lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael McCandless <luc...@mikemccandless.com>
Subject Re: Efficient optimization of large indexes?
Date Fri, 07 Aug 2009 16:15:17 GMT
On Thu, Aug 6, 2009 at 5:30 PM, Nigel<nigelspleen@gmail.com> wrote:
>> Actually IndexWriter must periodically flush, which will always
>> create new segments, which will then always require merging.  Ie
>> there's no way to just add everything to only one segment in one
>> shot.
>>
>
> Hmm, that makes sense now that you mention it.  And if you have to merge in
> the end anyway, there's no point to my idea of adding docs to a new index.

Well it's worth a shot at least ;)

> But addIndexes(IndexReader[]) as you suggest would solve that problem.

Right.  You could also set a massive mergeFactor to achieve the same thing.

>> Merge performance does seem rather slow... I recently profiled it and
>> was suprised to find that the merging of terms dict & postings was cpu
>> bound, even on a modern CPU (core i7 920) and with 3 merges running
>> concurrently.  I think most of the CPU cost comes from the pqueue
>> that's used to do the merge sort, plus read/writeVInt.  When Lucene
>> [eventually] switches to PForDelta, that should be more CPU friendly.
>
> That's interesting.  I recently did one of our big merges on a different
> server that has the same disks as the one I was using before, but a faster
> processor.  It seemed like the merge was quite a bit faster than usual
> (though it's possible I was fooled by other factors).

OK... so that supports the "merging is cpu bound" hypothesis.

> Also, it's tons of IO because for each merge it must read every single
>> byte and write nearly every single byte, so that's ~2X bytes moved.
>> Then, if you have more segments in your index than your mergeFactor,
>> multiple such merges are needed and you're looking at, at least, 4X
>> your index size in net bytes moved.  If you have CFS enabled, it's 8X
>> the index size.
>
> Not to get too sidetracked, but this reminds me of another question I meant
> to ask.  We use the compound format right now.  Our merge factor is
> relatively low, so switching to the non-compound format would certainly be
> possible without running into problems with too many open files.  Is there
> any significant speed different between compound and non-compound for
> indexing, searching, or merging?  (Searching for us would be the most
> important by far.)

Both indexing and searching should be faster if you don't use CFS.
For indexing it'll be a small impact... I'm less sure about searching
but I'd also expect the impact to be smallish.  If you test it, please
post back!

>>   * If possible, make sure you always add the same fields to your
>>    docs, in the same order (this results in consistent numbering of
>>    field name -> number).  This is very much an unexpected
>>    gotchya... the merging of stored fields and term vectors is much,
>>    much faster if the field numbers are identical.  LUCENE-1737 is
>>    open to fix Lucene so it consistently numbers automatically, but
>>    it's somewhat tricky because many places in Lucene assume the
>>    field names are densely packed.
>
> We generally do this already, but some of our fields are nullable and so for
> some documents the number-to-name mapping will be different.  Is there any
> value in adding dummy values like "NULL" in these cases?  That presumably
> adds overhead of its own, though.

If you store fields and term vectors then adding null fields should
speed up merging substantially and likely won't have much impact on
indexing time, time to retrieve docs, searching time or index size, I
think.  Post back :)

Mike

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


Mime
View raw message