ABSTRACT: Serverless computing represents a paradigm shift in cloud architecture by abstracting infrastructure management and allowing developers to focus solely on application logic. In this model, cloud providers dynamically allocate resources and charge only for the actual execution time, combining Function-as-a-Service (FaaS) and managed Backend-as-aService (BaaS) components. This research explores how serverless architectures impact performance and cost in enterprise-level applications, where scalability, reliability, and predictable expenses are critical. The study analyzes major performance factors such as cold starts, tail latency, and concurrency limits, along with cost dimensions including payper-execution billing, provisioned concurrency overhead, and data egress charges. Through literature review and controlled experimentation, we evaluate how workload characteristics—steady versus burst traffic—affect both latency and cost efficiency compared to container-based deployments. Results indicate that serverless architectures offer significant benefits for irregular or event-driven workloads due to their elasticity and operational simplicity. However, for sustained high-throughput or latency-sensitive enterprise systems, performance unpredictability and hidden costs can outweigh these advantages. The paper concludes by proposing a decision framework to help enterprises identify suitable workloads for serverless adoption and mitigate cost-performance tradeoffs through hybrid deployment strategies. The findings contribute to a clearer understanding of when serverless computing delivers genuine value and when traditional approaches remain more effective “This study contributes a reproducible evaluation framework and comparative analysis that can guide enterprise architects in cost-performance optimization.”
KEYWORDS: Serverless Computing; FaaS; Cold Start; Performance; Cost Optimization; Enterprise Applications.
SERVERLESS ARCHITECTURES IN CLOUD COMPUTING: PERFORMANCE AND COST TRADE-OFFS IN ENTERPRISE APPLICATIONS
MR. PIYUSH M SHINDE
MASTERS IN COMPUTER APPLICATIONS IDOL, MUMBAI UNIVERSITY