On Feb 1, 2007, at 1:26 PM, Alexey Petrenko wrote:
Guys,
is that right that we can publish IBM VME and JRockit VM versions for
Harmony with Harmony class library performance results on our
performance page?
I thought it was mentioned that at the momemnt, we can't do it for
JRockit because of limits contained in the license. You need to
double check that.
geir
SY, Alexey
2007/1/31, Tim Ellison <[EMAIL PROTECTED]>:
Henrik Stahl wrote:
>> If there are any design points in Harmony classlib or VM
>> interface that you think would unduly affect JRockit
>> achieving peak performance then we should talk about them too.
>
> Yes. Here's where it gets fun: Standard JRockit is based on class
> libraries that we license from Sun. This is not the case with J9
if I
> understand correctly.
J9 is used with the Sun class libraries in IBM Java SDK 5.0, so it
is a
similar situation.
> So if you run the JRockit JVM vs the JRockit eval VM for Harmony
and
> compare the performance, you will have a reasonable estimate on the
> performance of the *class library*. Not entirely apples-to-
apples, of
> course, but it's a start.
Yep. Harmony has made some design choices, for example in the
choice of
VM kernel classes, that may affect performance with JRockit. The
kernel
classes include those where the VM/JIT/GC has some intimate
knowledge of
their behaviour and shape, such as Class, Thread, and Reference.
The intent is that VMs can augment that set as they see fit with any
additional types that are 'kernel' for them, overriding the generic
classlib version.
Regards,
Tim
