preset value or the training is finished.
The process of training the KNN prediction model is described below, taking prediction of CPU resource usage as an example.
And selecting the historical data of the operated container from a historical database for training. In other words, historical data of the first n time periods of the operated container is used as a training set and input into a KNN prediction model, historical data of the last m time periods is used as a test set, and the hyperparameter k of the KNN prediction model is adjusted in the algorithm operation process.
The specific steps of the KNN prediction model for prediction are as follows:
step 1.1, finding time sequence of container i from historical data of container i in training set
A most similar time series
Wherein t represents the current time period, t' represents a certain time period before t, q represents q sub-time sequences before t as oneTime series. Then, the next time sequence with the container i is continuously found from the historical data of the container i in the historical database
A most similar time sequence is found until k most similar time sequences are found, and k nearest neighbor time sequence sets N are obtainedk(t)。
In specific implementation, the similarity between two time sequences can be obtained by the following formula 1-1:
of course, the present embodiment is not limited to using norm to calculate the similarity between two time series, and other distance algorithms such as manhattan distance, euclidean distance, etc. may also be used to calculate the similarity.
Next, the resource demand for a future period of time is estimated by the k most similar time series found earlier. Generally considered as a time series
And
has a similar trend, then
The next time period of and
with a similar trend for the next time period.
Step 1.2, according to vessel i
Set NkK in (t)
And k are
Resource usage for next time period
Determining a predicted value for resource usage by container i at time period t +1
In particular, the amount of the solvent to be used,
the predicted value can be obtained by equations 1-2:
through step 1.1 and step 1.2, only the predicted value of the resource usage amount of the container to be tested at the time of the next time period t +1 of the time period t can be predicted. To predict the predicted value p time periods after time period t, the predicted value at time period t +1 may be
As the resource usage amount of the container to be tested in the time period t +1, taking t +1 as the current time period, and then performing step 1.1 and step 1.2 again, i.e. searching and sequencing
And (3) obtaining a predicted value of the resource usage amount of the container i in the time period t +2 by the most similar k time sequences, taking the predicted value in the time period t +2 as the resource usage amount of the container i in the time period t +2, taking t +2 as the current time period, and then executing the step 1.1 and the step 1.2 again to obtain the predicted value of the resource usage amount of the container i in the time period t + 3. Repeating the above steps p times in this way, and obtaining a predicted value set of resource usage for p time periods in the future:
further, in order to guarantee the service quality, the maximum value in the prediction value set is selected as the resource prediction allocation value of the container i.
Further, the model inevitably generates errors in the prediction process, and in order to reduce error interference, an error correction term is further added to the KNN prediction model of this embodiment, that is, a prediction error e is added to the determined resource prediction allocation value as the final resource prediction allocation value of the container i:
wherein e is the average value of the historical prediction errors, and is obtained through the formula 1-4:
and continuously selecting other operated containers from the historical database as containers to be tested, and training the KNN prediction model for multiple times to determine the hyperparameter k in the KNN prediction model.
Similarly, a KNN prediction model for predicting the usage amount of the memory resources can be trained.
Step five: and (6) testing the model.
And after the prediction model is trained for multiple times on the training set, testing on the testing set, judging whether the trained model meets the accuracy requirement or not according to the test result, stopping model training if the accuracy meets the training requirement, and otherwise, continuing to train the model.
And after preprocessing the test set, inputting the test set into the trained model to obtain the container resource quota of the test sample. In the model test engineering, indexes of the model performance such as test precision, test time and the like of the model are sequentially calculated and recorded.
Referring to fig. 1, based on the prediction model constructed above, an embodiment of the present invention provides a method for predicting resource quotas of multi-tenant containers, including the following steps:
s101, a first preset number of historical time sequences with the similarity of the first time sequence of the container to be detected and the similarity of the first preset number of the historical time sequences are found from the historical database, the first time sequence is a sequence of resource usage of the container to be detected in a second preset number of time periods before a first time period, the resource usage is arranged in time sequence, the first time period is a time period before a time period to be predicted, and the historical time sequences are sequences of resource usage of the container to be detected in the second preset number of time periods before a designated historical time period, the resource usage is arranged in time sequence.
