lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Karl Wettin <karl.wet...@gmail.com>
Subject Re: instantiated contrib
Date Fri, 27 Aug 2010 08:47:19 GMT
Can you tell us what your queries are? Is it simple term queries,  
phrases, fuzzy, etc?

I think the bad guy here is the term "hotel", that so many documents  
contains it. You could try loading the full index to II and see how  
long time it take to match just that term and compare. And then try a  
more unique term.

If this indeed is the bad guy here are two solutions that might work  
depending on your needs:

Replace all queries for "hotel" with a filter and remove "hotel" from  
your term queries. This might decrease p/r for phrases etc.

Index and query using shingles to create more unique terms. This will  
also improve p/r for phrases if you don't already do phrase queries  
and if you do it should save you time at query time as you might no  
longer need the phrase queries. But the index size will grow.

I'm sure there are other solutions too, but to be quite honest I think  
your corpus is pushing the limit for when it make sense with an II. My  
guess is that you can get it perhaps 2x faster than a RAMDirectory if  
you tune it right.



	karl

27 aug 2010 kl. 05.34 skrev Li Li:

> if I index only 7k documents, the time comparison:
> time1: 7602331019 time2: 4246878035 total1: 10736 total2: 7393
> it seems II is faster than RAMDirectory.
>
> My indexed texts are all hotel names (chinese and english, litter  
> french).
> it has about 100k terms. terms such as hotel is very frequent and
> hotel name is very rare(exception Hotel Chain).
> so I guess it's distribution is a litter term is very frequent and
> other term is very rare.
>
> 2010/8/27 Karl Wettin <karl.wettin@gmail.com>:
>> My mail client died while sending this mail.. Sorry for any  
>> duplicate.
>>
>> It is strange that it should take 20 second to gather fields, this  
>> is the
>> only thing that really suprises me. I'd expect it to be instant  
>> compared to
>> RAMDirectory. It is hard to say from the information you provided.  
>> Did you
>> perhaps lazy load field values from your RAMDirectory and not  
>> retrieve them,
>> or something like that?
>>
>> Why your queries are slow is also hard to say, there can be many  
>> reaons. 70k
>> documents can be quite a few documents for II if they contain  
>> enough text.
>> Here are a few questions that may or may not be helpful:
>>
>> What is the content of the documents? Do they contain a lot of the  
>> same
>> text? Or are they all rather unique? The major thing that makes II  
>> faster
>> than RAMDirectory is that it does not have to deserialize values  
>> from the
>> bytestream. As the index grows binary searching for documents  
>> containing a
>> given term will start consume more time than deserializing the index.
>>
>> What speed do you see if you only load 10% (7k)?
>>
>> Did you see the graphics in the package level javadocs?
>> http://lucene.apache.org/java/3_0_2/api/all/org/apache/lucene/store/instantiated/package-summary.html
>>
>>
>>        karl
>>
>>
>> 26 aug 2010 kl. 09.24 skrev Li Li:
>>
>>> I have about 70k document, the total indexed size is about 15MB(the
>>> orginal text files' size).
>>>               dir=new RAMDirectory();
>>>               IndexWriter write=new IndexWriter(dir,...;
>>>               for(loop){
>>>                    writer.addDocument(doc);
>>>               }
>>>        writer.optimize();
>>>        writer.close();
>>>
>>>        IndexReader ir=IndexReader.open(dir,true);
>>>        InstantiatedIndex ii=new InstantiatedIndex(ir);
>>>        InstantiatedIndexReader iir=new InstantiatedIndexReader(ii);
>>>        is=new IndexSearcher(ir);
>>>        is2=new IndexSearcher(iir);
>>>
>>>             I calculate the time by:
>>>        long searchStart=System.nanoTime();
>>>        TopDocs docs=is.search(bQuery,Integer.MAX_VALUE);
>>>        long searchEnd=System.nanoTime();
>>>
>>>            I searched 10,000 documents and the time of RAMDirectory
>>> and instantiated
>>>            the time used is time1: 21s(21812978000 ns) time2:
>>> 20s(20713817000 ns)
>>>            I also calulate the time including get field value:
>>>               total1: 23852ms total2: 22610ms
>>>           it seems instantiated is not much faster than
>>> RAMDirectory. Is there any thing wrong I used? my max memory is 4GB
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
>>> For additional commands, e-mail: dev-help@lucene.apache.org
>>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
>> For additional commands, e-mail: java-user-help@lucene.apache.org
>>
>>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>


---------------------------------------------------------------------
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