Ok, so further investigation shows a fairly constant overhead of about 15-25 microsecs per invocation using standard mbeans. I've seen references to over 1M invocations/sec here:

http://www.samspublishing.com/articles/article.asp?p=27196&seqNum=4

What do I need to do to get that kind of performance?

What is the downside of exposing a shortcut like getDirectRef() and doing direct invocations rather than going through the mbean proxy?

TIA

Here's the code in question...

jboss-service.xml:



hello


no hands
<depends optional-attribute-name="HelloWorld" proxy-type="attribute">my.hello:service=hw1




HelloWorldMBean.java:

package hello;

import org.jboss.system.ServiceMBean;
public interface HelloWorldMBean extends ServiceMBean {
public void setMessage(String msg);
public void sayHi();

public void setHelloWorld(HelloWorldMBean hello);
public HelloWorldMBean getDirectRef();
public void echo();
}


HelloWorld.java:

package hello;

import org.jboss.system.ServiceMBeanSupport;
public class HelloWorld extends ServiceMBeanSupport implements HelloWorldMBean {

private double r;
private double incr = 1.7d;

private String message;
private HelloWorldMBean hello;

public HelloWorldMBean getDirectRef() {
return this;
}

public void setMessage(String msg) {
message = msg;
}

public void sayHi() {
r += Math.random() / incr;
}

public void setHelloWorld(HelloWorldMBean hello) {
this.hello = hello;
}

public void echo() {
hello.sayHi();
}

protected void startService() {
log.info("Starting " + message);
if (hello != null) {
double t1 = testEcho();
log.info("invoke ms " + t1 / 1000000.0);
}
}

protected void stopService() {
log.info("Stopping " + message);
}

private double testEcho() {

for (int i = 0; i < 50000; i++) {
hello.sayHi();
}

long start = System.currentTimeMillis();
for (int i = 0; i < 1000000; i++) {
hello.sayHi();
}
long elapsed = System.currentTimeMillis() - start;
double rate = (1000000 * 1000.0 / elapsed);
log.info("jmx " + rate);

return elapsed;
}
}