Re: [Neo4j] Index Performance(Was: Event framework has landed)

2010-06-30 Thread Mattias Persson
Since write operations are confined in a transaction, data is only written in tx.finish() så that's where neo4j does the invalidation of the searcher. I.e. assuming multiple write operations are performed in each transaction searchers need not be invalidated as often, and only when something is act

Re: [Neo4j] Index Performance(Was: Event framework has landed)

2010-06-30 Thread Atle Prange
Yes, but isn't that what you in practice have to do when using lucene as an index for values in neo4j? Data could be written all the time, and then you have to open a new reader every time you do a search, at least when the index for the value you are searching for has changed. How is that impleme

Re: [Neo4j] Index Performance(Was: Event framework has landed)

2010-06-29 Thread Mattias Persson
I think that calling writer.getReader() and closing it after each lookup causes it to be slow. Basically only create a new reader/searcher if the underlying writer has written something since last time you used the reader/searcher 2010/6/28 Atle Prange > I looked over the Lucene test. I am no Lu

Re: [Neo4j] Index Performance(Was: Event framework has landed)

2010-06-28 Thread Atle Prange
I looked over the Lucene test. I am no Lucene expert, so when i finally managed to actually find the documents i indexed, it was so slow i cannot imagine i am doing things right: package org.ogrm.test; import java.io.File; import java.io.IOException; import org.apache.lucene.analysis.KeywordAnal

Re: [Neo4j] Index Performance(Was: Event framework has landed)

2010-06-24 Thread Atle Prange
I guess i have to brush the dust of my tree knowledge then... -atle On Thu, 2010-06-24 at 09:43 +0200, Mattias Persson wrote: > 2010/6/23 Atle Prange > > > Hm, i'll have to fix that... > > > > Any thoughts on a Trie implementation? Would it be able to compete? > > > > I have no idea on performa

Re: [Neo4j] Index Performance(Was: Event framework has landed)

2010-06-24 Thread Mattias Persson
2010/6/23 Atle Prange > Hm, i'll have to fix that... > > Any thoughts on a Trie implementation? Would it be able to compete? > > I have no idea on performance or what would be the best approach. I though your alphabet-relationship-types approach sounded quite interesting. Or as a b-tree or some o

Re: [Neo4j] Index Performance(Was: Event framework has landed)

2010-06-23 Thread Atle Prange
Hm, i'll have to fix that... Any thoughts on a Trie implementation? Would it be able to compete? atle On Wed, Jun 23, 2010 at 11:04 AM, Mattias Persson wrote: > I think the lucene test is flawed since it never returns any results in > lookup method. That's why it's so fast :) > > 2010/6/22 A

Re: [Neo4j] Index Performance(Was: Event framework has landed)

2010-06-23 Thread Mattias Persson
I think the lucene test is flawed since it never returns any results in lookup method. That's why it's so fast :) 2010/6/22 Atle Prange > Started a new thread since the old got a bit long, if you want to > catch up read the thread "The event framework has landed". > > Okay, i changed the tests t

[Neo4j] Index Performance(Was: Event framework has landed)

2010-06-22 Thread Atle Prange
Started a new thread since the old got a bit long, if you want to catch up read the thread "The event framework has landed". Okay, i changed the tests to reflect a bit more realistic usage. The tests first inserts 1M entries to create a base of data. After that it makes reads and writes 1000 entr