Re: Sort Performance Problems across large dataset
Peter Hollas wrote: Currently we can issue a simple search query and expect a response back in about 0.2 seconds (~3,000 results) with the Lucene index that we have built. Lucene gives a much more predictable and faster average query time than using standard fulltext indexing with mySQL. This however returns result in score order, and not alphabetically. To sort the resultset into alphabetical order, we added the species names as a seperate keyword field, and sorted using it whilst querying. This solution works fine, but is unacceptable since a query that returns thousands of results can take upwards of 30 seconds to sort them. Are you using a Lucene Sort? If you reuse the same IndexReader (or IndexSearcher) then perhaps the first query specifying a Sort will take 30 seconds (although that's much slower than I'd expect), but subsequent searches that sort on the same field should be nearly as fast as results sorted by score. Doug - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
RE: Sort Performance Problems across large dataset
Sharon, The system that we are working on does recordset paging, so only the first 100 are returned on the first page. But then you are able to page through the rest of the of the results. The Lucene index is stored as a standard FSDirectory on a RAID filesystem, and it used only for the initial name lookup, to discover the SpeciesID. Further requests are handled by the mySQL backend database. The way that Lucene returns results by default is based on a document scoring algorithm, and so results most likely are out of alphabetical order. I think that we will likely have to use a custom hitcollector object to do the sorting. Many thanks, Peter. -Original Message- From: Xiaohong Yang (Sharon) [mailto:[EMAIL PROTECTED] Sent: 25 January 2005 00:37 To: Lucene Users List Subject: Re: Sort Performance Problems across large dataset Hi Peter, I just got on the list a few hours ago. I am still reading the source code. I am not going to send this to the list. I would like to know the .2 sec query time for 2 million fields, should it display only the first page (100 or so), not the whole 3000 found? It is very fast I agree. If the alphabetic index display only a link, not the content, then it should not be very slow since you only need to sort part of what a user need. May be display only the first A page, as it did with the regular scored results. Just my thought, might not work for you. Do you store the Lucene index in the database or in a text file? Best, Sharon LangPower Computing, Inc. http://www.indexingonline.com Peter Hollas [EMAIL PROTECTED] wrote: I am working on a public accessible Struts based species database project where the number of species names is currently at 2.3 million, and in the near future will be somewhere nearer 4 million (probably the largest there is). The species names are typically 1 to 7 words in length, and the broad requirement is to be able to do a fulltext search across them. It is also necessary to sort the results into alphabetical order by species name. Currently we can issue a simple search query and expect a response back in about 0.2 seconds (~3,000 results) with the Lucene index that we have built. Lucene gives a much more predictable and faster average query time than using standard fulltext indexing with mySQL. This however returns result in score order, and not alphabetically. To sort the resultset into alphabetical order, we added the species names as a seperate keyword field, and sorted using it whilst querying. This solution works fine, but is unacceptable since a query that returns thousands of results can take upwards of 30 seconds to sort them. My question is whether it is possible to somehow return the names in alphabetical order without using a String SortField. My last resort will be to perform a monthly index rebuild, and return results by index order (about a day to re-index!). But ideally there might be a way to modify the Lucene API to incorporate a scoring system in a way that scores by lexical order. Any ideas are appreciated! Many thanks, Peter. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
RE: Sort Performance Problems across large dataset
Peter, Let me continue to post to the list since I am here already. Both dynamic sort on-the-fly (hit counter) or presorted and stored index would give you the alphabetically sorted species name. Pre-stored alphabetical index would be faster. Even for 4 million names, the index should not take up too much room if you use integer to represent the sorted strings. ( The integer order represent the species name in alphabetical order.) Using the pre-stored index, and an array of score-sorted document hit list, it would be reasonably fast to compile the alphabetic list for display. This is just what I would design it, might not work for you. What type of database is this? Any example of use available on the web? Best, Sharon Peter Hollas [EMAIL PROTECTED] wrote: Sharon, The system that we are working on does recordset paging, so only the first 100 are returned on the first page. But then you are able to page through the rest of the of the results. The Lucene index is stored as a standard FSDirectory on a RAID filesystem, and it used only for the initial name lookup, to discover the SpeciesID. Further requests are handled by the mySQL backend database. The way that Lucene returns results by default is based on a document scoring algorithm, and so results most likely are out of alphabetical order. I think that we will likely have to use a custom hitcollector object