Unexpected cpu load and Solr incrase response time

[mikopacz <ka...@gmail.com>]

Hi, we currently have 2 servers running on JBoss container (master and slave)
with 20mln documents and about 3GB index size.
Java was executed with options:
*-Xms12G -Xmx12G -XX:NewSize=4G -XX:MaxNewSize=4G -XX:MaxPermSize=256m
-Dorg.jboss.resolver.warning=true -Dsun.rmi.dgc.client.gcInterval=3600000
-Dsun.rmi.dgc.server.gcInterval=3600000 -XX:+UseCompressedOops -XX:+UseTLAB
-XX:+UseConcMarkSweepGC -XX:+UseParNewGC -XX:+CMSParallelRemarkEnabled*

Commit duration on master is 5 minutes and we use Solr 3.3 (because in 3.4
we have a problem with dataimport
https://issues.apache.org/jira/browse/SOLR-2804). 

We have problem that occurs when server gets about *34* qps. *Do you have
any advice how to fix this problem?* I have attached the charts below. The
load and threads count increases between 19:00 and 20:00. On 20:10 we
reduced by half the number of queries.

http://lucene.472066.n3.nabble.com/file/n3527914/solr_users_reqs-day.png 
http://lucene.472066.n3.nabble.com/file/n3527914/load-day.png 
http://lucene.472066.n3.nabble.com/file/n3527914/threads-day.png 
http://lucene.472066.n3.nabble.com/file/n3527914/jboss_threads-day.png 
http://lucene.472066.n3.nabble.com/file/n3527914/cpu-day.png 
http://lucene.472066.n3.nabble.com/file/n3527914/_avg_response_query_time-22-11-2011_15_50_17.png 

--
View this message in context: http://lucene.472066.n3.nabble.com/Unexpected-cpu-load-and-Solr-incrase-response-time-tp3527914p3527914.html
Sent from the Solr - User mailing list archive at Nabble.com.