In this embodiment, the resource usage includes CPU usage or memory usage.
Taking the CPU resource usage amount of the predicted container as an example, finding the time sequence of the container i to be tested from the historical data of the container i to be tested in the historical database
A most similar historical time series
Wherein t represents the first time period indicated in step S101, which may also be referred to as a current time period; t' represents a certain time period before t, that is, the specified historical time period referred to in step S101; q denotes q sub-time series before t as one time series, i.e., the second preset number referred to in step S101. Then, continuously finding the next time sequence of the container i to be tested from the historical data of the container i to be tested in the historical database
Until k most similar historical time sequences are found, k nearest neighbor time sequence sets N are obtainedk(t) of (d). k is the hyperparameter of the KNN prediction model, i.e. the first predetermined number referred to in step S101.
In specific implementation, the similarity between two time sequences can be obtained by the following formula 1-1:
of course, the present embodiment is not limited to using norm to calculate the similarity between two time series, and other distance algorithms such as manhattan distance, euclidean distance, etc. may also be used to calculate the similarity.
S102, determining a predicted value of the resource usage of the container to be measured in the time period to be predicted according to the resource usage of the container to be measured in the first time period and the first resource usage and the second resource usage in the historical time sequences of the first preset number, wherein the first resource usage is the resource usage of the container to be measured in the appointed historical time period, and the second resource usage is the resource usage of the container to be measured in the next time period of the appointed historical time period.
Next, the resource demand for a future period of time is estimated by the k most similar time series found earlier. Generally considered as a historical time series
And
has a similar trend, then
The next time period of and
with a similar trend for the next time period.
Further, S102 specifically includes: calculating the difference value between the first resource usage amount and the second resource usage amount in each historical time sequence; calculating the average value of the difference values of the first preset number; and adding the average value to the resource usage of the container to be measured in the first time period to obtain a predicted value of the resource usage of the container to be measured in the time period to be predicted.
In practice, according to the container i to be tested
Set NkK in (t)
And k are
Resource usage for next time period
Determining the predicted value of the resource usage amount of the container i to be tested in the time period t +1
In particular, the amount of the solvent to be used,
the predicted value can be obtained by equations 1-2:
through steps S101 and S102, only the predicted value of the resource usage amount of the container to be measured at the time of the next time period t +1 after the time period t can be predicted. To predict the predicted value p time periods after time period t, the predicted value at time period t +1 may be
As the resource usage amount of the container to be tested in the time period t +1, t +1 is taken as the first time period (i.e. the current time period), and then step S101 and step S102 are executed again. Repeating the above steps p times in this way, the predicted value of the resource usage for p time periods in the future can be obtained.
Therefore, as shown in fig. 2, after step S102, the method of the present embodiment further includes the steps of:
s103, judging whether the number of the obtained predicted values reaches a third preset number or not; if the number of the obtained predicted values does not reach a third preset number, executing step S104; if the number of the obtained predicted values has reached the third preset number, step S105 is executed.
And S104, taking the time period to be predicted as a new first time period, taking the preset value of the resource usage of the container to be tested in the time period to be predicted as the resource usage of the container to be tested in the new first time period, and returning to execute the step S101 and the step S102.
And S105, selecting the maximum value in the predicted values of the resource usage of the third preset number as the predicted value of the resource usage of the container to be tested.
In specific implementation, the predicted value at the time period of t +1 is used
As the resource usage amount of the container i to be tested in the time period t +1, taking t +1 as the first time period (i.e. the current time period), and then performing step S101 and step S102 again, i.e. the search and sequence
Calculating the predicted value of the resource usage amount of the container i to be tested in the time period t +2 by using the k most similar historical time sequences; and taking the predicted value in the time period t +2 as the resource usage amount of the container i to be tested in the time period t +2, taking t +2 as the first time period (namely the current time period), and then executing the step S101 and the step S102 again to obtain the predicted value of the resource usage amount of the container i to be tested in the time period t + 3. Repeating the above steps p times in this way, and obtaining a predicted value set of resource usage for p time periods in the future:
in order to guarantee the service quality, the maximum value in the prediction value set is selected as the prediction value of the resource usage amount of the container i to be tested in p time periods in the future.
