This is an automated email from the ASF dual-hosted git repository.
robertlazarski pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/axis-axis2-java-core.git
The following commit(s) were added to refs/heads/master by this push:
new cf132dee66 Use L64X128MixRandom in the financial benchmark Monte Carlo
sample
cf132dee66 is described below
commit cf132dee663e4a0714deb053cfab720a18ce320f
Author: Robert Lazarski <[email protected]>
AuthorDate: Tue Oct 6 13:53:56 2026 -1000
Use L64X128MixRandom in the financial benchmark Monte Carlo sample
monteCarlo() drew its normals from java.util.Random.nextGaussian(), whose
atomic seed update and StrictMath polar method dominated the inner loop.
The JDK 17+ L64X128MixRandom makes the method about 3x faster on JDK 21;
seeded runs remain reproducible, but produce different paths than before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---
.../webservices/FinancialBenchmarkService.java | 34 ++++++++++++++--------
.../springboot/webservices/MonteCarloRequest.java | 4 +--
2 files changed, 24 insertions(+), 14 deletions(-)
diff --git
a/modules/samples/userguide/src/userguide/springbootdemo-tomcat11/src/main/java/userguide/springboot/webservices/FinancialBenchmarkService.java
b/modules/samples/userguide/src/userguide/springbootdemo-tomcat11/src/main/java/userguide/springboot/webservices/FinancialBenchmarkService.java
index 1215bd5d8c..c1be591caa 100644
---
a/modules/samples/userguide/src/userguide/springbootdemo-tomcat11/src/main/java/userguide/springboot/webservices/FinancialBenchmarkService.java
+++
b/modules/samples/userguide/src/userguide/springbootdemo-tomcat11/src/main/java/userguide/springboot/webservices/FinancialBenchmarkService.java
@@ -28,8 +28,9 @@ import java.util.Arrays;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
-import java.util.Random;
import java.util.UUID;
+import java.util.random.RandomGenerator;
+import java.util.random.RandomGeneratorFactory;
/**
* Java reference implementation of the Axis2/C Financial Benchmark Service.
@@ -302,17 +303,25 @@ public class FinancialBenchmarkService {
* and should be preserved verbatim in any re-implementation.
*
* <p>Sampling behavior:
- * Uses {@link Random#nextGaussian()} (polar method) for normal variates.
- * When {@code randomSeed != 0}, a seeded {@link Random} is used for
- * reproducibility (same seed → bit-identical output). When
+ * Uses the JDK 17+ {@code L64X128MixRandom} generator
+ * ({@link RandomGenerator#nextGaussian()}, modified ziggurat) for
+ * normal variates. When {@code randomSeed != 0}, a seeded instance is
+ * used for reproducibility (same seed → bit-identical output). When
* {@code randomSeed == 0}, a fresh unseeded instance gives
* non-deterministic results.
*
- * Warning for cross-implementation reproducibility: {@code
java.util.Random}
- * uses a linear congruential generator (LCG). An implementation in a
- * different language or using a different PRNG (e.g., xorshift128+,
- * PCG64, NumPy's default) will produce DIFFERENT numbers for the
- * SAME seed. Reproducibility is per-PRNG, not cross-PRNG.
+ * Do not switch back to {@code java.util.Random}: its
+ * {@code nextGaussian()} updates an atomic seed on every draw and uses
+ * the {@code StrictMath} polar method, and with ~252 draws per path it
+ * dominates the loop. Measured on JDK 21, swapping it for
+ * {@code L64X128MixRandom} made this method about 3x faster on both an
+ * Intel Xeon and an AMD EPYC server, with VaR unchanged within Monte
+ * Carlo error.
+ *
+ * Warning for cross-implementation reproducibility: an implementation
+ * in a different language or using a different PRNG (e.g.,
+ * xorshift128+, PCG64, NumPy's default) will produce DIFFERENT numbers
+ * for the SAME seed. Reproducibility is per-PRNG, not cross-PRNG.
*
* <p>Numerical edge cases:
* - <b>Guarded in the body below:</b> the variance accumulator
@@ -401,9 +410,10 @@ public class FinancialBenchmarkService {
double volSqrtDt = sigma * Math.sqrt(dt);
// ── PRNG: seeded for reproducibility, unseeded for production
─────────
- Random rng = request.getRandomSeed() != 0
- ? new Random(request.getRandomSeed())
- : new Random();
+ // L64X128MixRandom, not java.util.Random — see "Sampling behavior"
above.
+ RandomGenerator rng = request.getRandomSeed() != 0
+ ?
RandomGeneratorFactory.of("L64X128MixRandom").create(request.getRandomSeed())
+ : RandomGenerator.of("L64X128MixRandom");
logger.info(logPrefix + "starting " + nSims + " sims × " + nPeriods +
" periods" +
" (seed=" + request.getRandomSeed() + ", npy=" + npy + ")");
diff --git
a/modules/samples/userguide/src/userguide/springbootdemo-tomcat11/src/main/java/userguide/springboot/webservices/MonteCarloRequest.java
b/modules/samples/userguide/src/userguide/springbootdemo-tomcat11/src/main/java/userguide/springboot/webservices/MonteCarloRequest.java
index 830d69fdf4..0dbeb0bf98 100644
---
a/modules/samples/userguide/src/userguide/springbootdemo-tomcat11/src/main/java/userguide/springboot/webservices/MonteCarloRequest.java
+++
b/modules/samples/userguide/src/userguide/springbootdemo-tomcat11/src/main/java/userguide/springboot/webservices/MonteCarloRequest.java
@@ -79,7 +79,7 @@ package userguide.springboot.webservices;
*
* <h3>Reproducibility</h3>
* <p>Setting a non-zero {@code randomSeed} makes the run deterministic
- * AGAINST THIS IMPLEMENTATION (Java {@link java.util.Random}). The same
+ * AGAINST THIS IMPLEMENTATION (the JDK's {@code L64X128MixRandom}). The same
* seed will not produce the same output under other PRNGs — xorshift128+,
* PCG64, Mersenne Twister, NumPy's Generator — so seeded reproducibility
* is per-backend, not cross-backend. Callers comparing VaR numbers
@@ -162,7 +162,7 @@ public class MonteCarloRequest {
/**
* Random seed for reproducibility. 0 (default) → non-deterministic
* (each call yields different results). A non-zero value seeds
- * {@link java.util.Random} so repeated calls with the same seed give
+ * the JDK's {@code L64X128MixRandom} so repeated calls with the same seed
give
* bit-identical output.
*
* <p>Cross-PRNG caveat: the same seed in a different language or PRNG
