IBM's 2 msec latency GC j9

IBM sets real-time tempo for Java code with Metronome
By David Lammers
EE Times (09/12/05, 09:00:00 AM EDT)

Seems IBM is back in the Java-RT game. Three years ago at the Embedded Systems Conference in San Jose, I got a response from one of their main RTSJ-Java j9 developers that was pretty discouraging, but now they seem to be back. Here is an article about their 2 millisec worst case garbage collector called Metronome.
During our initial tests for our framework (in 2002), the then IBM j9 was overall a good performer, the only reason why we couldn't select it then was the unknown future of this VM in the realtime domain. I think even QNX wound up replacing the VM in their Neutrino/QNX6 RTOS (after a promising RT-Java start in 2001), since the direction of IBM wasn't clear.
Welcome back, IBM!

Also, the article contains interesting little side-kicks from Greg Bolella (distinguished engineer at Sun Microsystems Software and director of the real-time Java program):

"Both real-time garbage collection and scoped memory are necessary for these applications. Anyone, or any company, who intimates anything different just isn't clued in."

and

...The real-time-collection vs. scoped memory argument these IBM guys like to make is completely spurious..."

Greg was at IBM in charge of RT-Java when IBM kickstarted the RTSJ effort. He was the lead for the RTSJ 'JSR0001' for quite some time. He moved on to Sun right around the time when IBM appeared to loose interest in RT-Java.
Looks like he ain't afraid to say what he thinks...
In the end, he's right, it should not be a "either/or" choice for the consumer.