Further, after step S104, the method of the present embodiment further includes: and adding the predicted value of the resource usage of the container to be tested with the prediction error to obtain the final predicted value of the resource usage of the container to be tested.
In order to reduce error interference, an error correction term is further added to the KNN prediction model of this embodiment, that is, a prediction error e is added to the determined resource prediction allocation value as a final resource prediction allocation value of the container i:
wherein e is the average value of the historical prediction errors, and is obtained through the formula 1-4:
and determining the resource quota of the container to be tested according to the final resource usage predicted value of the container to be tested, for example, taking the final resource usage of the container to be tested as the future resource quota of the container to be tested.
Compared with the traditional multi-tenant container resource quota planning method, the method provided by the embodiment of the invention does not simply guess the container resource quota according to the service type and experience, and accurately fits the change of the container resource consumption by utilizing a KNN prediction model and combining the characteristics of container self-similarity, time correlation and the like, so that the future resource demand of the container can be scientifically and accurately predicted, and the multi-tenant container resource quota is reasonably planned. The method of the embodiment of the invention not only reduces the time of data analysis in the process of planning the resource quota of the multi-tenant container, but also improves the accuracy of the prediction result, thereby more effectively managing the resources of the container cloud platform.
Before the container is deployed, the historical database does not have any historical operating data of the container, so that the resource demand of the container cannot be predicted by a time sequence regression method. However, there are many resource usage data of other containers in the history database, and a new container application Service needs to accurately allocate an appropriate initial resource amount, so as to avoid serious resource waste caused by the allocation of the appropriate initial resource amount, and ensure that an SLA (Service-Level agent) is not violated. In this embodiment, by matching the pre-run containers with the historical data of the existing containers (i.e., the run containers) in the database, the historical database not only includes the CPU resource usage and the memory resource usage of each run container, but also includes
Properties of a run Container
(e.g., container mirror, runtime command, service port, run state, etc.). And finding a container which is most similar to the user pre-run container from the plurality of operated containers, and using the peak value of the resource usage amount of the most similar container as the initial resource allocation amount of the pre-run container.
To this end, the method of this embodiment further includes the steps of: before the container to be detected runs, finding out a target container with the highest similarity to the container to be detected from the run containers according to the attribute data of the run containers; and taking the maximum value of the resource usage amount of the target container as the initial resource allocation amount of the container to be tested. The above steps are processes of predicting the initial resource usage of the container to be tested by using the similarity prediction model.
Further, according to the attribute data of the operated containers, finding the target container with the highest similarity to the container to be tested from the operated containers includes: calculating the norm of the attribute data of the container to be tested and the attribute data of each operated container; and taking the operated container corresponding to the maximum norm as a target container with the highest similarity to the container to be detected.
In this embodiment, the attribute data includes, but is not limited to, at least one of the following data: container mirroring, runtime commands, service ports, and run state. The attribute data may be represented in the form of a vector, e.g.
To calculate the attribute data and each operated container convenientlySimilarity of attribute data of the container.
Taking the CPU resource usage of the prediction container as an example, the specific steps of predicting by using the similarity prediction model are as follows:
k
step 2.1, carrying out normalization operation on the attribute value a of the pre-operation container, and obtaining an attribute vector representing the attribute of the pre-operation container through normalization
Using equation 2-1, calculate the attribute vector xi of the attributes of the pre-run container and the L2 norm of the attributes of each run container xj stored in the history database,
and 2.2, obtaining a container j which minimizes the DISTANCE according to the formula 2-2, wherein the container j is the container which is most similar to the pre-operation container.
Step 2.3, in order to avoid reducing the service quality, selecting a maximum value from the resource usage of the most similar container j as the initial resource allocation amount of the pre-run container, wherein the specific calculation method is shown in formula 2-2:
wherein, the value of l can be determined according to actual requirements.
