• Benchmark
• Results
• Comparison

JPA Performance Benchmark (JPAB)

Comparison of Hibernate with H2 server vs DataNucleus with Derby embedded

Each of the following tables focuses on a specific database operation, where the last table presents average results comparison.

Speed comparison of JPA database persistence operations (normalized score, higher is better)

Transaction Size =>	Few Entities		Many Entities		Average Score
	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded
Basic Person Test	4.6	4.6	3.0	2.8	3.8	3.7
Element Collection Test	3.7	2.4	1.1	1.6	2.4	2.0
Inheritance Test	6.7	4.6	3.1	2.7	4.9	3.7
Indexing Test	8.2	7.6	4.2	4.8	6.2	6.2
Graph (Binary Tree) Test	3.4	1.6	3.9	1.4	3.7	1.5
Multithreading Test	6.8	8.6	5.0	3.3	5.9	5.9
All Tests	5.6	4.9	3.4	2.8	4.5	3.8

The results above show that in general Hibernate with H2 server is slightly more efficient than DataNucleus with Derby embedded in persisting JPA entity objects to the database.

A large performance gap has been detected when using graphs of objects with large transaction size. Comparing the normalized speed of DataNucleus with Derby embedded database (1.4) to the normalized speed of Hibernate with H2 database server (3.9) reveals that in that case, Hibernate with H2 server is 2.8 times faster than DataNucleus with Derby embedded.

Speed comparison of JPA database retrieval operations (normalized score, higher is better)

Retrieval Size =>	Few Entities		Many Entities		Average Score
	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded
Basic Person Test	3.1	6.8	6.0	14.9	4.6	10.9
Element Collection Test	1.0	5.4	1.1	7.6	1.0	6.5
Inheritance Test	2.9	8.8	7.6	14.8	5.2	11.8
Indexing Test	2.6	9.2	7.0	14.1	4.8	11.6
Graph (Binary Tree) Test	0.42	9.9	1.1	15.8	0.78	12.9
Multithreading Test	6.5	11.9	10.5	11.5	8.5	11.7
All Tests	2.8	8.7	5.6	13.1	4.2	10.9

The results above show that in general DataNucleus with Derby embedded is more efficient than Hibernate with H2 server in retrieving JPA entity objects from the database. Comparing the normalized speed of Hibernate with H2 database server (4.2) to the normalized speed of DataNucleus with Derby embedded database (10.9) reveals that in these tests, DataNucleus with Derby embedded is 2.6 times faster than Hibernate with H2 server.

A huge performance gap has been detected when using graphs of objects with small retrieval size. Comparing the normalized speed of Hibernate with H2 database server (0.42) to the normalized speed of DataNucleus with Derby embedded database (9.9) reveals that in that case, DataNucleus with Derby embedded is 23.6 times faster than Hibernate with H2 server.

Speed comparison of JPA database query operations (normalized score, higher is better)

Retrieval Size =>	Few Entities		Many Entities		Average Score
	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded
Basic Person Test	3.7	39.1	1.0	7.5	2.3	23.3
Element Collection Test	3.7	46.0	0.63	6.7	2.2	26.3
Inheritance Test	3.2	7.7	1.4	3.8	2.3	5.7
Indexing Test	1.7	0.053	5.8	8.0	3.8	4.0
Multithreading Test	1.1	failed	0.88	failed	1.0	failed
All Tests	2.7	23.2	1.9	6.5	2.3	14.8

DataNucleus with Derby embedded has failed in 2 tests (see exceptions).

The results above show that in general DataNucleus with Derby embedded is much more efficient than Hibernate with H2 server in executing the tested JPA queries. Comparing the normalized speed of Hibernate with H2 database server (2.3) to the normalized speed of DataNucleus with Derby embedded database (14.8) reveals that in these tests, DataNucleus with Derby embedded is 6.4 times faster than Hibernate with H2 server.

A huge performance gap has been detected when using JPA element collections with small retrieval size. Comparing the normalized speed of Hibernate with H2 database server (3.7) to the normalized speed of DataNucleus with Derby embedded database (46.0) reveals that in that case, DataNucleus with Derby embedded is 12.4 times faster than Hibernate with H2 server.

