RE: iSeries JVM version

["Urbanek, Marty" <Marty_Urbanek@xxxxxxxxxxxx>]

------------------------------

> date: Mon, 24 Jan 2005 08:16:02 -0600
> from: "Joe Pluta" <joepluta@xxxxxxxxxxxxxxxxx>
> subject: RE: iSeries JVM version

> I'm not saying ANYTHING definitively, Marty.  I'm just saying what the
> numbers said.  The 24-way model 840-2420 scored about 80,000 on JDK
> 1.3.0, while the 32-core (which I assume is 16 2-core CPUs) i5 595
> scored 880,000 on JDK 1.4.2.

> What exactly does this mean?  I don't know, but it's a damned curious
> number, don't you think?

Oh, yeah, it definitely piqued my interest.


>> Have you compared the total MHz available on the 840-2420 24-way
versus
>> the 595 16-way? Could this be more a matter of increased CPU cycles to
run
>> the JVM rather than JVM code?

>I don't know anything except what I told you, and what I told you is
>what I read from the SPEC JBB2000 test results:
>http://www.spec.org/osg/jbb2000/.  Like Don, I'd love to hear what Blair
>Wyman has to say on this issue.

Me, too. Blair is the one who enlightened me to thinking about iSeries Java 
performance in terms of how many cycles are available to run the JVM rather 
than the machine's height, width, weight, CPW, cost, etc. (my gripe being "but 
this little PC over here kicked our _ss at Java execution")

I wish they would have run that jbb2000 test against the i5 595 at JDK 1.3.1 as 
well, to highlight the improvement due to JVM versus hardware.

Did you have any other reference to iSeries JDK 1.4.2 switching to a "standard 
JVM" and/or having major performance gains? I was not aware iSeries had dropped 
the "platform unique" JVM. I agree, for a major performance benefit, dump it. 

Thanks,
-Marty