Year: 2024
Type: article
Abstract: Machine Learning Operations (MLOps) are designed to accelerate the development of high-quality machine learning (ML) models by reducing the deployment cycle and improving overall efficiency. Despite its promise, the concept of MLOps remains underexplored, with unclear implications for research and practical application. Current research has primarily focused on developing individual ML models, overlooking the complexities of deploying and managing integrated ML systems in real- world scenarios. A comprehensive understanding of system interactions is crucial, particularly when using multi-container services, which necessitate robust and effective monitoring solutions. In response, this study proposes a novel model, the Multi-Containers Monitoring Model, which leverages ML techniques such as Bidirectional Long Short-Term Memory (Bi- LSTM) and State-Action-Reward-State-Action (SARSA) to address these challenges. The proposed model enables the effective scaling and monitoring of MLOps systems by interpreting and managing interactions between containers. It also expands software deployment capabilities across various settings, enhancing software release performance. The results demonstrate that Multi-Containers Monitoring Model improves deployment cycles by up to 24.55%, reduces build length cycles by up to 13%, and decreases response time by up to 50.03%. This study offers a significant advancement in utilizing MLOps for real-world ML system monitoring and deployment.
Author: Zeinab Shoieb Elgamal
Institution: Helwan University
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Keywords: MLOps, container monitoring, BiLSTM, SARSA
Open Access status: diamond
OpenAlex ID: W7217793384
