I have a fairly large Xapian index (3.7M docs, 280+GB in 5 Flint indexes), and I'm surprised by how slow phrase searching is in comparison to other queries. A search that (estimates that it) returns 20K documents for ' a ( b NEAR c )' takes 631 seconds, while ' a b c ' takes less than a second to return an estimate of 40K docs. Both searches return exactly the same top 10 documents. I understand why phrase searching is more expensive than a simple 'AND' query, but 10.5 minutes vs under a second is not what I'd expect. Anyone have hints on what I could do to improve performance? Xapian 1.0.6, Win2k3 Server, Dual-quad proc, 8GB RAM, 500GB of 15K RPM disk, RAID 5. Thanks for any help /cco
On Wed, Aug 27, 2008 at 01:40:57PM -0700, Chris Olds wrote:> Anyone have hints on what I could do to improve performance? > > Xapian 1.0.6, Win2k3 Server, Dual-quad proc, 8GB RAM, 500GB of 15K RPM disk, > RAID 5.Profile to show us where the time is spent. I've no idea what's a good profiling tool for Windows, but if you're able to reboot into Linux temporarily, see: http://trac.xapian.org/wiki/ProfilingXapian Cheers, Olly