HIVEMemoryUsage

By default, HIVE now uses KEA++ with an H2-based vocabulary implementation, kea.vocab.VocabularyH2. Prior versions of HIVE used an in-memory implementation, kea.vocab.VocabularySesame. To improve indexing performance, you can switch to the in-memory implementation. Memory requirements are detailed below.

When using the KEA++ indexer with the in-memory vocabulary, depending on the size of the vocabularies used with HIVE, the service may require a significant amount of system memory.

To increase the maximum JVM heap size, use the -Xmx option. To limit the proportion of VM spent in garbage collection before an OutOfMemoryError is thrown, use -XX:UseGCOverheadLimit. In Tomcat, this can be set using the CATALINA_OPTS variable, as in catalina.sh:

CATALINA_OPTS="-Xmx1512m -XX:-UseGCOverheadLimit"

Below are per-vocabulary memory requirements (this is in addition to basic Tomcat overhead). Memory requirements are slightly higher during the vocabulary import process.

Agrovoc	115MB
LCSH	800MB
MeSH	100MB
NBII	45MB
TGN	1.5GB

To operate HIVE with all vocabularies using the KEA++ indexer requires ~3.5GB of heap.