IMAP-GCG: Edge Container Resource Scheduling and Configuration Method Based on Improved MAPPO and GCN-GRU

Container technology has been widely used in industrial development, including system deployment and program management, and has achieved very good results. However, in the current edge network environment, there are delay problems caused by container cold start and user terminal task changes, which need to be solved by designing a flexible and efficient edge container resource configuration method. In response to this problem, this paper proposes a method for dynamic container resource allocation called Improved Multi-Agent PPO with GCN-GRU (IMAP-GCG). This method mainly implements the prediction of terminal workload, container resource pre-configuration and container resource scheduling methods. First, we propose a timing prediction model with GCN and GRU as the encoder and LSTM as the decoder to predict the temporal periodicity of user requests, so as to deploy containers in advance. Then, a MAPPO algorithm based on trusted region policy optimization is proposed to schedule container resources, which improves the performance of container resource scheduling. Experiments on two workloads show that the IMAP-GCG model proposed in this paper is superior to the existing advanced multi-agent reinforcement learning model in terms of CPU utilization, bandwidth utilization, and task average delay. The task type forecasting accuracy on this indicator is due to the existing advanced time series forecasting models.