Of course, the embodiment is not limited to using norm to calculate the similarity between the pre-run container and the run container, and may also use other distance algorithms, such as manhattan distance, euclidean distance, and the like.
As shown in fig. 3, based on the same inventive concept as the multi-tenant container resource quota predicting method, the embodiment of the present invention further provides a multi-tenant container resource quota predicting apparatus 30, a similar sequence extracting module 301 and a first predicting module 302.
The similar sequence extraction module 301 is configured to find a first preset number of historical time sequences from the historical database, where the first preset number of historical time sequences is a sequence in which resource usage of the container to be tested in a second preset number of time periods before a first time period is arranged in a time sequence, the first time period is a time period before a time period to be predicted, and the historical time sequences are sequences in which resource usage of the container to be tested in a second preset number of time periods before a specified historical time period is arranged in a time sequence.
The first prediction module 302 is configured to determine a predicted value of the resource usage of the to-be-predicted container in the to-be-predicted time period according to the resource usage of the to-be-detected container in the first time period and the first resource usage and the second resource usage in the first preset number of historical time sequences, where the first resource usage is the resource usage of the to-be-detected container in the specified historical time period, and the second resource usage is the resource usage of the to-be-detected container in the next time period of the specified historical time period.
Further, as shown in fig. 4, the multi-tenant container resource quota predicting apparatus 30 according to the embodiment of the present invention further includes a loop determining module 303, a parameter resetting module 304, and a predicted value selecting module 305.
The loop judgment module 304 is configured to judge whether the number of the obtained predicted values reaches a third preset number, and if the number of the obtained predicted values does not reach the third preset number, execute the parameter resetting module 303; if the number of the obtained predicted values has reached the third preset number, the predicted value selecting module 305 is executed.
The parameter resetting module 303 is configured to use the time period to be predicted as a new first time period, use a preset value of the resource usage amount of the container to be tested in the time period to be predicted as the resource usage amount of the container to be tested in the new first time period, and execute the similar sequence extracting module 301 and the first predicting module 302.
The predicted value selecting module 305 is configured to select a maximum value of the predicted values of the resource usage amounts of the third preset number as the predicted value of the resource usage amount of the container to be tested.
Further, the device 30 for predicting resource quota of multi-tenant container in this embodiment further includes an error correction module 306, configured to add the prediction error to the predicted value of the resource usage of the container to be tested to obtain a final predicted value of the resource usage of the container to be tested after selecting a maximum value of the predicted values of the resource usage of the third preset number as the predicted value of the resource usage of the container to be tested.
Further, the first prediction module 302 is specifically configured to: calculating the difference value between the first resource usage amount and the second resource usage amount in each historical time sequence; calculating the average value of the difference values of the first preset number; and adding the average value to the resource usage of the container to be measured in the first time period to obtain a predicted value of the resource usage of the container to be measured in the time period to be predicted.
Further, the multi-tenant container resource quota predicting device 30 of this embodiment further includes an initial predicting module, configured to, before the to-be-tested container runs, find, according to the attribute data of the run container, a target container with the highest similarity to the to-be-tested container from the run container; and taking the maximum value of the resource usage amount of the target container as the initial resource allocation amount of the container to be tested.
Further, the initial prediction module is specifically configured to: calculating the norm of the attribute data of the container to be tested and the attribute data of each operated container; and taking the operated container corresponding to the maximum norm as a target container with the highest similarity to the container to be detected.
In this embodiment, the attribute data includes but is not limited to at least one of the following: container mirroring, runtime commands, service ports, and run state.
In this embodiment, the resource usage includes, but is not limited to, CPU usage or memory usage.
Further, the multi-tenant container resource quota predicting apparatus 30 of this embodiment further includes a data collection module, which is configured to pull, in a pull manner, historical data of the container to be tested from a server running the container and store the historical data in a historical database.
The multi-tenant container resource quota predicting device and the multi-tenant container resource quota predicting method provided by the embodiment of the invention adopt the same inventive concept, can obtain the same beneficial effects, and are not repeated herein.
