CN114461538A - Cloud computing application memory management method based on real-time content prediction and historical resource occupation - Google Patents

Cloud computing application memory management method based on real-time content prediction and historical resource occupation Download PDF

Info

Publication number
CN114461538A
CN114461538A CN202210089562.2A CN202210089562A CN114461538A CN 114461538 A CN114461538 A CN 114461538A CN 202210089562 A CN202210089562 A CN 202210089562A CN 114461538 A CN114461538 A CN 114461538A
Authority
CN
China
Prior art keywords
memory
time
application program
occupation
cloud computing
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
Application number
CN202210089562.2A
Other languages
Chinese (zh)
Inventor
刘东海
徐育毅
庞辉富
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Youyun Software Co ltd
Beijing Guangtong Youyun Technology Co ltd
Original Assignee
Hangzhou Youyun Software Co ltd
Beijing Guangtong Youyun Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hangzhou Youyun Software Co ltd, Beijing Guangtong Youyun Technology Co ltd filed Critical Hangzhou Youyun Software Co ltd
Priority to CN202210089562.2A priority Critical patent/CN114461538A/en
Publication of CN114461538A publication Critical patent/CN114461538A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/0223—User address space allocation, e.g. contiguous or non contiguous base addressing
    • G06F12/023—Free address space management
    • G06F12/0253—Garbage collection, i.e. reclamation of unreferenced memory
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

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

Abstract

The invention relates to the field of cloud environment, in particular to a cloud computing application memory management method based on real-time content prediction and historical resource occupation. The beneficial effects of the invention are as follows: the invention combines real-time memory occupation prediction and historical resource usage statistical records, provides an integrated application memory expansion scheme, and improves the operation efficiency of application programs in the cloud computing environment under the condition of avoiding memory errors.

