Escalabilidade em sistemas web: uma análise comparativa entre Java e Rust em arquitetura monolítica

Abstract

The expansion of web applications has intensified the pursuit of scalability, and monolithic architecture remains the starting point for many systems. As load increases, migration toward distributed architectures is often assumed to be inevitable, overlooking an earlier variable: the execution ecosystem of the language, encompassing its memory management model, framework, and concurrency model. This work compares two functionally equivalent implementations of a web address shortening service, one in Java with Spring Boot and embedded Tomcat server, and another in Rust with Actix-web on the asynchronous Tokio runtime, subjected to progressive loads of up to 1,500 requests per second in containers limited to half a processing core and 512 megabytes of memory. Three garbage collection algorithms of the Java virtual machine were evaluated as an independent variable. Results show that the Rust implementation maintained a stable median latency between 260 and 272 microseconds with a fixed footprint of 7 megabytes, while the default Java configuration exhibited 95th-percentile latency peaks of up to 196 milliseconds and approximately 420 megabytes of memory consumption. The latest-generation concurrent collector, designed for minimal latency, failed the acceptance thresholds in all three runs due to contention over the fractional processing capacity. The sequential collector, the simplest available, outperformed the other Java configurations in latency, variance, and memory consumption. Within the investigated scenario, the choice of execution ecosystem and collection algorithm proved to be a determining factor in the performance limits of the monolith before architectural migration.