67 0

시스템 성능 개선을 위한 기계학습 기반의 동시접속자 수 예측 모델

Title
시스템 성능 개선을 위한 기계학습 기반의 동시접속자 수 예측 모델
Other Titles
A Predictive Model for the Number of Concurrent Users based on Machine Learning for Improvement of System Performance
Author
이정근
Alternative Author(s)
Lee, Jung Keun
Advisor(s)
조인휘
Issue Date
2020-02
Publisher
한양대학교
Degree
Master
Abstract
현대의 업무 처리는 대부분이 웹기반 환경의 시스템으로 구성되어 있다. 웹기반의 업무환경은 다수의 사용자가 동시에 웹 애플리케이션 서버에 접근 시도가 증가하게 되었으며 그에 맞게 늘어난 동시접속자 수를 얼마나 잘 수용할 수 있는가 하는 것이 중요하게 되었다. 동시접속자가 일시적으로 몰리게 되면 웹서비스의 속도가 느려지면서 업무처리가 지연되는 경우가 발생된다. 만약 현재 시점의 동시접속자를 예측하여 제공하게 된다면 웹서비스를 제공하는 측면에서 매우 유용할 것이다. 이에 동시접속자 수를 예측할 수 있는 모델을 제안해 보고자 한다. 본 논문에서는 웹 애플리케이션 동시접속자 수를 예측을 위해 기계학습 알고리즘 중 회귀분석과 딥러닝 방식의 순환신경망(RNN) 알고리즘을 훈련데이터 세트 비율을 단계적(10% ~ 90%)으로 높여가면서 최적의 정확도 값을 예측하였다. 회귀분석의 경우 훈련세트 비율이 커질수록 오차율이 감소하였으며, 순환신경망(RNN)의 경우는 훈련세트 비율과 오차율의 관계가 비례하지는 않았다. 훈련세트 비율이 증가됨에 따라 오차율의 증감을 반복하는 패턴을 보였으며 훈련세트 비율이 70% 이후 오차율이 증가되었다. 동시접속자 수 예측에 있어 회귀분석과 순환신경망(RNN)을 통해 오차율을 비교한 결과 순환신경망(RNN)의 평균제곱근오차(RMSE) 값이 회귀분석 결과값 보다 최소값으로 나와 가장 적합한 알고리즘임을 확인하였다.| Most modern business processes consist of web-based systems. Web-based work environment has increased the number of users attempting to access the web application server at the same time, and how well it can accommodate the increased number of concurrent users Temporary congestion of concurrent users can slow down web services and delay business processes. If we predict and provide concurrent users at the present time, it will be very useful in terms of providing web services. Therefore, we propose a model that can predict the number of concurrent users. In this paper, the regression analysis and deep learning RNN algorithm among the machine learning algorithms for predicting the number of concurrent users of web applications is increased by increasing the training data set step by step (10% to 90%). Was predicted. In the regression analysis, the error rate decreased as the training set ratio increased. In the case of RNN, the relationship between the training set ratio and the error rate was not proportional. As the training set ratio increased, the pattern of repeating the error rate increased and decreased, and the training rate increased after 70%. As a result of comparing the error rate through the regression analysis and the RNN, it was confirmed that the mean square error (RMSE) value of the RNN was the minimum value than the result of the regression analysis.
Most modern business processes consist of web-based systems. Web-based work environment has increased the number of users attempting to access the web application server at the same time, and how well it can accommodate the increased number of concurrent users Temporary congestion of concurrent users can slow down web services and delay business processes. If we predict and provide concurrent users at the present time, it will be very useful in terms of providing web services. Therefore, we propose a model that can predict the number of concurrent users. In this paper, the regression analysis and deep learning RNN algorithm among the machine learning algorithms for predicting the number of concurrent users of web applications is increased by increasing the training data set step by step (10% to 90%). Was predicted. In the regression analysis, the error rate decreased as the training set ratio increased. In the case of RNN, the relationship between the training set ratio and the error rate was not proportional. As the training set ratio increased, the pattern of repeating the error rate increased and decreased, and the training rate increased after 70%. As a result of comparing the error rate through the regression analysis and the RNN, it was confirmed that the mean square error (RMSE) value of the RNN was the minimum value than the result of the regression analysis.
URI
http://dcollection.hanyang.ac.kr/common/orgView/000000111069https://repository.hanyang.ac.kr/handle/20.500.11754/122956
Appears in Collections:
GRADUATE SCHOOL OF ENGINEERING[S](공학대학원) > ELECTRICAL ENGINEERING AND COMPUTER SCIENCE(전기ㆍ전자ㆍ컴퓨터공학과) > Theses (Master)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE