lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mark Harwood (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (LUCENE-4069) Segment-level Bloom filters for a 2 x speed up on rare term searches
Date Tue, 10 Jul 2012 08:59:36 GMT

    [ https://issues.apache.org/jira/browse/LUCENE-4069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13410145#comment-13410145
] 

Mark Harwood commented on LUCENE-4069:
--------------------------------------

bq. So now we are close to 1M lookups/sec for a single thread!

Cool!

bq. I wonder if somehow we can do a better job picking the right sized bit vector up front?

bq. You basically need to know up front how many unique terms will be in the given field for
this segment right?

Yes - the job of anticipating the number of unique keys probably has 2 different contexts:
1) Net new segments e.g. guessing up front how many docs/keys a user is likely to generate
in a new segment before the flush settings kick in.
2) Merged segments e.g. guessing how many unique keys survive a merge operation

Estimating key volumes in context 1 is probably hard without some additional hints from the
end user. Arguably the BloomFilterFactory.getSetForField() method already represents where
this setting can be controlled.
In context 2 where potentially large merges occur we could look at adding an extra method
to BloomFilterFactory to handle this different context e.g. something like
   FuzzySet getSetForMergeOpOnField(FieldInfo fi, OneMerge mergeContext)
Based on the size of the segments being merged and volumes of deletes a more appropriate size
of Bloom bitset could be allocated based on a worst-case estimate.
Not sure how we get the OneMerge instance fed through the call stack - could that be held
somewhere on a ThreadLocal as generally useful context?




                
> Segment-level Bloom filters for a 2 x speed up on rare term searches
> --------------------------------------------------------------------
>
>                 Key: LUCENE-4069
>                 URL: https://issues.apache.org/jira/browse/LUCENE-4069
>             Project: Lucene - Java
>          Issue Type: Improvement
>          Components: core/index
>    Affects Versions: 3.6, 4.0
>            Reporter: Mark Harwood
>            Priority: Minor
>             Fix For: 4.0, 3.6.1
>
>         Attachments: BloomFilterPostingsBranch4x.patch, LUCENE-4069-tryDeleteDocument.patch,
LUCENE-4203.patch, MHBloomFilterOn3.6Branch.patch, PKLookupUpdatePerfTest.java, PKLookupUpdatePerfTest.java,
PKLookupUpdatePerfTest.java, PKLookupUpdatePerfTest.java, PrimaryKeyPerfTest40.java
>
>
> An addition to each segment which stores a Bloom filter for selected fields in order
to give fast-fail to term searches, helping avoid wasted disk access.
> Best suited for low-frequency fields e.g. primary keys on big indexes with many segments
but also speeds up general searching in my tests.
> Overview slideshow here: http://www.slideshare.net/MarkHarwood/lucene-bloomfilteredsegments
> Benchmarks based on Wikipedia content here: http://goo.gl/X7QqU
> Patch based on 3.6 codebase attached.
> There are no 3.6 API changes currently - to play just add a field with "_blm" on the
end of the name to invoke special indexing/querying capability. Clearly a new Field or schema
declaration(!) would need adding to APIs to configure the service properly.
> Also, a patch for Lucene4.0 codebase introducing a new PostingsFormat

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message