[
https://issues.apache.org/jira/browse/LUCENE-4598?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Michael McCandless updated LUCENE-4598:
---------------------------------------
Attachment: LUCENE-4598.patch
I made a hacked up patch to test how a specialized (payloads, dgap/vint,
counting) 2nd pass aggregation would perform (attached):
{noformat}
Task QPS base StdDev QPS comp StdDev
Pct diff
LowTerm 29.28 (1.2%) 31.01 (1.4%)
5.9% ( 3% - 8%)
MedTerm 14.28 (1.5%) 16.19 (1.4%)
13.4% ( 10% - 16%)
HighTerm 4.05 (2.4%) 5.05 (1.5%)
24.6% ( 20% - 29%)
{noformat}
So ... I think we should provide specialized impls for the common cases, and
make it easy to for users to use (eg FacetsCollector.create(FSP) or something).
These results are similar to what I saw with the single-valued DocValue
collector at
https://issues.apache.org/jira/browse/LUCENE-4600?focusedCommentId=13527566&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13527566
> Change PayloadIterator to not use top-level reader API
> ------------------------------------------------------
>
> Key: LUCENE-4598
> URL: https://issues.apache.org/jira/browse/LUCENE-4598
> Project: Lucene - Core
> Issue Type: Improvement
> Components: modules/facet
> Reporter: Michael McCandless
> Assignee: Shai Erera
> Fix For: 4.1, 5.0
>
> Attachments: LUCENE-4598.patch, LUCENE-4598.patch, LUCENE-4598.patch,
> LUCENE-4598.patch
>
>
> Currently the facet module uses MultiFields.* to pull the D&PEnum in
> PayloadIterator, to access the payloads that store the facet ords.
> It then makes heavy use of .advance and .getPayload to visit all docIDs in
> the result set.
> I think we should get some speedup if we go segment by segment instead ...
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
