Implementation of Continuous Integration and Continuous Delivery (CI/CD) on Deep Learning Models with Google Cloud Platform Case Study of Automatic Problem Generation

Abstract

Ineffective deployment processes can negatively impact application releases to users, while unstable systems can undermine user experience. This research aims to implement Docker and Kubernetes Pipelines through the DevOps MLOps approach to enhance deployment process effectiveness and analyze the performance of Docker and Kubernetes as alternative environments for deploying the AQG web server. The research methodology follows Research and Development with Development and Operations procedures encompassing planning, coding & building, testing, releasing, deploying & operating, and monitoring. The study focuses on deploy time and web server performance using Docker and Kubernetes Pipelines, assessed through load testing analysis. The subject of this research is the performance outcomes of the web server utilizing Docker and Kubernetes Pipelines. The research outcomes include development of a web server for the Automatic Question Generator service using Docker and Kubernetes Pipelines. Load testing evaluations demonstrate that the web server with Docker and Kubernetes Pipelines exhibits greater stability in error rate, throughput, and response time as threads increase from 5,000 to 10,000. Moreover, the total deploy time decreased from 9 minutes 1 second to 3 minutes 40 seconds, indicating a 245% increase in efficiency