Design and Performance Comparison of Docker Container Based Deep Learning Model Management System for Real-Time Analysis 


Vol. 46,  No. 2, pp. 390-400, Feb.  2021
10.7840/kics.2021.46.2.390


PDF
  Abstract

Deep learning technology is known to be very effective in inferring the results of high-dimensional data and can be applied to various business areas. With the development of the deep learning framework, various artificial intelligence services are emerging. In particular, Smart City, which has recently become a hot topic, provides artificial intelligence services based on various data from cities. However, most artificial intelligence services have limitations in efficiently using resources only to satisfy the inference accuracy of models using deep learning technology. Even when configuring a web or platform architecture for actual service application, costs for operating various infrastructure technologies from data inflow, pre-processing, model learning, and serving are incurred. In the case of smart city services, operating costs are expected to be large because they are operated based on vast amounts of data. In this study, various architectures are designed for the development, distribution, and management of stable web services equipped with deep learning models, and the performance of the architectures is compared. Performance comparisons were made for a total of four architectures: Embedded function architecture, which is a general function execution method, and Flask API, Fast API, and TFServing architectures, which are REST methods. As a result of the experiment, TFServing showed the best performance in terms of data processing speed, but when considering the types of models and customization that can be mounted, Fast API, which shows sufficient speed, was also shown to be sufficient to develop a stable architecture.

  Statistics
Cumulative Counts from November, 2022
Multiple requests among the same browser session are counted as one view. If you mouse over a chart, the values of data points will be shown.


  Cite this article

[IEEE Style]

M. Lee, M. Kang, I. Kim, J. Kim, "Design and Performance Comparison of Docker Container Based Deep Learning Model Management System for Real-Time Analysis," The Journal of Korean Institute of Communications and Information Sciences, vol. 46, no. 2, pp. 390-400, 2021. DOI: 10.7840/kics.2021.46.2.390.

[ACM Style]

Mo-se Lee, Min-su Kang, In-ho Kim, and Jae-hun Kim. 2021. Design and Performance Comparison of Docker Container Based Deep Learning Model Management System for Real-Time Analysis. The Journal of Korean Institute of Communications and Information Sciences, 46, 2, (2021), 390-400. DOI: 10.7840/kics.2021.46.2.390.

[KICS Style]

Mo-se Lee, Min-su Kang, In-ho Kim, Jae-hun Kim, "Design and Performance Comparison of Docker Container Based Deep Learning Model Management System for Real-Time Analysis," The Journal of Korean Institute of Communications and Information Sciences, vol. 46, no. 2, pp. 390-400, 2. 2021. (https://doi.org/10.7840/kics.2021.46.2.390)