Agreed, Solr uses random access bitsets everywhere so I'm thinking this could be an improvement or at least a great option to enable and try out. I'll update LUCENE-1536 so we can benchmark.
On Thu, Aug 27, 2009 at 4:06 AM, Michael McCandless<luc...@mikemccandless.com> wrote: > On Thu, Aug 27, 2009 at 6:30 AM, Grant Ingersoll<gsing...@apache.org> wrote: > >>> I am wondering... are new SOLR filtering features faster than standard >>> Lucene queries like >>> {query} AND {filter}??? >> >> The new filtering features in Solr are just doing what Lucene started doing >> in 2.4 and that is using skipping when possible. It used to be the case in >> both Lucene and Solr that the filter was only every applied after scoring >> but before insertion into the Priority Queue. That is now fixed. > > I think performance of filtering can still be further improved, within > Lucene... it's still very much a work in progress. > > EG if a filter is random access (eg RAM resident as a bit set), which > I think for Solr is frequently the case (?), it ought to be applied > just like we now apply deleted documents (LUCENE-1536 is opened for > this). This can result in sizable performance gains, especially for > more complex queries and no-so-dense filters. > > Mike >