Description

Cloud computing application memory management method based on real-time content prediction and historical resource occupation
Technical Field
The invention relates to the field of cloud computing, in particular to a cloud computing application memory management method based on real-time content prediction and historical resource occupation.
Background
With the maturity of big data and cloud computing technologies, the high performance computing demand of enterprises for big data processing is increasing. Various applications are deployed on the cloud to enable efficient parallel computing and on-demand allocation of computational resources. But in the face of an unlimited increase in the amount of data, the number of applications, and the complexity of computational tasks, physical resources are ultimately limited.
In a cloud computing environment, memory management of applications has been a difficult problem in the industry. Although the advent of many big data computing platform systems solved the technical challenges of parallel computing. However, when processing large application computing tasks that run for long periods of time, memory errors often occur, which devote both developer's historical time and computing resource investment. The best way to solve such problems is to increase the physical resources of the cloud computing environment, but doing so is not only costly but also results in unnecessary redundancy. Currently, there is also a scheme for scheduling resources by predicting the memory occupation of an application program, but such a method is not real-time enough, and does not effectively balance the performance of the application program with the memory overhead.
Currently, research on application memory management schemes of cloud computing platforms is less. The invention CN113296880A provides a container-based application management method. By configuring a serverless computing system implemented on a container basis, two states are set for an application: online and low power consumption. For the requirements of capacity reduction and capacity expansion, an application instance of the application is switched between two states respectively. The method defines the online and low-power consumption states and the switching steps in detail, but does not provide the switching conditions, and cannot realize automatic real-time memory management, so that the method is not suitable for an enterprise cloud computing platform with massive application data, limited physical resources and complex computing tasks.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a cloud computing application memory management method based on real-time content prediction and historical resource occupation by combining real-time memory occupation prediction and historical resource usage statistical records.
The object of the present invention is achieved by the following technical means. Aiming at the problem of memory management of an application program in a cloud environment, the invention realizes real-time prediction of a memory at the next moment based on memory occupation time sequence data of a fixed time window of the application program, combines historical resource usage statistical records of the full life cycle of the application program, and realizes an integrated automatic memory recovery method for cloud computing application through a reinforcement learning model. The cloud computing application memory management method based on real-time content prediction and historical resource occupation comprises the following steps:
(1) application program in given cloud computing environment
Figure DEST_PATH_IMAGE001
Checking points at fixed time intervals T, recording application program
Figure 756100DEST_PATH_IMAGE001
In the past memory occupation situation of n check points, the application program
Figure 1136DEST_PATH_IMAGE001
Memory occupation record at the t-th check point, i.e. memory occupation record at time t
Figure DEST_PATH_IMAGE002
Comprises the following steps:
Figure DEST_PATH_IMAGE003
(2) inputting the application program at the time t based on the long-time memory network LSTM
Figure 515294DEST_PATH_IMAGE001
(application i) memory footprint record
Figure 238400DEST_PATH_IMAGE002
Application program with output at t +1 moment
Figure 521614DEST_PATH_IMAGE001
Memory footprint prediction
Figure DEST_PATH_IMAGE004
：
Figure DEST_PATH_IMAGE005
In each step of iterative operation later, at a given time t, the application program at the next time, namely the time t +1, is predicted based on the memory occupation records of the previous n check points
Figure 701666DEST_PATH_IMAGE001
Memory footprint of, the vector
Figure 171961DEST_PATH_IMAGE004
Application program in cloud computing environment as time t
Figure 964337DEST_PATH_IMAGE001
The first part of the state evaluation, namely:
Figure DEST_PATH_IMAGE006
；
(3) setting a log manager to record resource usage records of all application programs in a full life cycle, and specifically comprising the following steps: maximum memory occupation max _ memory, minimum memory occupation min _ memory, maximum CPU occupation max _ CPU, minimum CPU occupation min _ CPU, called times call _ count, total running time total _ time, average running time average _ time, maximum running time max _ time, minimum running time min _ time, called average interval average _ interval, called maximum interval max _ interval, called minimum interval min _ interval, and until t, the application program
Figure 102057DEST_PATH_IMAGE001
History of
Figure DEST_PATH_IMAGE007
Comprises the following steps:
Figure DEST_PATH_IMAGE008
the vector
Figure 954476DEST_PATH_IMAGE007
Evaluating a second part as a state of the cloud computing environment, namely:
Figure DEST_PATH_IMAGE009
therefore, the application program in the cloud computing environment at the time t is obtained
Figure 177646DEST_PATH_IMAGE001
The state of (a) was evaluated as:
Figure DEST_PATH_IMAGE010
；
(4) based on the enhanced learning model DQN, the input of the DQN model is
Figure 508134DEST_PATH_IMAGE011
And performing memory management on a given application program in the cloud environment.
The invention has the beneficial effects that: the invention provides an integrated application memory expansion scheme by combining real-time memory occupation prediction and historical resource usage statistical records, and improves the operation efficiency of application programs in a cloud computing environment under the condition of avoiding memory errors.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
Detailed Description
The invention will be described in detail below with reference to the following drawings:
as shown in fig. 1, the present invention provides a cloud computing application memory management method based on real-time content prediction and historical resource occupancy, which includes the following steps:
a) application in a given cloud computing environment
Figure DEST_PATH_IMAGE012
. Setting check point at fixed time interval T, recording application program
Figure 500360DEST_PATH_IMAGE012
Memory footprint of n checkpoints in the past. Application program
Figure 759566DEST_PATH_IMAGE001
Memory footprint record at the tth checkpoint (hereinafter referred to as time t)
Figure 266770DEST_PATH_IMAGE013
Comprises the following steps:
Figure 745156DEST_PATH_IMAGE003
b) inputting an application program at the time t based on a long-time memory network LSTM (well-known algorithm)
Figure 654206DEST_PATH_IMAGE012
Memory usage record of
Figure 317269DEST_PATH_IMAGE013
Application program with output at t +1 moment
Figure 515032DEST_PATH_IMAGE012
Memory usage prediction:
Figure 859426DEST_PATH_IMAGE005
in each step of iterative operation later, at a given time t, the application program at the next time, namely the time t +1, is predicted based on the memory occupation records of the previous n check points
Figure 685299DEST_PATH_IMAGE012
The memory footprint of. The vector is used as an application program in a cloud computing environment at the time t
Figure 722525DEST_PATH_IMAGE012
The first part of the state evaluation, namely:
Figure 876426DEST_PATH_IMAGE006
c) setting a log manager to record resource usage records of all application programs in a full life cycle, and specifically comprising the following steps: maximum memory occupation max _ memory, minimum memory occupation min _ memory, maximum CPU occupation max _ CPU, minimum CPU occupation min _ CPU, called times call _ count, total running time total _ time, average running time average _ time, maximum running time max _ time, minimum running time min _ time, called average interval average _ interval, called maximum interval max _ interval, called minimum interval min _ interval, and until t, the application program
Figure 24511DEST_PATH_IMAGE012
The history of (c) is:
Figure 704891DEST_PATH_IMAGE008
the vector evaluates a second part as a state of the cloud computing environment, namely:
Figure 913018DEST_PATH_IMAGE009
thus obtaining the application program in the cloud computing environment at the time t
Figure 554215DEST_PATH_IMAGE012
The state of (a) is evaluated as:
Figure 240412DEST_PATH_IMAGE010
；
d) and performing memory management on a given application program in the cloud environment based on a reinforcement learning model DQN (known method). What needs to be specifically stated in DQN includes state space, action space, and rewards. Wherein the definition of the continuous state space has been defined in detail in a, b, c.
We define the memory change action space for an application as follows:
1. unchanged, the current memory is not processed
2. 0.8, memory occupation is reduced to 0.8 times of the current value
3. 0.6, memory occupancy is reduced to 0.6 times of the current value
4. 0.4, memory usage is reduced to 0.4 times of the current value
5. 0.2, memory occupancy is reduced to 0.2 times of the current value
6. And Kill terminates the running of the current application program and recycles all the memory.
The reward function defines:
the purpose of the scheme is to reduce the memory overhead of the application program as much as possible on the premise of ensuring the normal and efficient operation of the application program, so that the reward function of the memory management of the application program is composed of the following points
1. The memory resource occupies M, the negative feedback is carried out, the less the occupation is, the higher the reward is
2. The running time T of the application program, negative feedback, the shorter the running time, the higher the reward
Figure DEST_PATH_IMAGE014
Here the reward normalization process is performed based on the maximum minimum memory usage and run time in the history of different applications,
Figure 804992DEST_PATH_IMAGE015
and
Figure DEST_PATH_IMAGE016
respectively, the weights of the two prizes.
The input of the DQN model is the state
Figure 387283DEST_PATH_IMAGE017
That is, the vector for predicting the memory occupation of the application program at the time t +1 and splicing the history records of the full life cycle of the application program before the time t is output as a 6-dimensional vectorThe vectors of (1) correspond to Q values of 6 actions, respectively, and are based on the selected action
Figure DEST_PATH_IMAGE018
An algorithm to
Figure 109251DEST_PATH_IMAGE019
To randomly select action with a small probability to
Figure DEST_PATH_IMAGE020
The operation action corresponding to the maximum Q value of the DQN output is selected. And optimizing the DQN network based on experience replay of data collected in the experience pool. Finally, a cloud computing application memory recovery scheme based on real-time memory prediction and historical records is achieved.
The invention has the characteristics that: the invention carries out memory prediction through LSTM, but the prediction is not the end point of our task, we take the prediction result as an intermediate result, the intermediate result is finally served by the memory scaling action, and the scaling action is carried out through DQN.
It should be understood that equivalent substitutions and changes to the technical solution and the inventive concept of the present invention should be made by those skilled in the art to the protection scope of the appended claims.

