Aswathnarayan Muthukrishnan Kirubakaran

Cloud-native architectures deployed on Amazon Web Services (AWS) increasingly rely on microservices and Internet of Things (IoT) platforms to support scalable, real-time, and data-intensive applications. While these paradigms improve agility and elasticity, they introduce significant challenges in secure data integration, governance, and regulatory compliance due to highly distributed data flows and expanded attack surfaces. This paper presents a secure cloud-native data integration architecture on AWS that unifies microservices-based application workloads, IoT data ingestion pipelines, and database-level protection using Oracle 19c Transparent Data Encryption (TDE). The proposed architecture maps established design patterns such as service meshes, edge and fog computing, and streaming-based ETL to AWS-native services, enabling encrypted data exchange in transit and robust protection of data at rest without application modifications. A prototype implementation using Amazon EKS, AWS IoT Core, AWS Glue, and Oracle 19c demonstrates scalable ingestion, low-latency processing, and minimal performance overhead from encryption. Experimental results validate that the architecture achieves improved scalability and tail-latency characteristics while maintaining strong data-at-rest security and compliance readiness. The paper further identifies research gaps in AI-driven automation, unified observability, and adaptive security frameworks for future cloud-native and multi-cloud data integration environments.

Replication is fundamental to achieving availability and durability in cloud storage systems, yet the operational characteristics of failover under controlled disruptions remain insufficiently quantified in reproducible settings. This paper presents an empirical evaluation of replication and failover behavior in Amazon Web Services (AWS) using systematic synthetic failure injection. We measure Recovery Time Objective (RTO), Recovery Point Objective (RPO), tail recovery latency (P95/P99), replication lag distributions, latency amplification, and throughput degradation under zone-level and region-level failure scenarios. Results demonstrate that synchronous multi-availability-zone replication achieves low and stable RTO with negligible RPO, whereas asynchronous cross-region replication introduces measurable lag and significantly amplified tail recovery. By decomposing failover into detection, control-plane redirection, and client stabilization phases, the study provides a structured methodology for analyzing recovery dynamics and offers evidence-based guidance for replication strategy selection in stateful cloud workloads.

Public cloud platforms provide multiple storage abstractions, including object storage, block storage, and managed file systems, each designed to balance scalability, durability, and operational flexibility. Despite their widespread adoption, selecting an appropriate storage model for data-intensive workloads remains a non-trivial architectural decision due to differences in latency, throughput, and I/O behavior. This paper presents a systematic empirical evaluation of these storage abstractions under controlled experimental conditions to provide evidence based guidance for workload-driven storage selection. We benchmark sequential, random, and mixed I/O workloads, measuring sustained throughput, input/output operations per second (IOPS), latency characteristics, and performance stability. The results reveal distinct performance profiles across storage models. Block storage consistently delivers high IOPS and low latency, making it suitable for transactional and latency-sensitive workloads. Object storage demonstrates strong performance for large sequential transfers and bulk data movement. Managed file storage provides shared access semantics with moderate performance trade-offs. Rather than advocating a universally superior storage solution, the findings emphasize aligning storage selection with dominant workload characteristics. By quantifying observable performance differences across abstractions, this study provides practical architectural insights for analytics systems, distributed applications, and cloud-native data platforms.

End-to-end autonomous steering models map raw sensory observations directly to continuous control actions, reducing the hand-engineering burden of modular autonomy stacks. However, two practical barriers limit deployability: (i) steering prediction is inherently temporal and frame-wise convolutional neural networks (CNNs) often exhibit oscillations and degraded recovery under covariate shift, and (ii) centralized training requires aggregating large-scale driving data that are distributed, bandwidth-intensive, and privacy-sensitive. This paper presents a federated temporal autonomous steering framework combining CNN-based spatial perception with Long Short-Term Memory (LSTM) sequence modeling under Federated Learning. Multiple clients (simulated vehicles) train locally on heterogeneous, non-identically distributed (non-IID) driving data and periodically synchronize with a server via Federated Averaging. We formalize the temporal imitation learning objective, present a sequence-aware model architecture and preprocessing pipeline, and define a federated optimization protocol with communication and non-IID considerations. Extensive experiments compare centralized CNN, centralized CNN-LSTM, and federated CNN-LSTM baselines under steering imbalance and distribution shifts induced by curvature bias and lighting perturbations. Results show that temporal modeling reduces steering oscillation variance and improves closed-loop stability, while federated training achieves comparable accuracy to centralized training with privacy-preserving data locality. We also analyze convergence behavior under heterogeneous clients, communication cost per round, and ablations on sequence length, local epochs, and client sampling. The proposed framework provides a reproducible foundation for distributed, privacy-preserving end-to-end control.