JVM CPU USAGE

Question

I am working on load tests for c to jni to java. While doing these tests, it appears as if the first or initial instantiation of a class hits the jvm the hardest & uses any where from 25 % to 15 % of the cpu. This number eventually goes down, dropping to around 8.7% in the first few seconds then slowing dropping to .6 after 2 minutes. It takes ruffly 45 minutes for it to drop to 0.0 & stay there. Subsequent hits do not appear to affect this number greatly once it has leveled out, so after the initial hit, we are good to go.
The issue is that we have 300 users on unix. Approximately 2/3 of those users will login approximately the same time. I can drop the number to .8 if I delay 60 seconds prior to instantiating the first class. Is there anyway to get around this issue. Are there some settings in the jvm that I could utilize to drop this number? I have searched the internet & not found anything useful. Is this just typical of the jvm? Any help or insights would be greatly appreciated!
Thanks, Janet.

Answer 1

I am working on load tests for c to jni to java.
While doing these tests, it appears as if the first
or initial instantiation of a class hits the jvm the
hardest & uses any where from 25 % to 15 % of theDo you mean that you've observed that the initial instantiation of any class takes more time than the subsequent ones? It might be class load (in many cases, a class would be loaded when a first instance of this class is created). Or if you include in "instantiation" the execution of the constructor code, and for some classes constructors are large enough, then what you see may be initial interpreted execution of the constructor (maybe combined with class loading), that takes much time. Then this constructor and/or the code it calls gets compiled into machine code, which reduces execution time.
BTW, how (which tool/method) do you measure that class instantiation uses 25% of CPU time? Percentage of CPU time seems like not the most appropriate metrics for such a (normally short) event.
cpu. This number eventually goes down, dropping to
around 8.7% in the first few seconds then slowing
dropping to .6 after 2 minutes. It takes ruffly 45
minutes for it to drop to 0.0 & stay there.
Subsequent hits do not appear to affect this number
greatly once it has leveled out, so after the initial
hit, we are good to go.
The issue is that we have 300 users on unix.
Approximately 2/3 of those users will login
approximately the same time. I can drop the number^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Each starting his/her own JVM or what? Why long first instantiation time affects each of them, not just the first one?
to .8 if I delay 60 seconds prior to instantiating
the first class. Is there anyway to get around this
issue. Are there some settings in the jvm that IIf you have multiple JVMs indeed, an ideal solution would be a "multitasking JVM" like the experimental one called MVM, once developed at Sun, but still not productized after a few years. It works as kind of a Java-level OS, capable of starting multiple applications, well isolated from each other, within a single JVM. Otherwise, I can only suggest to look at your code and check why this instantiation takes that much time, and how the code can be improved. Perhaps some things can be delayed or done later on demand, etc.
could utilize to drop this number? I have searched
the internet & not found anything useful. Is this
just typical of the jvm? Any help or insights would
be greatly appreciated!
Thanks, Janet.