Scalable Hybrid Deep Learning–Driven Container Scheduling Framework  for Reliable Cloud Computing

Authors: Dr. P. Jaya Prakash1, Anipakula Suma2, Gampa Sathish Kumar3, Embeti Koushik4, KuramHemanth Babu5
1Associate Professor, Department of Information Technology, Sri Venkateswara college of Engineering, India
2Department of Information Technology, Sri Venkateswara college of Engineering, India
3Department of Information Technology, Sri Venkateswara college of Engineering, India
4Department of Information Technology, Sri Venkateswara college of Engineering, India
5Department of Information Technology, Sri Venkateswara college of Engineering, India
Emails:1pokalajayaprakash@gmail.com,2sumanaidu018@gmail.com,3sathishraj9346@gmail.com,4
embetikoushik@gmail.com,5kuramhemanth57@gmail.com
Corresponding Author/guide: Dr. P. Jaya Prakash, M. Tech, Ph.D, Associate Professor, Dept of InformationTechnology.

Abstract-Efficient container scheduling in cloudcomputing dynamically allocates CPU and memory resources to containers based on predicted workloads, maximizing resource utilization while avoiding node overload and underutilization. Existing container scheduling techniques select nodes based on initial resource allocations and user requirements, but often suffer inefficiencies due to over- or under-allocation of resources, leading to wasted capacity or service disruptions. These systems use AI models to predict workloads but struggle with irregular noise in load patterns and complex model structures that limit prediction accuracy and scheduling efficiency.The proposed future system uses the DeHyFo hybrid deep learning model to accurately predict future CPU and memory usage by decomposing workloads into linear and irregular components through multiple linear regression and the LightTS model. It reduces resource waste and node overload by scoring predictions with an efficient resource utilization function (SERU) for optimal scheduling, improving resource efficiency andsignificantly lowering node overload incidents. This system enhances service reliability by dynamically adapting to workload changes using historical container data and integrates seamlessly with Kubernetes for efficient deployment and resource management, effectively handling diverse and irregular workloads while providing better service quality and cost savings compared to existing methods .Keywords:Efficientcontainer scheduling, Kubernetes, DeHyFo hybrid deep learning model,LightTS model, Linear Regression, cloud computing