[jira] [Created] (LUCENE-7121) BKDWriter should not store ords when documents are single valued

["Michael McCandless (JIRA)" <ji...@apache.org>]

Michael McCandless created LUCENE-7121:
------------------------------------------

             Summary: BKDWriter should not store ords when documents are single valued
                 Key: LUCENE-7121
                 URL: https://issues.apache.org/jira/browse/LUCENE-7121
             Project: Lucene - Core
          Issue Type: Improvement
            Reporter: Michael McCandless
            Assignee: Michael McCandless
             Fix For: master, 6.1


Since we now have stats for points fields, it's easy to know up front whether the field you are about to build a BKD tree for is single valued or not.

If it is single valued, we can optimize space by not storing the ordinal to identify a point, since its docID also uniquely identifies it.

This saves 4 bytes per point, which for the 1D case is non-trivial (12 bytes down to 8 bytes per doc), and even for the 2D case is good reduction (16 bytes down to 12 bytes per doc).

This is an optimization ... I won't push it into 6.0.




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org