One other question: are you using real query logs or a set of unique queries? With real query logs, the caches will warm up after a while (tens of minutes) and performance will improve.
With a set of unique queries, you are mostly measuring Solr cache misses. For us, that is about 4X slower, and we have a small index. wunder On 10/8/08 11:17 AM, "Rajiv2" <[EMAIL PROTECTED]> wrote: > > Hi, thanks for responding so quickly, > >> 6-12 seconds seems really long and 15 million docs is nothing on a >> machine like this. Are you sure the issue is in Solr? How are you >> measuring the 6-12 seconds? > > I'm looking at the <QTime> value in the Solr response. > >> Assuming it is Solr... > >> How often are you indexing? How often do you commit and get new >> searchers? What's your JVM heap size? Are you warming? Is your >> index optimized? Did you turn off the compound file system? > > This is basically a test that I'm doing and it's not in production yet, so I > did a one time index and I haven't committed any new documents. > - JVM heap size is 12 GB > - I am autowarming > - Index is optimized > - useCompoundFile is false > >> You said you've "done most of the optimizations", can you be specific? > > - I have a minimum # of stored fields, 5 out of 25. > - My index is optimized > - HashDocSet is set to around 75000 > - I've setup autowarming queries > - Haven't warmed sort fields because I'm not doing any sorting > - Not using any solid state drives > - Using filters instead of queries for filtering. > >> >> >> thanks, >> Rajiv >> >> -- >> View this message in context: >> http://www.nabble.com/Need-help-with-Solr-Performance-tp19881808p19881808.htm>> l >> Sent from the Solr - User mailing list archive at Nabble.com. >> > > -------------------------- > Grant Ingersoll > > Lucene Helpful Hints: > http://wiki.apache.org/lucene-java/BasicsOfPerformance > http://wiki.apache.org/lucene-java/LuceneFAQ > > > > > > > > > >
