[jira] [Commented] (LUCENE-7399) Speed up flush of points v2

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

    [ https://issues.apache.org/jira/browse/LUCENE-7399?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15402919#comment-15402919 ] 

Michael McCandless commented on LUCENE-7399:
--------------------------------------------

Maybe the visitor should also take {{BytesRef}}?  Codec impls could
read a whole byte[] values block in at once; maybe that's a savings.
We can explore that separately in a v3 issue!  We could also fix
{{BKDWriter.writeCommonPrefixes}} to save the copy there, though
that's just once per leaf block.

Shouldn't the {{assertHistogram}} call be called in an {{assert}}?  It
seems to be called directly now.

It looks like you also removed {{Sorter.insertionSort}} in favor of
{{binarySort}}.  Maybe add a javadoc to {{Sorter.binarySort}} saying
it's O(N^2) and is only used once we have recursed down to <= 20 items
to sort?

Have you tweaked 20 to see if that's a good value?  Sorting BKD points
is rather costly since when we swap, we swap whole values (docID,
maybe ord, then the byte[] value for this field).


> Speed up flush of points v2
> ---------------------------
>
>                 Key: LUCENE-7399
>                 URL: https://issues.apache.org/jira/browse/LUCENE-7399
>             Project: Lucene - Core
>          Issue Type: Improvement
>            Reporter: Adrien Grand
>            Assignee: Adrien Grand
>            Priority: Minor
>         Attachments: LUCENE-7399.patch
>
>
> There are improvements we can make on top of LUCENE-7396 to get ever better flush performance.



--
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