Optimized Resource Allocation in Cloud Computing using Rotation Invariant Coordinate Convolutional Neural Network (RIC-CNN)

Abstract

This paper presents the design and implementation of a Rotation-Invariant Coordinate Convolutional Neural Network (RIC-CNN)–based framework for optimized resource allocation in cloud computing environments. The proposed solution leverages workload data from the Bitbrains dataset and transforms it into image-like tensor representations to enable intelligent scheduling decisions. The model focuses on addressing critical challenges in cloud computing such as high energy consumption, SLA (Service-Level Agreement) violations, and inefficient resource utilization. By combining coordinate convolution and rotation-invariant convolution layers, the RIC-CNN captures spatial dependencies and input geometry more robustly than traditional CNNs. Performance evaluation using 1500 time-series workload traces demonstrated that the proposed approach achieved up to 56.5% reduction in average response time, 14.1% fewer SLA violations, and 1.5% improvement in energy efficiency when compared to baseline scheduling algorithms like Round Robin and Least Loaded. Radar plots and box plots further visualized the superior performance of RIC-CNN across multiple cloud performance metrics. These results validate the effectiveness of deep learning models in dynamic, real-time cloud resource management, making this approach suitable for scalable, multi-objective optimization in edge and cloud environments.

• The study proposes a novel RIC-CNN model for multi-objective cloud resource scheduling, leveraging spatial features from real-time workload tensors.


• Using the Bitbrains dataset, the approach demonstrated up to 56.5% improvement in response time, 14.1% reduction in SLA violations, and energy savings of 1.5% compared to traditional schedulers.


• The model's architecture with CoordConv and rotation-invariant layers enabled superior generalization across rotated input tensors, validating its robustness and scalability.