to do the sorting. Many thanks, Peter. -Original Message- From: Xiaohong Yang (Sharon) [mailto:[EMAIL PROTECTED] Sent: 25 January 2005 00:37 To: Lucene Users List Subject: Re: Sort Performance Problems across large dataset Hi Peter, I just got on the list a few hours ago. I am still reading the source code. I am not going to send this to the list. I would like to know the .2 sec query time for 2 million fields, should it display only the first page (100 or so), not the whole 3000 found? It is very fast I agree. If the alphabetic index display only a link, not the content, then it should not be very slow since you only need to sort part of what a user need. May be display only the first A page, as it did with the regular scored results. Just my thought, might not work for you. Do you store the Lucene index in the database or in a text file? Best, Sharon LangPower Computing, Inc. http://www.indexingonline.com Peter Hollas wrote: I am working on a public accessible Struts based species database project where the number of species names is currently at 2.3 million, and in the near future will be somewhere nearer 4 million (probably the largest there is). The species names are typically 1 to 7 words in length, and the broad requirement is to be able to do a fulltext search across them. It is also necessary to sort the results into alphabetical order by species name. Currently we can issue a simple search query and expect a response back in about 0.2 seconds (~3,000 results) with the Lucene index that we have built. Lucene gives a much more predictable and faster average query time than using standard fulltext indexing with mySQL. This however returns result in score order, and not alphabetically. To sort the resultset into alphabetical order, we added the species names as a seperate keyword field, and sorted using it whilst querying. This solution works fine, but is unacceptable since a query that returns thousands of results can take upwards of 30 seconds to sort them. My question is whether it is possible to somehow return the names in alphabetical order without using a String SortField. My last resort will be to perform a monthly index rebuild, and return results by index order (about a day to re-index!). But ideally there might be a way to modify the Lucene API to incorporate a scoring system in a way that scores by lexical order. Any ideas are appreciated! Many thanks, Peter. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Sort Performance Problems across large dataset
I am working on a public accessible Struts based species database project where the number of species names is currently at 2.3 million, and in the near future will be somewhere nearer 4 million (probably the largest there is). The species names are typically 1 to 7 words in length, and the broad requirement is to be able to do a fulltext search across them. It is also necessary to sort the results into alphabetical order by species name. Currently we can issue a simple search query and expect a response back in about 0.2 seconds (~3,000 results) with the Lucene index that we have built. Lucene gives a much more predictable and faster average query time than using standard fulltext indexing with mySQL. This however returns result in score order, and not alphabetically. To sort the resultset into alphabetical order, we added the species names as a seperate keyword field, and sorted using it whilst querying. This solution works fine, but is unacceptable since a query that returns thousands of results can take upwards of 30 seconds to sort them. My question is whether it is possible to somehow return the names in alphabetical order without using a String SortField. My last resort will be to perform a monthly index rebuild, and return results by index order (about a day to re-index!). But ideally there might be a way to modify the Lucene API to incorporate a scoring system in a way that scores by lexical order. Any ideas are appreciated! Many thanks, Peter. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Sort Performance Problems across large dataset
Hi, do you optimize the index? Do you tried to implement a own hit collector? Stefan Am 25.01.2005 um 01:01 schrieb Peter Hollas: I am working on a public accessible Struts based species database project where the number of species names is currently at 2.3 million, and in the near future will be somewhere nearer 4 million (probably the largest there is). The species names are typically 1 to 7 words in length, and the broad requirement is to be able to do a fulltext search across them. It is also necessary to sort the results into alphabetical order by species name. Currently we can issue a simple search query and expect a response back in about 0.2 seconds (~3,000 results) with the Lucene index that we have built. Lucene gives a much more predictable and faster average query time than using standard fulltext indexing with mySQL. This however returns result in score order, and not alphabetically. To sort the resultset into alphabetical order, we added the species names as a seperate keyword field, and sorted using it whilst querying. This solution works fine, but is unacceptable since a query that returns thousands of results can take upwards of 30 seconds to sort them. My question is whether it is possible to somehow return the names in alphabetical order without using a String SortField. My last resort will be to perform a monthly index rebuild, and return results by index order (about a day to re-index!). But ideally there might be a way to modify the Lucene API to incorporate a scoring system in a way that scores by lexical order. Any ideas are appreciated! Many thanks, Peter. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --- company:http://www.media-style.com forum: http://www.text-mining.org blog: http://www.find23.net - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Sort Performance Problems across large dataset
