Work

Serverless Auto-Scaling E-Commerce Order Processing on AWS: Design and Simulation-Based Evaluation Under Flash-Sale Traffic

Year: 2026
Type: preprint
Abstract: Online retailers face traffic that can jump by more than an order of magnitude within seconds during flash sales, and capacity planned in advance either fails under that surge or sits idle the rest of the time. This paper presents the design of an event-driven order-processing back end built entirely from managed AWS services—API Gateway, Lambda, SQS, DynamoDB, and SNS—and evaluates it against two conventional alternatives, a fixed pool of virtual-machine workers and a threshold-based VM auto-scaler. Because the evaluation must be repeatable without incurring production-scale cloud cost, we use a discrete-event simulation with explicitly stated, recalibratable parameters and report means with 95% confidence intervals over ten independent replications. For a three-minute surge to 400 requests/s, the serverless design completed 100.0% of requests with a 95th-percentile latency of 0.25 s, whereas the fixed pool completed 70.2% and the VM auto-scaler 82.5%. A second experiment shows that placing a queue between order acceptance and order completion eliminated the 22.2% order loss seen when a capacity-limited downstream stage was called synchronously, at the price of a confirmation delay that peaked near one minute. We discuss the limits of the model and outline the measurements needed to validate it on a live deployment.
Language: English
Open Access status: green
OpenAlex ID: W7236696895
Cites:
Cited by:
Related to: 10