[
https://issues.apache.org/jira/browse/LUCENE-1483?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12654922#action_12654922
]
Michael McCandless commented on LUCENE-1483:
--------------------------------------------
Hmmmmmmm.
OK I think I see what could explain this: insertion into the pqueue is
fairly costly. So, because we now make 100 pqueues, each gathering
top N results, we are paying much more insertion cost overall than the
single queue that IndexSearcher(MultiReader) uses.
So.... how about still doing the searches per-sub-reader(searcher),
but, make a HitCollector that gathers the results into a single
pqueue, passing that HitCollector to each sub-searcher?
If that turns out OK, then I think it would make LUCENE-1471 moot
because we should similarly change MultiSearcher to use a single
shared pqueue.
Actually I think this approach should be a bit faster, because there
is some very small method call overhead to how MultiReader implements
TermDocs/Positions by "concatenating" its sub-readers. So by pushing
Searcher down onto each SegmentReader we should gain a bit, but it
could very well be in the noise. For this reason we may in fact want
to do this same thing for the "normal" (sort by relevance)
IndexSearcher.search.
I wish I thought of this sooner. Sorry for the runaround Mark!
> Change IndexSearcher to use MultiSearcher semantics for sorted searches
> -----------------------------------------------------------------------
>
> Key: LUCENE-1483
> URL: https://issues.apache.org/jira/browse/LUCENE-1483
> Project: Lucene - Java
> Issue Type: Improvement
> Affects Versions: 2.9
> Reporter: Mark Miller
> Priority: Minor
> Attachments: LUCENE-1483.patch
>
>
> Here is a quick test patch. FieldCache for sorting is done at the individual
> IndexReader level and reloading the fieldcache on reopen can be much faster
> as only changed segments need to be reloaded.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
