Tim Ellison's misguided remarks

Tim Ellison of the Harmony project has attempted to discredit my blog post on Intel Nehalem.  I'm quite aware of IBM's stellar performance on a "similar" platform, however I wasn't at the liberty to comment on it until a public statement from IBM was made.  Now that has happened...

Congratulations to IBM J9!  Your SPECjbb2005 score is impressive and I look forward to beating it shortly :-)  No, seriously, you've done a great job.  However,  why don't you submit a SPECjvm2008 result?  Actually no company besides Sun has submitted a result!  Why is that?  From my perspective the benchmark is being avoided because no one can beat the Sun result.  If a JVM was truly the fastest on the planet it would step up and beat Sun at SPECjvm2008.  Same applies for single-JVM SPECjbb2005, looking at your current results a single-JVM run IBM J9 should look good.  Come on, make me eat my words.

Look for a blog about SPECjvm2008 on Nehalem in the next few days...

All of our performance optimizations for Intel's latest processor are found in the OpenJDK JVM now.  Sun HotSpot JVM is open source, you can find all the source here.  Sun Java 6 Updates and Performance releases are Sun's distribution of Open JDK.  All performance optimizations make their way into Open JDK, HotSpot optimizations have made it into Open JDK faster than the library changes.  The latest performance release will run on hs15, Open JDK runs on hs15.  To say that we're not open source is simply wrong.

Sun and Open JDK take reliability and stability very seriously.  It is the foundation of a vast majority of Java deployments out there today, let alone its use on the client.  Spouting complaints about how slow we integrate performance optimizations into Open JDK is misleading/misguided.  As for TreeMap, I have to say we're very close.  Its under-review for our outbound open-source contributions.  I look forward to announcing it in the weeks to come, and also look forward to Tim's £500 donation.  Its slow because we are diligent about overall performance impact, and strict when it comes compatibility.  SPECjbb2005 is small potatoes when considering reliability and stability.


Dave, you've been around long enough to know (I hope) that I wasn't directing my comments towards you personally, and if that was the impression I apologize.

Performance leadership is, as you say, an ephemeral claim, and I was also careful to refer to the openness of the class library changes -- so I stand by my comment that 'today "Sun Java powered by the HotSpot Server VM" is not the fastest and not open source'. I fully expect that to change as all Java implementers strive to deliver improvements.

Posted by Tim Ellison on April 02, 2009 at 09:09 PM EDT #

I don't take any of this personally. I enjoy a blog flame toss from time to time (and I hope you do too!). I understand and respect your opinions, we just don't agree at the moment.


Posted by Davd Dagastine on April 02, 2009 at 11:26 PM EDT #

...and long may the mutual passion to improve Java continue. Peace.

Posted by Tim Ellison on April 03, 2009 at 12:18 AM EDT #

Performance always sound great, but what about availability of the VMs? \*None\* of the cited VMs is currently available for public download.
IBM lets me download JDK 6 SR4 (they used SR5),
Oracle JRockit's download page lists R 27.6.0 (they used P 28.0.0)
Sun currently offers 6u6-p (benchmarked was 6u14-p)

Posted by Stefan on April 03, 2009 at 06:38 AM EDT #

SPEC benchmarks are highly competitive, and many times JVM vendors leverage SPEC's 3 month availability rule for JVMs. Speaking for Sun, Java Update 6 Performance Release will be available in June (around Java One).

Posted by David Dagastine on April 03, 2009 at 08:33 AM EDT #

Can someone please elaborate a bit on the (openess of the) class library mentioned here? What's this all about?


Posted by Christian on April 03, 2009 at 12:13 PM EDT #

Christian, take a look at my latest blog post.


Posted by David Dagastine on April 07, 2009 at 11:49 AM EDT #

'Sun Java 6 Updates and Performance releases are Sun's distribution of Open JDK.'

No they're not. For one thing, OpenJDK is JDK7 isn't it? The Sun update releases will be moving/have moved to hs14 which has been forked privately by Sun. And I'm not even going to bother with all the class library differences... I don't know about the performance releases, but as far as I'm aware, the 6 update releases are based off a completely different internal Teamware repository.

'All performance optimizations make their way into Open JDK,'

Eventually :)

'The latest performance release will run on hs15, Open JDK runs on hs15.'

But one is JDK6, and one is JDK7. OpenJDK6 has hs11. IcedTea6 has replaced this with hs14 for the performance benefits, but recent updates have been made in private, while public development continues on the unstable hs15.

Posted by Andrew John Hughes on April 08, 2009 at 03:01 PM EDT #

Post a Comment:
Comments are closed for this entry.



« July 2016