Peformance of setting JVM attribute or callback.

Question

I have a program that receives feeds in the native world and then calls the JVM through JNI with a callback like receivedFeed(attr1, attr2 ....attr20). There are a large number of attributes being passed back for each callback.
The question I have has to do with performance.
Is it faster to use this callback approach where the attributes are put in the signature and then reassembled on the java side into an object? Compared to directly setting each attribute on the Java object from within the native world and then prompting the JVM with a callback like updated();

Answer 1

Very true. I would normally agree. Wrap the parameters in an object to minimize the number of arguments and make it flexible.
Specifically, I am thinking about performance and I am willing to compromise a bit on the beauty of the solution, if it is faster.
I have receiving a feed that has about 20 data items. Mostly ints and longs, and maybe a string. This format is relatively fixed and is not likely to change any time soon. If it does, it is all of my own code so the changes are easy.
In the parameter passing approach, my assumption is that I would have a callback into the JVM with a long signature like
receiveFeedCallback(jint1,jint2,jint3,...jint20);
For the object wrapping approach, I would first have to construct an object. and then do the callback:
jvalue args[4];
     args[0].i=(int) 0;
     args[1].i=(int) val1;
     args[2].i=(int) val2;
     args[3].i=(int) val3;
jobject trans = jniENV_->NewObjectA(clazz,"<init>",args);
jniENV_->CallStaticVoidMethod(jObj,_method_JNITEST_callback,feedTransaction);
In the end, I probably need to know what the overhead of creating a class in Java vs. C++ is compared to the parameter passing approach.
I may be doing this callback about 20(or more) times per millisecond. About 85% of the time in my program is on this one step so I would like to try the fastest approach possible.
In the end, I am trying to write a test case to test the performance....which leads me to another questions and likely another thread in the JNI world....to be posted soon.
Thanks for the help.