Based on the same inventive concept as the multi-tenant container resource quota prediction method, an embodiment of the present invention further provides an electronic device, which may be specifically a desktop computer, a portable computer, a smart phone, a tablet computer, a Personal Digital Assistant (PDA), a server, and the like. As shown in fig. 5, the electronic device 50 may include a processor 501, a memory 502, and a transceiver 503. The transceiver 503 is used to receive and transmit data under the control of the processor 501.
The processor 501 may be a CPU (central processing unit), an ASIC (Application Specific integrated circuit), an FPGA (Field Programmable Gate Array), or a CPLD (Complex Programmable Logic Device), and implements the multi-tenant container resource quota prediction method in any of the above embodiments according to an obtained program instruction by calling a program instruction stored in a memory.
An embodiment of the present invention provides a computer-readable storage medium, configured to store computer program instructions for the electronic device, where the computer program instructions include a program for executing the method for predicting resource quotas of multi-tenant containers.
The computer storage media may be any available media or data storage device that can be accessed by a computer, including but not limited to magnetic memory (e.g., floppy disks, hard disks, magnetic tape, magneto-optical disks (MOs), etc.), optical memory (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor memory (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND FLASH), Solid State Disks (SSDs)), etc.
The above embodiments are only used to describe the technical solutions of the present application in detail, but the above embodiments are only used to help understanding the method of the embodiments of the present invention, and should not be construed as limiting the embodiments of the present invention. Variations or substitutions that may be readily apparent to one skilled in the art are intended to be included within the scope of the embodiments of the present invention.
Claims (10)
1. A multi-tenant container resource quota prediction method is characterized by comprising the following steps:
a similar sequence extraction step, namely finding a first preset number of historical time sequences with the similarity of a first time sequence of a container to be detected, wherein the similarity of the first preset number of historical time sequences is ranked at the top, the first time sequence is a sequence of resource usage of the container to be detected in a second preset number of time periods before a first time period, the resource usage of the container to be detected is ranked in a time sequence, the first time period is a time period before a time period to be predicted, and the historical time sequence is a sequence of resource usage of the container to be detected in the second preset number of time periods before a specified historical time period, the resource usage of the container to be detected is ranked in the time sequence;
and a predicting step, namely determining a predicted value of the resource usage of the container to be tested in the time period to be predicted according to the resource usage of the container to be tested in the first time period and a first preset number of resource usage and a second resource usage in the historical time sequence, wherein the first resource usage is the resource usage of the container to be tested in the appointed historical time period, and the second resource usage is the resource usage of the container to be tested in the next time period of the appointed historical time period.
2. The method of claim 1, further comprising, after the predicting step:
a circulation step, namely taking the time period to be predicted as a new first time period, and taking a preset value of the resource usage of the container to be tested in the time period to be predicted as the resource usage of the container to be tested in the new first time period;
repeating the similar sequence extraction step, the prediction step and the circulation step until the predicted values of the resource usage amount of the third preset number are obtained;
and selecting the maximum value in the predicted values of the resource usage of the third preset number as the predicted value of the resource usage of the container to be tested.
3. The method according to claim 2, wherein after selecting a maximum value of the predicted values of the resource usage by a third preset number as the predicted value of the resource usage by the container to be tested, the method further comprises:
and adding the prediction error to the predicted value of the resource usage of the container to be tested to obtain a final predicted value of the resource usage of the container to be tested.
4. The method according to any one of claims 1 to 3, wherein the predicting step comprises:
calculating the difference value of the first resource usage amount and the second resource usage amount in each historical time sequence;
calculating the average value of the difference values of a first preset number;
and adding the average value to the resource usage of the container to be measured in the first time period to obtain a predicted value of the resource usage of the container to be measured in the time period to be predicted.
5. The method of any of claims 1 to 3, further comprising:
before the container to be detected runs, according to the attribute data of the run container, finding a target container with the highest similarity to the container to be detected from the run container;
and taking the maximum value of the resource usage amount of the target container as the initial resource allocation amount of the container to be tested.
