Hi, Did you commit (reopen the searcher) during the performance degradation period and did any of your queries use sort? If so, perhaps your JVM is accumulating those thrown-away FieldCache objects and then GC has more and more garbage to clean up, causing pauses and lowering your overall throughput.
Otis -- Sematext -- http://sematext.com/ -- Lucene - Solr - Nutch ________________________________ From: wojtekpia <wojte...@hotmail.com> To: solr-user@lucene.apache.org Sent: Monday, February 9, 2009 6:05:26 PM Subject: Performance degradation caused by choice of range fields In my schema I have two copies of my numeric fields: one with the original value (used for display, sort), and one with a rounded version of the original value (used for range queries). When I use my rounded field for numeric range queries (e.g. q=RoundedValue:[100 TO 1000]), I see very consistent results under load. My hit rate stays the same (at ~23 hits/sec) throughout long running load tests. When I use my original field for range queries, I get performance degradation over time (while under load), rather than consistently worse throughput. For the first 15 minutes, I see throughput similar to my throughput with rounded values, of about 23 hits/second. For the next 15 minutes, I'm down to about 20 hits/second. For the next 15 minutes, I'm down to about 18 hits/second, etc. I expected worse performance by using the non-rounded original value, but I didn't expect degradation. I expected to see throughput of X < 23 hits/second, but consistent at all times. I don't understand why my performance gets worse over time. Any ideas why? I have ~1000 unique values in my rounded field, and ~ 100,000 unique values in my un-rounded field. Thanks. Wojtek -- View this message in context: http://www.nabble.com/Performance-degradation-caused-by-choice-of-range-fields-tp21924197p21924197.html Sent from the Solr - User mailing list archive at Nabble.com.
