Newer Versions

v2.1
v2.0

Older Versions

v1.6
v1.7

LucidWorks Enterprise v1.8

Support Resources

LucidWorks Forum
KnowledgeBase

This is the documentation for LucidWorks Enterprise v1.8. The most current release is v2.1.

Skip to end of metadata
Go to start of metadata

A number of configuration items can be manipulated for better performance when benchmarking LucidWorks Enterprise. Implementing some of these optimizations may require directly configuring Solr via schema.xml and solrconfig.xml. See the Solr documentation at http://wiki.apache.org/solr/ for more details on Solr customizations that may be right for your implementation.

  • Ensure that you are running the JVM in server mode.
  • Allocate only as much memory as needed to the JVM heap. The rest should be left free to allow the operating system to cache as much of the Lucene index files as possible.

Improving indexing speed

  • Minimize indexing the same content in more than one field. Each field should be either indexed on its own or Solr's copyField functionality can be used to copy it to an indexed catch-all field.
  • Avoid storing the same content more than once. The target field of copyField commands should almost never be stored.
  • Avoid commits during the indexing process. Turn off Solr auto-commit and avoid explicitly committing until indexing has completed.

Improving Search speed

  • Perform a variety of searches before starting any timings. This warms up the server JVM, and causes parts of the index, commonly used sort fields and filters to be cached by the operating system.
  • Search in as few fields as possible. A single indexed catch-all text field containing the contents of all the other searchable fields (generated by copyField commands) will be faster to search than a multi-field query across many indexed fields.
  • If necessary, turn off relevancy enhancers such as proximity phrase queries, date recency boosts, and synonym expansion to generate benchmarks for comparison with later tests when those features are re-enabled.
  • Retrieve the minimum number of stored fields that still provide a optimal search experience for users.
  • Only retrieve the number of documents that are immediately necessary. The start and rows query arguments may be used to request pages of results.
  • For a large index (on *NIX), force key parts of the indexed portion into operating system cache by changing to the index directory and executing cat *.prx *.frq *.tis > /dev/null
  • Optimize the index occasionally to minimize the number of segments in the index. When benchmarking, the number of segments in the index should be noted.
  • Review the section on Wildcards at Start of Terms if leading wildcards have been enabled for important performance considerations.
Enter labels to add to this page:
Please wait 
Looking for a label? Just start typing.