Efficient Big Data Processing in Financial Sector Applications: The Evaluation of Distributed Computing Architectures

Abstract—Financial institutions are increasingly challenged by the influx of high-volume, high- velocity, and heterogeneous data streams, including transaction records and real-time market feeds. Conventional ETL pipelines and monolithic data warehouse systems fall short of delivering the low- latency responses, scalable throughput, and precise processing guarantees required for critical operations such as fraud detection, algorithmic trading, and real-time risk management. This paper presents a detailed examination of distributed computing paradigms—including batch, micro-batch, and streaming—as well as architectural patterns such as Lambda, Kappa, and hybrid frameworks, specifically adapted for financial applications. We introduce a containerized hybrid Lambda-Kappa model deployed on Kubernetes[12], integrating Apache Kafka [6] for event ingestion, Apache Flink [5] (augmented with GPU powered processing) for real-time processing, and Apache Spark [2][13] for batch computation. Our 60-node prototype achieves 1.2 million events per second with p99 latency under 0.7 seconds and demonstrates nearly linear scalability (R2 = 0.99), reducing operational costs by approximately 25%. The paper also discusses system resilience, compliance and security considerations, and outlines future research directions in serverless orchestration [13], adaptive autoscaling, and privacy-aware analytics. Keywords—Big data, distributed computing, financial analytics, real-time streaming, Lambda architecture, Kubernetes, GPU acceleration.