How much faster is Java 8?

by Geoffrey De Smet - March 19, 2014
Optimization

Java SE 8 was released yesterday.
Traditionally, every new major JRE version comes with a free performance boost.
Do we get another free lunch? And how big is the gain this time?

Let’s benchmark it.

Benchmark methodology

Run the same code with 3 different JRE versions (SunJDK 1.6.0_26, OpenJDK 1.7.0_51 and OpenJDK 1.8.0).
The code itself was written for Java 6 (both in syntax and JDK API’s usage) and compiled for Java 6 with OpenJDK 1.7.
Each run takes about 55 minutes.
VM arguments: -Xmx1536M -server
Software: Linux 3.2.0-59-generic-pae
Hardware: Intel® Xeon® CPU W3550 @ 3.07GHz
Each run solves 13 planning problems with OptaPlanner.
Each planning problem runs for 5 minutes. It starts with a 30 second JVM warm up which is discarded.
Solving a planning problem involves no IO (except a few milliseconds during startup to load the input).
A single CPU is completely saturated.
It constantly creates many short lived objects, and the GC collects them afterwards.
The benchmarks measure the number of scores that can be calculated per millisecond. Higher is better.
Calculating a score for a proposed planning solution is non-trivial:
it involves many calculations, including checking for conflicts between every entity and every other entity.

To reproduce these benchmarks locally, build optaplanner from source
and run the main class
GeneralOptaPlannerBenchmarkApp.

Benchmark results

Executive summary

My observations:

On the biggest dataset (Machine Reassignment B10), which dwarfs any of the other datasets in size,
Java 8 is 20% faster than Java 7, which was already 17% faster than Java 6.
In some cases, Java 8 is slower than Java 7.
Specially for the course scheduling datasets, Java 8 is 6% slower than Java 7.
Hopefully new releases of Java 8 will resolve this performance regression soon.
On average, Java 8 is only 1% faster than Java 7. This while Java 7 is already 16% faster than Java 6.
Despite that this is the first final release of OpenJDK 8, I did not find any regressions in Java 8.
OptaPlanner’s examples are 100% reproducible, so as expected, the different JRE’s give the exact same results at every single iteration.

Raw benchmark numbers

Conclusion

On the big datasets, Java 8 is clearly faster. And this without changing a line of code.
On average, the result is less convincing (with the current release), but a free lunch is always welcome.

Update (18-JUN-2015): Java 8 build 0 versus build 45

I wrote this blog more than a year ago. Meanwhile both Java 8 and OptaPlanner have had a number of releases.
How much do the Java 8 updates affect performance?

To make a fair comparison, I ‘ve (re)ran the latest OptaPlanner code on both the original Java 8 release (build 0) and the latest release (build 45).
These are the results:

JDK	Cloud balancing 200c	Cloud balancing 800c	Machine reassignment B1	Machine reassignment B10	Course scheduling c7	Course scheduling c8	Exam scheduling s2	Exam scheduling s3	Nurse rostering m1	Nurse rostering mh1	Sport scheduling nl14
SunJDK 1.6.0_26	46462	44963	94567	28655	5473	6989	6954	3785	3232	2948	1977
OpenJDK 1.7.0_51	53683	49798	116553	33733	6182	7848	8243	4606	3645	3377	2445
OpenJDK 1.8.0	54687	51625	135102	40529	5798	7357	8048	4441	3637	3324	2321
6 ⇒ 7	15.54%	10.75%	23.25%	17.72%	12.95%	12.29%	18.54%	21.69%	12.78%	14.55%	23.67%
7 ⇒ 8	1.87%	3.67%	15.91%	20.15%	-6.21%	-6.26%	-2.37%	-3.58%	-0.22%	-1.57%	-5.07%
Dataset scale	120k	1920k	500k	250000k	217k	145k	1705k	1613k	18k	12k	4k

On average, Java 8 build 45 is 2.07% faster than Java 8 build 0. That’s welcome!

JDK	Cloud balancing 200c	Cloud balancing 800c	Machine reassignment B1	Machine reassignment B10	Course scheduling c7	Course scheduling c8	Exam scheduling s2	Exam scheduling s3	Nurse rostering m1	Nurse rostering mh1	Sport scheduling nl14
OpenJDK 1.8.0	63404	55295	126743	38260	6072	7825	10518	7176	3872	3558	1252
OpenJDK 1.8.0_45	64715	56472	131941	36175	6499	8084	10406	7217	3905	3595	1354
1.8.0 ⇒ 1.8.0_45	2.07%	2.13%	4.10%	-5.45%	7.03%	3.31%	-1.06%	0.57%	0.85%	1.04%	8.15%
Dataset scale	120k	1920k	500k	250000k	217k	145k	1705k	1613k	18k	12k	4k

Author

KIE Community