yes the classloader is jar based and so the classloaders are not the same and so the calls are serialized, i.e. normal invocation of RMI. The throughput is with network invocations in your case (although 10ms is high, we have it running in <1ms *with network* on win2k) However what you are seeing is normal, the optimization kicks in with classes in one JAR. (Even though your throughput isn't quite high enough, what VM are you using?) It tells me that we should clearly stress that to get optimizations going you must package in one jar? |THE PROBLEM: | |When I package all three beans in a single jar (and describe in a single |ejb-jar.xml file) I can get throughput for a given case of 2,581 |WorkObjects | |per second. that's roughly 0.3ms per invocation which is decent | |When I package each bean in a separate jar (each with its own |META-INF/ejb-jar.xml file) the throughput drops to 90 WorkObjects per |second (all other parameters equal) | that's roughly 10ms per invocation and that is price you pay for your network serialization (it is high by the way, you are obviously serializing big objects). (again we get <1ms per invocation on certain beans on win2k) |QUESTION: | |Is there any explaination for this? | yep, we optimize the first case, the second is "standard EJB", and it just means that the object you are passing in invocations are *big* marc | |Thanks, |Scott Morrison | | | | |-- |-------------------------------------------------------------- |To subscribe: [EMAIL PROTECTED] |To unsubscribe: [EMAIL PROTECTED] |Problems?: [EMAIL PROTECTED] | | -- -------------------------------------------------------------- To subscribe: [EMAIL PROTECTED] To unsubscribe: [EMAIL PROTECTED] Problems?: [EMAIL PROTECTED]