On the other hand, DataNucleus with Derby embedded is slower, for instance, when using database indexes with small retrieval size. Comparing the normalized speed of DataNucleus with Derby embedded database (0.053) to the normalized speed of Hibernate with H2 database server (1.7) reveals that in that case, DataNucleus with Derby embedded is 32.1 times slower than Hibernate with H2 server.

Speed comparison of JPA database update operations (normalized score, higher is better)

Transaction Size =>	Few Entities		Many Entities		Average Score
	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded
Basic Person Test	2.9	4.0	1.8	3.8	2.4	3.9
Element Collection Test	2.6	3.7	1.1	4.5	1.9	4.1
Inheritance Test	3.2	4.4	2.5	4.8	2.9	4.6
Indexing Test	2.9	5.2	2.7	4.2	2.8	4.7
Graph (Binary Tree) Test	1.2	2.7	0.43	1.5	0.79	2.1
Multithreading Test	5.6	failed	2.4	failed	4.0	failed
All Tests	3.1	4.0	1.8	3.7	2.5	3.9

DataNucleus with Derby embedded has failed in 2 tests (see exceptions).

The results above show that in general DataNucleus with Derby embedded is more efficient than Hibernate with H2 server in updating JPA entity objects in the database.

A large performance gap has been detected when using JPA element collections with large transaction size. Comparing the normalized speed of Hibernate with H2 database server (1.1) to the normalized speed of DataNucleus with Derby embedded database (4.5) reveals that in that case, DataNucleus with Derby embedded is 4.1 times faster than Hibernate with H2 server.

Speed comparison of JPA database removal operations (normalized score, higher is better)

Transaction Size =>	Few Entities		Many Entities		Average Score
	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded
Basic Person Test	3.8	3.5	2.4	1.3	3.1	2.4
Element Collection Test	1.9	1.9	1.1	1.6	1.5	1.7
Inheritance Test	3.7	2.4	2.5	1.3	3.1	1.9
Indexing Test	7.6	6.6	3.1	1.9	5.3	4.3
Graph (Binary Tree) Test	0.99	0.81	0.91	0.93	0.95	0.87
Multithreading Test	6.1	failed	7.3	failed	6.7	failed
All Tests	4.0	3.0	2.9	1.4	3.5	2.2

DataNucleus with Derby embedded has failed in 2 tests (see exceptions).

The results above show that in general Hibernate with H2 server is more efficient than DataNucleus with Derby embedded in deleting JPA entity objects from the database.

Comparison of JPA/Database speed - the averages (normalized score, higher is better)

Transaction/Retrieval Size	Few Entities		Many Entities		Average Score
	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded	Hibernate H2 server	DataNucleus Derby embedded
Basic Person Test	3.6	11.6	2.8	6.1	3.2	8.8
Element Collection Test	2.6	11.9	1.0	4.4	1.8	8.1
Inheritance Test	4.0	5.6	3.4	5.5	3.7	5.5
Indexing Test	4.6	5.7	4.5	6.6	4.6	6.2
Graph (Binary Tree) Test	1.5	3.7	1.6	4.9	1.6	4.3
Multithreading Test	5.2	10.3	5.2	7.4	5.2	8.8
All Tests	3.7	8.0	3.2	5.7	3.4	6.9

The results above show that in general DataNucleus with Derby embedded is more efficient than Hibernate with H2 server in performing JPA database operations. Comparing the normalized speed of Hibernate with H2 database server (3.4) to the normalized speed of DataNucleus with Derby embedded database (6.9) reveals that in these tests, DataNucleus with Derby embedded is 2.0 times faster than Hibernate with H2 server.

A large performance gap has been detected when using JPA element collections with small transaction/retrieval size. Comparing the normalized speed of Hibernate with H2 database server (2.6) to the normalized speed of DataNucleus with Derby embedded database (11.9) reveals that in that case, DataNucleus with Derby embedded is 4.6 times faster than Hibernate with H2 server.

Other Head to Head DBMS/JPA Comparisons

JPA Performance Benchmark - Copyright © 2010, 2011, 2012 ObjectDB Software Ltd. - All rights reserved.
The benchmark program is released as open source under the GPL license.