Repository navigation
JMOA 2.1: a measured PetClinic RAM win
JMOA 2.1 is the first release backed by a direct, held-out comparison against a documented strict no-JMOA Spring Boot deployment. It combines build-time JVM transformation and packaging with an evidence-gated workflow: semantic equivalence, Linux process and cgroup attribution, same-artifact controls, blinded reduction, and immutable publication records.
| Metric | Accepted JMOA R41F versus strict no-JMOA B0E |
|---|---|
| Process PSS | −15,241.5 KiB (−14.88 MiB) |
memory.current |
−17,033,216 bytes (−16.24 MiB) |
| Private Dirty | −15,252 KiB |
| Held-out direction | 12/12 favorable blocks |
| Exact paired sign test | p = 0.00048828125 |
| PSS bootstrap 95% interval | [−16,109.5, −15,052.5] KiB |
The campaign completed 81/81 sessions: 5 qualification, 20 same-artifact controls, 8 screening, and 48 held-out sessions. All frozen RAM, semantic, lifecycle, and product-cost gates passed.
The engineering tradeoff is disclosed with the win: median lifecycle CPU increased 14.71%, median startup increased 1.652 s, and median request/p95 latency changes were both 0 ms.
This is a deliberately narrow claim: the exact accepted R41F JMOA fat-JAR deployment versus the documented strict no-JMOA B0E exploded-Boot deployment on PetClinic. It is packaging-inclusive, not a universal Spring Boot or content-only claim.
Read the engineering record
- JMOA 2.1 technical paper
- Human-readable PetClinic result
- Machine-readable result
- Independent portfolio case study
Release contents
- JMOA Maven plugin 2.1.0, runtime library 2.1.0, sources, and Maven POMs
- SHA-256 checksum list and commit-bound release manifest
- Technical paper plus human- and machine-readable result artifacts
Every uploaded asset was verified after publication against its local SHA-256 digest. The release tag resolves to commit 54c65dcb86102f954b3076998e976c79762803be.