[
https://issues.apache.org/jira/browse/LUCENE-7774?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15962939#comment-15962939
]
Adrien Grand commented on LUCENE-7774:
--------------------------------------
We were able to achieve significant gains by switching to radix sort
(LUCENE-7306, see annotation O at
http://people.apache.org/~mikemccand/geobench.html#index-times) for the
in-memory part. Maybe we should explore doing something similar for offline
sorting?
> Allow BKDWriter buffer sizes to be configurable
> -----------------------------------------------
>
> Key: LUCENE-7774
> URL: https://issues.apache.org/jira/browse/LUCENE-7774
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Alan Woodward
>
> BKDWriter has a 16Mb internal buffer it uses for sorting points, which then
> spills over to on-disk sorting if there are too many values. The difference
> in performance between in-buffer and on-disk sorting is pretty stark, and it
> would be nice to make this buffer size configurable so that users can choose
> to use more RAM during indexing to speed things up.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
