[
https://issues.apache.org/jira/browse/LUCENE-6015?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Adrien Grand updated LUCENE-6015:
---------------------------------
Attachment: LUCENE-6015.patch
Here is a patch proposal that updates the shift value from 14 to 10. This seems
to work fine with the luceneutil benchmarks (see above) and with
http://people.apache.org/~jpountz/sparse.html which reports that the sparse
impl is both faster to build and to iterate when the set contains less than
maxDoc/1000 documents (1000 ~ 2^10).
A shift value of 10 practically means that FixedBitSet will be preferred to
SparseFixedBitSet when more than 1/16 of the long values contain at least one
bit that is set.
> Revisit DocIdSetBuilder's heuristic to switch to FixedBitSet
> ------------------------------------------------------------
>
> Key: LUCENE-6015
> URL: https://issues.apache.org/jira/browse/LUCENE-6015
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Adrien Grand
> Assignee: Adrien Grand
> Priority: Minor
> Attachments: LUCENE-6015.patch
>
>
> DocIdSetBuilder starts with a SparseFixedBitSet and then upgrades to a
> FixedBitSet when the cardinality grows larger than maxDoc >>> 14. However
> Robert improved SparseFixedBitSet performance quite significantly in
> LUCENE-6003 so we should see if it makes sense to update this heuristic.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
