JRocket and WLS 6.1 SP2

Question

I was wanting to find out if anyone has had any experiences using JRocket with
Weblogic Server 6.1 SP2. I know that BEA doesn't really support that configuration
but we have gotton it to run and were curious if anyone else had.
We are experiencing some initial wierdness and were wondering if it was a symptom
of using JRocket with 6.1 or perhaps it was the way things are supposed to work.
We are load testing our application. Once we start hitting the server and several
users are doing some processing the CPU usage goes high and our response times
increase. But then after a few minutes the CPU usage goes down and response times
decrease also. This happens even though at this point several more users have
logged onto the system. About 30 minutes later into the test we experience a similar
situation where CPU usage and response times go up.
Is this where JRocket is performing optimizations?
Thanks!
Rich

Answer 1

Yes, JRockit continues to optimize the code. The first native code produced
is not optimal (although a lot faster than interpreted code). JRockit will
monitor the system and find out where the application spends most of its
time, and then these methods will be regenerated but with a lot of
optimizations turned on in the codegenerator. This is done in the
background, though, and should not impair performance.
Regards,
/Staffan
"Rich Llaca" <[email protected]> skrev i meddelandet
news:[email protected]...
>
>
Thanks Staffan.
I guess I didn't really realize that JRocket would be compiling all themethods
that it ran through. I thought that as it ran that it compiled thosehotspots
only at first but still did standard interpretation.
So after the code path has run through at least once...does JRocketcontinue to
optimize as needed even though the code has been compiled into nativecode?
>
Thanks Again,
Rich
"Staffan Larsen" <[email protected]> wrote:
Rich,
There shouldn't be any problems in running WebLogic Server 6.1 on top
of
JRockit 7.0.
I think what you see is the code generator in JRockit taking CPU time,
at
least for your first peak. Since JRockit does not have an interpreter,
every
method is compiled to machine code the first time it is called. This
takes
some time (and is also why JRockit generally starts slower than
interpreter-based VMs). What I would suggest is that you run through
the
system before you apply the high load. Just have your load script run
through every step once before you run it with many users. This will
make
sure that most of the machine code needed to run your application has
been
generated.
The second peak is not obvious. Does your load script change behavior
at
about this time, thus requiring more code to be generated? It could also
be
the optimizations, but I don't think they should be noticeable.
You can start JRockit with -Xverbose:codegen to see a message for each
generated method. This will print a whole lot of messages, but with this
you
can see if the peaks corresponds to codegeneration.
Regards,
/Staffan
"Rich Llaca" <[email protected]> skrev i meddelandet
news:[email protected]...
I was wanting to find out if anyone has had any experiences using
JRocket
with
Weblogic Server 6.1 SP2. I know that BEA doesn't really support thatconfiguration
but we have gotton it to run and were curious if anyone else had.
We are experiencing some initial wierdness and were wondering if itwas a
symptom
of using JRocket with 6.1 or perhaps it was the way things are supposedto
work.
We are load testing our application. Once we start hitting the serverand
several
users are doing some processing the CPU usage goes high and our
response
times
increase. But then after a few minutes the CPU usage goes down andresponse times
decrease also. This happens even though at this point several moreusers
have
logged onto the system. About 30 minutes later into the test we
experience
a similar
situation where CPU usage and response times go up.
Is this where JRocket is performing optimizations?
Thanks!
Rich