6. The method according to claim 5, wherein the finding a target container with the highest similarity to the container to be tested from the executed containers according to the attribute data of the executed containers comprises:
calculating the norm of the attribute data of the container to be tested and the attribute data of each operated container;
and taking the operated container corresponding to the maximum norm as a target container with the highest similarity to the container to be detected.
7. The method of claim 1, wherein the resource usage comprises CPU usage or memory usage.
8. A multi-tenant container resource quota predicting apparatus, comprising:
the system comprises a similarity sequence extraction module, a similarity sequence extraction module and a comparison module, wherein the similarity sequence extraction module is used for finding a first preset number of historical time sequences which are ranked at the top with the similarity of a first time sequence of a container to be detected, the first time sequence is a sequence formed by arranging the resource usage of the container to be detected in a second preset number of time periods before a first time period according to the time sequence, the first time period is a time period before a time period to be predicted, and the historical time sequence is a sequence formed by arranging the resource usage of the container to be detected in the second preset number of time periods before a specified historical time period according to the time sequence;
the first prediction module is used for determining a predicted value of the resource usage of the container to be predicted in the time period to be predicted according to the resource usage of the container to be predicted in the first time period and a first preset number of first resource usage and second resource usage in the historical time sequence, the first resource usage is the resource usage of the container to be predicted in the appointed historical time period, and the second resource usage is the resource usage of the container to be predicted in the next time period of the appointed historical time period.
9. An electronic device comprising a transceiver, a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the transceiver is configured to receive and transmit data under control of the processor, and wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the program.
10. A computer-readable storage medium, on which computer program instructions are stored, which program instructions, when executed by a processor, carry out the steps of the method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811534570.3A CN111327655A (en) | 2018-12-14 | 2018-12-14 | Multi-tenant container resource quota prediction method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811534570.3A CN111327655A (en) | 2018-12-14 | 2018-12-14 | Multi-tenant container resource quota prediction method and device and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111327655A true CN111327655A (en) | 2020-06-23 |
Family
ID=71168397
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811534570.3A Pending CN111327655A (en) | 2018-12-14 | 2018-12-14 | Multi-tenant container resource quota prediction method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111327655A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111813631A (en) * | 2020-07-15 | 2020-10-23 | 江苏方天电力技术有限公司 | Resource situation visualization and analysis method for cloud data center |
| CN112001116A (en) * | 2020-07-17 | 2020-11-27 | 新华三大数据技术有限公司 | Cloud resource capacity prediction method and device |
| CN112016797A (en) * | 2020-07-15 | 2020-12-01 | 北京淇瑀信息科技有限公司 | KNN-based resource quota adjusting method and device and electronic equipment |
| CN114185642A (en) * | 2021-11-12 | 2022-03-15 | 联奕科技股份有限公司 | Intelligent campus development method and system based on container management platform |
| CN116257360A (en) * | 2023-03-09 | 2023-06-13 | 上海道客网络科技有限公司 | Method and system for planning container group resources based on historical usage data |
| CN117472586A (en) * | 2023-12-05 | 2024-01-30 | 支付宝(杭州)信息技术有限公司 | Training method and device of time sequence model, and memory management method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108092797A (en) * | 2017-11-21 | 2018-05-29 | 北京奇艺世纪科技有限公司 | A kind of Container Management method and device |
| CN108920153A (en) * | 2018-05-29 | 2018-11-30 | 华南理工大学 | A kind of Docker container dynamic dispatching method based on load estimation |
| US20180349797A1 (en) * | 2017-06-02 | 2018-12-06 | Oracle International Corporation | Data driven methods and systems for what if analysis |
-
2018