Hi Peter, I just got on the list a few hours ago. I am still reading the source code. I am not going to send this to the list. I would like to know the .2 sec query time for 2 million fields, should it display only the first page (100 or so), not the whole 3000 found? It is very fast I agree. If the alphabetic index display only a link, not the content, then it should not be very slow since you only need to sort part of what a user need. May be display only the first A page, as it did with the regular scored results. Just my thought, might not work for you. Do you store the Lucene index in the database or in a text file? Best, Sharon LangPower Computing, Inc. http://www.indexingonline.com Peter Hollas [EMAIL PROTECTED] wrote: I am working on a public accessible Struts based species database project where the number of species names is currently at 2.3 million, and in the near future will be somewhere nearer 4 million (probably the largest there is). The species names are typically 1 to 7 words in length, and the broad requirement is to be able to do a fulltext search across them. It is also necessary to sort the results into alphabetical order by species name. Currently we can issue a simple search query and expect a response back in about 0.2 seconds (~3,000 results) with the Lucene index that we have built. Lucene gives a much more predictable and faster average query time than using standard fulltext indexing with mySQL. This however returns result in score order, and not alphabetically. To sort the resultset into alphabetical order, we added the species names as a seperate keyword field, and sorted using it whilst querying. This solution works fine, but is unacceptable since a query that returns thousands of results can take upwards of 30 seconds to sort them. My question is whether it is possible to somehow return the names in alphabetical order without using a String SortField. My last resort will be to perform a monthly index rebuild, and return results by index order (about a day to re-index!). But ideally there might be a way to modify the Lucene API to incorporate a scoring system in a way that scores by lexical order. Any ideas are appreciated! Many thanks, Peter. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Sort Performance Problems across large dataset
On Jan 24, 2005, at 7:01 PM, Peter Hollas wrote: I am working on a public accessible Struts based Well there's the problem right there :)) (just kidding) To sort the resultset into alphabetical order, we added the species names as a seperate keyword field, and sorted using it whilst querying. This solution works fine, but is unacceptable since a query that returns thousands of results can take upwards of 30 seconds to sort them. 30 seconds... wow. My question is whether it is possible to somehow return the names in alphabetical order without using a String SortField. My last resort will be to perform a monthly index rebuild, and return results by index order (about a day to re-index!). But ideally there might be a way to modify the Lucene API to incorporate a scoring system in a way that scores by lexical order. What about assigning a numeric value field for each document with the number indicating the alphabetical ordering? Off the top of my head, I'm not sure how this could be done, but perhaps some clever hashing algorithm could do this? Or consider each character position one digit in a base 27 (or 27 to include a space) and construct a number for that? (though that would be an enormous number and probably too large) - sorry my off-the-cuff estimating skills are not what they should be. Certainly sorting by a numeric value is far less resource intensive than by String - so perhaps that is worth a try? At the very least, give each document a random number and try sorting by that field (the value of the field can be Integer.toString()) to see how it compares performance-wise. Erik - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Sort Performance Problems across large dataset
Peter,
Currently we can issue a simple search query and expect a response back
in about 0.2 seconds (~3,000 results)
You may want to try something like the following (I do this in FishEye,
seems to be performant for moderately large field-spaces).
Use a custom HitCollector, and store all the matching doc-ids in a
java.util.BitSet. This will still give you your 0.2second performance.
Then, use a TermDocs iterator to visit each term in your species name
field, printing out (or whatever) each species name if it contains a
docid in your bitset. Something like this pseudocode:
BitSet docs = doSearch(query); // 0.2seconds
TermEnum te = reader.terms(new Term(species-name, ));
TermDocs td = reader.termDocs();
Term t = te.term();
while (t!=null t.field().equals(species-name)) {
td.seek(te);
while (td.next()) {
int docid = td.doc();
if (docs.get(docid)) {
print match: + docid;
break; // try next term
}
}
if (!te.next()) {
break;
}
t = te.term();
}
te.close();
td.close();
Now, with 2.3 million (or 4 million!) species names, I'm not sure how
fast it will be to iterate through all the species-name termdocs. But
I would be interested to find out; if you give this a code a try, could
you report back your results?
=Matt
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
