An Experimental Evaluation of a Hybrid SQL–NoSQL Architecture for Scalable E-Commerce Analytics

Abstract

The rapid growth of data volume, structural heterogeneity, and real-time processing demands in modern e-commerce platforms challenges the scalability and flexibility of single-model database systems. Hybrid database architectures, which integrate multiple storage paradigms within a coordinated framework, offer a promising alternative; however, controlled empirical evaluation across heterogeneous models remains limited. This study presents a structured experimental assessment of a hybrid SQL–NoSQL architecture integrating relational, document-oriented, in-memory, column-family, graph, and streaming components within a unified deployment environment. The architecture incorporates PostgreSQL, MongoDB, Redis, Cassandra, and Neo4j, coordinated through Kafka-based streaming with Change Data Capture (CDC) synchronization using Debezium. Performance evaluation focused on reproducible benchmarking of representative subsystems under standardized hardware constraints. Key performance indicators—including throughput, response latency, and scalability under increasing concurrency—were measured using established workload simulation tools. Results demonstrate clear workload-dependent specialization: relational processing maintained stable transactional behavior, document-oriented storage supported higher ingestion throughput, and in-memory caching achieved minimal access latency. The findings indicate that distributing workload categories across specialized storage engines improves operational balance compared with monolithic deployments. Beyond quantitative benchmarking, the study provides a reproducible architectural framework illustrating how coordinated multi-model systems can support scalable and adaptable data infrastructures for real-time e-commerce analytics.