Startup Latency Analysis in Java Frameworks for Serverless AWS Lambda Deployments.

Cold start latency in serverless computing, particularly in Java-based AWS Lambda functions, presents a significant challenge for latency-sensitive applications. This study investigates the performance characteristics of three modern Java frameworks - Spring Boot, Micronaut, and Quarkus - deployed on AWS Lambda using the ARM64 (Graviton2) architecture. It evaluates cold start latency across three deployment configurations: managed runtime (with and without SnapStart) and GraalVM native images. Metrics were collected at varying memory allocations using Java 21. Results show that Quarkus consistently outperforms others in cold start latency on standard JVM, while SnapStart and GraalVM significantly reduce the number of cold starts and achieve sub-second latency, respectively. We discuss the implications of these findings for choosing a Java framework and runtime strategy on AWS Lambda, considering the trade-offs in deployment time, complexity, and performance. The paper concludes with recommendations for leveraging SnapStart and native images to mitigate cold start issues in Java serverless applications on ARM64.