Claims (2)

1. A cloud computing application memory management method based on real-time content prediction and historical resource occupation is characterized by comprising the following steps: the method comprises the following steps:
(1) application program in given cloud computing environment
Figure 128384DEST_PATH_IMAGE001
Checking points at fixed time intervals T, recording application program
Figure 763633DEST_PATH_IMAGE002
In the past memory occupation situation of n check points, the application program
Figure 733251DEST_PATH_IMAGE002
Memory occupation record at the t-th check point, i.e. memory occupation record at time t
Figure 518673DEST_PATH_IMAGE003
Comprises the following steps:
Figure 785575DEST_PATH_IMAGE004
；
(2) inputting the application program at the time t based on the long-time memory network LSTM
Figure 266760DEST_PATH_IMAGE002
Memory usage record of
Figure 924006DEST_PATH_IMAGE003
Application program with output at t +1 moment
Figure 778698DEST_PATH_IMAGE002
Memory footprint prediction
Figure 903037DEST_PATH_IMAGE005
：
Figure 286614DEST_PATH_IMAGE006
In each step of iterative operation later, at a given time t, the application program at the next time, namely the time t +1, is predicted based on the memory occupation records of the previous n check points
Figure 493473DEST_PATH_IMAGE002
The memory usage of (a) is,
Figure 900925DEST_PATH_IMAGE005
application program in cloud computing environment as time t
Figure 814524DEST_PATH_IMAGE002
The first part of the state evaluation, namely:
Figure 696898DEST_PATH_IMAGE007
；
(3) setting a log manager to record resource usage records of all application programs in a full life cycle, and specifically comprising the following steps: maximum memory occupation max _ memory, minimum memory occupation min _ memory, maximum CPU occupation max _ CPU, minimum CPU occupation min _ CPU, called times call _ count, total running time total _ time, average running time average _ time, maximum running time max _ time, minimum running time min _ time, called average interval average _ interval, called maximum interval max _ interval, called minimum interval min _ interval, and until t, the application program
Figure 331666DEST_PATH_IMAGE002
History of
Figure 262582DEST_PATH_IMAGE008
Comprises the following steps:
Figure 358583DEST_PATH_IMAGE009
Figure 883629DEST_PATH_IMAGE008
evaluating a second part as a state of the cloud computing environment, namely:
Figure 268343DEST_PATH_IMAGE010
therefore, the application program in the cloud computing environment at the time t is obtained
Figure 737371DEST_PATH_IMAGE002
The state of (a) was evaluated as:
Figure 894070DEST_PATH_IMAGE011
；
(4) based on the enhanced learning model DQN, the input of the DQN model is
Figure 118247DEST_PATH_IMAGE012
And performing memory management on a given application program in the cloud environment.
2. The cloud computing application memory management method based on real-time content prediction and historical resource occupancy according to claim 1, wherein: in the reinforced learning model DQN, the DQN,
the memory change action space for an application is defined as follows:
(1) if the memory is not changed, the current memory is not processed;
(2) 0.8, reducing the memory occupation to 0.8 times of the current value;
(3) 0.6, reducing the memory occupation to 0.6 times of the current value;
(4) 0.4, reducing the memory occupation to 0.4 times of the current value;
(5) 0.2, reducing the memory occupation to 0.2 times of the current value;
(6) and Kill, stopping running the current application program and recycling all the memories;
the reward function r defines:
Figure 459098DEST_PATH_IMAGE013
memory resource occupation M and application program running time T, reward normalization processing is carried out according to the maximum and minimum memory occupation and running time in the history records of different application programs,
Figure 519765DEST_PATH_IMAGE014
and
Figure 528042DEST_PATH_IMAGE015
respectively, the weights of the two prizes.
CN202210089562.2A 2022-01-26 2022-01-26 Cloud computing application memory management method based on real-time content prediction and historical resource occupation Pending CN114461538A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210089562.2A CN114461538A (en) 2022-01-26 2022-01-26 Cloud computing application memory management method based on real-time content prediction and historical resource occupation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210089562.2A CN114461538A (en) 2022-01-26 2022-01-26 Cloud computing application memory management method based on real-time content prediction and historical resource occupation

