solr/tomcat performance.

[adm1n <ev...@gmail.com>]

Hi,

I'm running solr+tomcat with the following configuration:
I have 16 slaves, which are being queried by aggregator, while aggregator
being queried by the users.
My slaveUrls variable in solr.xml (on aggregator) looks like - '<property
name="slaveUrls"
value="host01/slave01,host02/slave02,host03/slave03,...,host16/slave16" />'
I'm running it on linux machine (not dedicated, there are some other 'heavy'
processes) with 16 quads CPUs and 66GB Ram.

I ran some tests and I saw, that when I did 400 concurrent requests to
aggregator the host stopped to respond until I restart the tomcat. I tried
to 'play' with tomcat's/java configuration a little, but it didn't help me
much and the main issue was memory usage and timeouts. Currently I'm using
the following settings:
Java:
-Xms256m -Xmx8192m
I tried to tweak -XX:MinHeapFreeRatio setting, but from what I could see no
memory was returned to OS.
Tomcat:
<Executor name="HTTPThreadPool" namePrefix="HTTPThread-"
                        maxThreads="8000"
                        minSpareThreads="4000"/>

<Connector executor="HTTPThreadPool" port="8080" protocol="HTTP/1.1"
redirectPort="8443" URIEncoding="UTF-8"
                        maxHttpHeaderSize="8388608"
                        enableLookups="false"
                        acceptCount="100"
                        connectionTimeout="10000" />

Assuming I'll have ~1000 requests/second done to aggregator, on how many
aggregators should I balance the loading? Or may be I can achieve better
performance only by tweaking the current system?


Any help/advise will be appreciated,
Thanks!

--
View this message in context: http://lucene.472066.n3.nabble.com/solr-tomcat-performance-tp3727199p3727199.html
Sent from the Solr - User mailing list archive at Nabble.com.