Updated benchmark numbers in premain README

ioi.lam at oracle.com ioi.lam at oracle.com
Tue Sep 30 19:48:08 UTC 2025


I've updated the benchmark numbers using the latest premain tip (as of 
2025/09/30):

https://github.com/openjdk/leyden/blob/premain/README.md#premain-aot-cache-summary

There's quite a bit of improvement due to the new compiler tuning 
changes by Aleksey. I also added a "2 cores only" mode to 
emulate microservices that are allocated very few CPUs.


Here are the results of {premain AOT} vs {JDK 25, no CDS, no AOT}. Note 
that premain is generally much more beneficial with the 2 Cores Only 
case, except for PetClinic:


Benchmark        Desktop/Server Class (28 Cores)   2 Cores Only
*----------------------------------------------------------------------------
*Helidon Quick Start                3.59x 4.11x  ++
JavacBenchApp 50 source files      2.21x                3.17x ++++
Micronaut First App Demo           2.91x                4.90x +++++++
Quarkus Getting Started Demo       2.97x                3.74x +++
Spring-boot Getting Started Demo   4.13x                4.70x ++
Spring PetClinic Demo              3.33x                3.03x -- ???



Changes from last time ( 
https://github.com/openjdk/leyden/blob/46e03d000efd1b2784ad4dcd4c83310ace498ec0/README.md 
)

The 2 Core Only case see a lot of improvements with the compiler tuning. 
PetClinic is again an outlier.

Benchmark           Desktop/Server Class (28 Cores)   2 Cores Only
----------------------------------------------------------------------------
Helidon Quick Start                3.60 -> 3.59x  ~      3.53 -> 4.11x  ++
JavacBenchApp 50 source files      2.17 -> 2.21x  ~      2.74 -> 3.17x  ++
Micronaut First App Demo           2.85 -> 2.91x  +      4.39 -> 4.90x  ++
Quarkus Getting Started Demo       2.73 -> 2.97x  ++     2.95 -> 3.74x  +++
Spring-boot Getting Started Demo   3.96 -> 4.13x  ++     3.70 -> 4.70x  +++
Spring PetClinic Demo              3.24 -> 3.33x  +      3.04 -> 3.03x  ~

(The +, -, ~ are calculated by my eyeballs so not very scientific :-)


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/leyden-dev/attachments/20250930/740af565/attachment.htm>


More information about the leyden-dev mailing list