Lately I have been learning more about running ColdFusion applications in JEE application servers and servlet containers. The product I have focused on primarily is Apache’s Tomcat, a very free, open source, and scalable JEE platform. So far I have enjoyed the experience.

However recently I have been asked to do some performance testing on various portions of the application I work on for my day job. And today I came across disturbing results. JRun was actually performing faster than Tomcat! Eeep! This seemed to me a travesty considering the age and lack of support for the JRun product.

So originally my JVM arguments looked like this.

set JAVA_OPTS=-Xms256m -Xmx512m -XX:+UseConcMarkSweepGC -XX:GCTimeRatio=19

Basically this is telling Tomcat to run with a minimum heap size of 256MB, a maximum heap size of 512MB, use the concurrent sweep garbage collector. The -XX:GCTimeRatio argument tells the JVM to have a goal of spending 5% of its time on garbage collection, the remaining being devoted to application time. With this configuration I was finding JRun to be the faster server. Why??

Then I noticed a little parameter in my JRun JVM configuration, -server. After a little searching I have found that the HotSpot JVM has two modes of operation: -client and -server. The difference appears to be how the memory manager and compiler optimizes memory and class compiling/instantiation. For the client configuration it apparently optimizes speed more toward new generation objects, while the server configuration flag optimized for better perm generation management.

For the application I was working on this made a big difference! Adding that flag to my JVM arguments enhanced performance in Tomcat for this app, and made me much happier! So now my JVM argument setting in catalina.bat looks like this.

set JAVA_OPTS=-server -Xms256m -Xmx512m -XX:+UseConcMarkSweepGC -XX:GCTimeRatio=19