for JVM options, could somebody comment whether it is suitable for a busy
server?

On Wed, Jan 12, 2011 at 9:58 PM, Glen Mazza <[email protected]> wrote:

> You might wish to create a Metro client (
> http://www.jroller.com/gmazza/entry/soap_client_tutorial) and run the same
> tests on it.  It would be interesting to see response time differences, and
> it may give more indication whether the bottleneck is in CXF or your
> hardware.
>
> Glen
>
>
> On 12.01.2011 08:48, Christopher Cheng wrote:
>
>> I was stress testing a CXF client with 100 connections to SOAP server
>> using
>> JMeter. The result is that for 1 connection, it takes about 5 seconds. For
>> 100 connections, it takes 30 seconds for each thread and load average is
>> around 11-14
>>
>> Is it a problem with CXF or is it a hardware performance problem? What
>> should I do to make it better?
>>
>> The hardware:
>> - Xeon E5520 x 2
>> - 24GB RAM
>> - CentOS 5.3
>> - 64 bit jdk 1.6.0_22-b04
>> - Resin 3.1
>>
>> JVM options:
>> -J-server
>> -XX:NewSize=128m
>> -XX:MaxNewSize=128m
>> -XX:MaxPermSize=512m
>> -XX:SurvivorRatio=5
>> -Xmn1024m
>> -Xss2048k</jvm-arg>
>> -Xms5g
>> -Xmx5g
>> -XX:ParallelGCThreads=20
>> -XX:+UseConcMarkSweepGC
>> -XX:+UseParNewGC
>> -XX:TargetSurvivorRatio=90
>> -XX:MaxTenuringThreshold=31
>> -Dorg.apache.cxf.Logger=org.apache.cxf.common.logging.Log4jLogger
>> -Xdebug
>> -Dcom.sun.management.jmxremote
>>
>>
>

Reply via email to