lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael McCandless <luc...@mikemccandless.com>
Subject Re: [jira] Commented: (LUCENE-1458) Further steps towards flexible indexing
Date Sat, 22 Nov 2008 13:46:38 GMT
Thinking about this more...

One way to compress the terms index in RAM and on disk would be to use
a pre-compiled (during indexing) letter FST (finite state transducer).

It would be similar to a letter-trie, which is similar to a binary
tree except at each node you have N letters as children, so that
common prefixes are shared.  But an FST also shares the common
suffixes, and "outputs" the data you need (in our case the long
offsets into the main terms dict file) in the middle of the word when
you've identified its unique prefix.

To look up a term in the index, you walk letter by letter down this
tree.  At some point you may find no letter matching your term, at
which point you'd fallback to the highest letter lower than the one
you wanted (since we want to look up highest term <= target).
Eventually you will hit an output, which you record as the offset into
main terms file.  After the output there's only 1 edge leaving every
node (in the common suffix part of the FST) but you have to keep
walking those edges to fully reconstruct the term you just found.

I'm sure there are clever implementations that would minimize RAM
usage.

If we [eventually] make the terms index pluggable then people can
explore these different approaches...

Mike

Jason Rutherglen wrote:

> It would be nice to have btree like features such as previous(), min  
> and max.  Also a unique sequence id per term that enables faster  
> lookup if the term id is known.
>
> On Wed, Nov 19, 2008 at 1:38 PM, Michael McCandless <lucene@mikemccandless.com 
> > wrote:
>
> I think we wouldn't do any term compression for the btree, at least  
> for the parts loaded in RAM (we don't today, ie, we create the full  
> Term or String as an array).
>
> For the parts left on disk we should be able to do something similar  
> to what we do today, eg for child nodes only encode the "delta" wrt  
> the parent node?
>
> Mike
>
>
> Jason Rutherglen wrote:
>
> Michael B: Are you interested in making column stride fields  
> realtime and use the btree for the terms?  This is an idea I started  
> on I called tag index where the postings are divided into blocks.   
> The blocks can then be replaced in memory with periodic flush to  
> disk as the in ram postings grows.
>
> Michael M: How would the term compression be handled in a btree model?
>
> On Wed, Nov 19, 2008 at 2:29 AM, Michael McCandless (JIRA) <jira@apache.org 
> > wrote:
>
>   [ https://issues.apache.org/jira/browse/LUCENE-1458?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12648971

> #action_12648971 ]
>
> Michael McCandless commented on LUCENE-1458:
> --------------------------------------------
>
> bq. So something like a B+Tree would probably work better.
>
> I agree, btree is a better fit, though we don't need insertion &  
> deletion operations since each segment is write once.
>
> > Further steps towards flexible indexing
> > ---------------------------------------
> >
> >                 Key: LUCENE-1458
> >                 URL: https://issues.apache.org/jira/browse/LUCENE-1458
> >             Project: Lucene - Java
> >          Issue Type: New Feature
> >          Components: Index
> >    Affects Versions: 2.9
> >            Reporter: Michael McCandless
> >            Assignee: Michael McCandless
> >            Priority: Minor
> >             Fix For: 2.9
> >
> >         Attachments: LUCENE-1458.patch, LUCENE-1458.patch,  
> LUCENE-1458.patch
> >
> >
> > I attached a very rough checkpoint of my current patch, to get early
> > feedback.  All tests pass, though back compat tests don't pass due  
> to
> > changes to package-private APIs plus certain bugs in tests that
> > happened to work (eg call TermPostions.nextPosition() too many  
> times,
> > which the new API asserts against).
> > [Aside: I think, when we commit changes to package-private APIs such
> > that back-compat tests don't pass, we could go back, make a branch  
> on
> > the back-compat tag, commit changes to the tests to use the new
> > package private APIs on that branch, then fix nightly build to use  
> the
> > tip of that branch?o]
> > There's still plenty to do before this is committable! This is a
> > rather large change:
> >   * Switches to a new more efficient terms dict format.  This still
> >     uses tii/tis files, but the tii only stores term & long offset
> >     (not a TermInfo).  At seek points, tis encodes term & freq/prox
> >     offsets absolutely instead of with deltas delta.  Also, tis/tii
> >     are structured by field, so we don't have to record field number
> >     in every term.
> > .
> >     On first 1 M docs of Wikipedia, tii file is 36% smaller (0.99 MB
> >     -> 0.64 MB) and tis file is 9% smaller (75.5 MB -> 68.5 MB).
> > .
> >     RAM usage when loading terms dict index is significantly less
> >     since we only load an array of offsets and an array of String  
> (no
> >     more TermInfo array).  It should be faster to init too.
> > .
> >     This part is basically done.
> >   * Introduces modular reader codec that strongly decouples terms  
> dict
> >     from docs/positions readers.  EG there is no more TermInfo used
> >     when reading the new format.
> > .
> >     There's nice symmetry now between reading & writing in the codec
> >     chain -- the current docs/prox format is captured in:
> > {code}
> > FormatPostingsTermsDictWriter/Reader
> > FormatPostingsDocsWriter/Reader (.frq file) and
> > FormatPostingsPositionsWriter/Reader (.prx file).
> > {code}
> >     This part is basically done.
> >   * Introduces a new "flex" API for iterating through the fields,
> >     terms, docs and positions:
> > {code}
> > FieldProducer -> TermsEnum -> DocsEnum -> PostingsEnum
> > {code}
> >     This replaces TermEnum/Docs/Positions.  SegmentReader emulates  
> the
> >     old API on top of the new API to keep back-compat.
> >
> > Next steps:
> >   * Plug in new codecs (pulsing, pfor) to exercise the modularity /
> >     fix any hidden assumptions.
> >   * Expose new API out of IndexReader, deprecate old API but emulate
> >     old API on top of new one, switch all core/contrib users to the
> >     new API.
> >   * Maybe switch to AttributeSources as the base class for  
> TermsEnum,
> >     DocsEnum, PostingsEnum -- this would give readers API  
> flexibility
> >     (not just index-file-format flexibility).  EG if someone wanted
> >     to store payload at the term-doc level instead of
> >     term-doc-position level, you could just add a new attribute.
> >   * Test performance & iterate.
>
> --
> 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: java-dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-dev-help@lucene.apache.org
>
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-dev-help@lucene.apache.org
>
>


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


Mime
View raw message