Edge computing resource scheduling method based on container elastic scaling.

Area of Science:

• Computer Science
• Distributed Systems
• Artificial Intelligence

Background:

• Edge computing requires efficient resource and bandwidth management for real-time data processing.
• Traditional container scaling strategies suffer from slow response times, poor resource utilization, and unpredictable loads.
• Containerization is fundamental to edge computing due to its performance benefits.

Purpose of the Study:

• To propose a novel edge computing resource scheduling method based on elastic container scaling.
• To address limitations of traditional container scaling, including response time and resource utilization.
• To adapt edge computing to dynamic application load patterns and traffic surges.

Main Methods:

• Developed a container load prediction model, Trend Enhanced-Temporal Convolutional Network (TE-TCN), using an encoder-decoder structure with dual-input ResNet.
• Modeled container elastic scaling as a multi-objective optimization problem using Markov Decision Process (MDP).
• Designed a reinforcement learning-based predictive container scaling strategy utilizing TE-TCN predictions and a time-varying action space.

Main Results:

• The TE-TCN model demonstrated accurate container load change prediction on benchmark and real-world datasets.
• The proposed strategy reduced average response time by 16.2% during burst loads.
• Average CPU utilization increased by 44.6% during jitter loads.

Conclusions:

• The proposed TE-TCN model and reinforcement learning-based scaling strategy effectively optimize edge computing resource management.
• The approach enhances system responsiveness and resource efficiency in dynamic edge environments.
• This method provides a robust solution for handling unpredictable traffic and load variations in edge computing.