- 2018-12-14 CN CN201811534570.3A patent/CN111327655A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180349797A1 (en) * | 2017-06-02 | 2018-12-06 | Oracle International Corporation | Data driven methods and systems for what if analysis |
| CN108092797A (en) * | 2017-11-21 | 2018-05-29 | 北京奇艺世纪科技有限公司 | A kind of Container Management method and device |
| CN108920153A (en) * | 2018-05-29 | 2018-11-30 | 华南理工大学 | A kind of Docker container dynamic dispatching method based on load estimation |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111813631A (en) * | 2020-07-15 | 2020-10-23 | 江苏方天电力技术有限公司 | Resource situation visualization and analysis method for cloud data center |
| CN112016797A (en) * | 2020-07-15 | 2020-12-01 | 北京淇瑀信息科技有限公司 | KNN-based resource quota adjusting method and device and electronic equipment |
| CN112016797B (en) * | 2020-07-15 | 2024-03-01 | 北京淇瑀信息科技有限公司 | KNN-based resource quota adjustment method and device and electronic equipment |
| CN112001116A (en) * | 2020-07-17 | 2020-11-27 | 新华三大数据技术有限公司 | Cloud resource capacity prediction method and device |
| CN114185642A (en) * | 2021-11-12 | 2022-03-15 | 联奕科技股份有限公司 | Intelligent campus development method and system based on container management platform |
| CN114185642B (en) * | 2021-11-12 | 2023-11-17 | 联奕科技股份有限公司 | Intelligent campus development method and system based on container management platform |
| CN116257360A (en) * | 2023-03-09 | 2023-06-13 | 上海道客网络科技有限公司 | Method and system for planning container group resources based on historical usage data |
| CN116257360B (en) * | 2023-03-09 | 2023-09-08 | 上海道客网络科技有限公司 | Method and system for planning container group resources based on historical usage data |
| CN117472586A (en) * | 2023-12-05 | 2024-01-30 | 支付宝(杭州)信息技术有限公司 | Training method and device of time sequence model, and memory management method and device |
| CN117472586B (en) * | 2023-12-05 | 2024-03-12 | 支付宝(杭州)信息技术有限公司 | Training method and device of time sequence model, and memory management method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111327655A (en) | Multi-tenant container resource quota prediction method and device and electronic equipment | |
| US10713073B2 (en) | Systems and methods for identifying cloud configurations | |
| US20190122078A1 (en) | Search method and apparatus | |
| CN110969290B (en) | Runoff probability prediction method and system based on deep learning | |
| CN106549772A (en) | Resource prediction method, system and capacity management device | |
| CN110727506B (en) | SPARK parameter automatic tuning method based on cost model | |
| US11650968B2 (en) | Systems and methods for predictive early stopping in neural network training | |
| CN112200296B (en) | Network model quantization method and device, storage medium and electronic equipment | |
| US10248462B2 (en) | Management server which constructs a request load model for an object system, load estimation method thereof and storage medium for storing program | |
| CN116244069A (en) | Capacity expansion and contraction method and device, electronic equipment and readable storage medium | |
| Cordingly et al. | Function memory optimization for heterogeneous serverless platforms with cpu time accounting | |
| CN110716761A (en) | Automatic and self-optimizing determination of execution parameters of software applications on an information processing platform | |
| CN117318033B (en) | Power grid data management method and system combining data twinning | |
| CN116737334A (en) | Task scheduling and data set label updating method and device and electronic equipment | |
| JPWO2018168695A1 (en) | Distributed machine learning device, distributed machine learning method, and distributed machine learning program | |
| EP3580676A1 (en) | Sample-based multidimensional data cloning | |
| JP7424373B2 (en) | Analytical equipment, analytical methods and analytical programs | |
| CN107665185A (en) | A kind of customer behavior analysis method and relevant apparatus based on short message service | |
| CN113610225A (en) | Quality evaluation model training method and device, electronic equipment and storage medium | |
| CN112861951B (en) | Image neural network parameter determining method and electronic equipment | |
| Glavan et al. | Cloud environment assessment using clustering techniques on microservices dataset | |
| CN117807411B (en) | Server performance index prediction method and device and electronic equipment | |
| US20230351146A1 (en) | Device and computer-implemented method for a neural architecture search | |
| US11928562B2 (en) | Framework for providing improved predictive model | |
| CN117933446A (en) | Wind power generation prediction method, device, network 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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200623 |