Files

46 lines
2.5 KiB
Plaintext

framework-benchmarks — consolidated results
Sandbox: 2 vCPU / 7.8GiB RAM Linux container, GraalVM CE 25.4.4.1.1 (JDK 25.0.4.1.1)
Versions under test: Spring Boot 4.1.1, Quarkus 3.40.1 (LTS), Micronaut 5.2.2
All numbers are the average of 3 kept runs after discarding one cold-cache run (startup/memory),
or one 15-second run at concurrency 20 (throughput). Raw per-run JSON in 01-04 in this directory.
STARTUP TIME (seconds, cold process to first 200 response)
framework JVM native native is Nx faster
spring-boot 6.318 0.132 47.9x
quarkus 2.185 0.035 62.4x
micronaut 2.507 0.041 61.1x
RESIDENT MEMORY (MB, RSS right after first successful request)
framework JVM native native is Nx smaller
spring-boot 229.2 79.5 2.9x
quarkus 171.6 33.6 5.1x
micronaut 180.0 46.2 3.9x
THROUGHPUT (requests/sec, GET /widgets, 20 concurrent stdlib-http.client workers, 15s)
framework JVM req/s native req/s
spring-boot 2275.7 4795.1
quarkus 3791.4 4243.8
micronaut 4276.2 4961.2
DEPLOYMENT ARTIFACT SIZE (MB)
framework JVM (fat jar / fast-jar dir) native executable
spring-boot 23 73
quarkus 21 44
micronaut 14 54
NATIVE BUILD TIME (wall clock, 2 vCPU, this sandbox — native-image is CPU-bound and
parallelizes with available cores, so a bigger CI runner will be meaningfully faster)
quarkus: 6m22s (succeeded on the first attempt)
micronaut: 7m15s (succeeded on the 4th attempt — 3 earlier attempts failed in
~42-50s each over a Netty/logback static-initialization clash;
see docs/output/06-micronaut-native-build-failures.txt)
spring-boot: 8m25s (succeeded on the first attempt, after routing around the
native-maven-plugin's execution binding — see repo README)
Cross-checks against the independent 2025-10 benchmark by gillius.org (different
hardware: Apple M2 Pro laptop vs this shared 2-vCPU container; different app: a
Postgres-backed CRUD app vs this repo's deliberately DB-free one; Spring Boot 3.5.6/
Quarkus 3.28.1/Micronaut 4.5.4 vs the current GA versions here) — see post for the full
comparison table. Qualitatively consistent: native beats JVM mode by roughly one to two
orders of magnitude on both startup and memory, in both benchmarks, on both machines.