Publications (1)

Publication Number Publication Date
CN114461538A true CN114461538A (en) 2022-05-10

Family

ID=81411022

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210089562.2A Pending CN114461538A (en) 2022-01-26 2022-01-26 Cloud computing application memory management method based on real-time content prediction and historical resource occupation

Country Status (1)

Country Link
CN (1) CN114461538A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116719648A (en) * 2023-08-10 2023-09-08 泰山学院 Data management method and system for computer system
WO2024114484A1 (en) * 2022-12-02 2024-06-06 中国科学院深圳先进技术研究院 Serverless computing adaptive resource scheduling method and system and computer device

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024114484A1 (en) * 2022-12-02 2024-06-06 中国科学院深圳先进技术研究院 Serverless computing adaptive resource scheduling method and system and computer device
CN116719648A (en) * 2023-08-10 2023-09-08 泰山学院 Data management method and system for computer system
CN116719648B (en) * 2023-08-10 2023-11-07 泰山学院 Data management method and system for computer system

Similar Documents

Publication Publication Date Title
CN111309479B (en) Method, device, equipment and medium for realizing task parallel processing
CN114461538A (en) Cloud computing application memory management method based on real-time content prediction and historical resource occupation
CN110531996B (en) Particle swarm optimization-based computing task unloading method in multi-micro cloud environment
US11928599B2 (en) Method and device for model compression of neural network
CN108764458B (en) Method and system for reducing storage space consumption and calculation amount of mobile equipment
CN111274111B (en) Prediction and anti-aging method for microservice aging
CN113806094A (en) Cloud platform resource dynamic scheduling method based on deep learning
CN110728372B (en) Cluster design method and cluster system for dynamic loading of artificial intelligent model
CN113342504A (en) Intelligent manufacturing edge calculation task scheduling method and system based on cache
Loukopoulos et al. A pareto-efficient algorithm for data stream processing at network edges
US20220343146A1 (en) Method and system for temporal graph neural network acceleration
Zamirai et al. Sieve: Speculative inference on the edge with versatile exportation
CN114301911A (en) Task management method and system based on edge-edge cooperation
CN113344073A (en) Daily load curve clustering method and system based on fusion evolution algorithm
CN111427687A (en) Artificial intelligence cloud platform
WO2021262139A1 (en) Distributed machine learning models
CN110969259A (en) Processing core with data-dependent adaptive rounding
CN110647381A (en) Virtual machine resource balancing and deployment optimizing method
Wu et al. An estimation of distribution algorithm to optimize the utility of task scheduling under fog computing systems
CN117787444B (en) Intelligent algorithm rapid integration method and device for cluster countermeasure scene
CN112148464B (en) Method and system for unloading mobile edge computing task
CN116414556B (en) Heterogeneous embedded equipment power distribution system and method based on redundant calculation force
CN116842994B (en) Dynamic optimization method and system for execution efficiency of multiple neural networks
CN117319373A (en) Data transmission method, device, electronic equipment and computer readable storage medium
CN118018610A (en) Cloud edge cooperative resource scheduling method, system, electronic equipment and readable 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