From: Tim Bradshaw on
On 2009-11-01 20:48:51 +0000, "joswig(a)corporate-world.lisp.de"
<joswig(a)lisp.de> said:

> IBM offers a $125000 processor for mainframes which have JVM support
> in microcode:

Did you read the Wikipedia article at all?

"Currently a zAAP engine contains the same underlying hardware as any
other mainframe engine. However, IBM uses microcode to limit code
execution to Java and XML workloads only. As with Integrated Facility
for Linux (IFL) and zIIP, zAAPs are not specifically optimized to run
Java faster or better. IBM does not change the "software model" or MSU
rating on its mainframes with the installation of zAAPs. Consequently
IBM's customers do not incur software license charges for their zAAPs,
reducing their costs of running Java- and XML-oriented applications."

In other words: it is the same HW as a normal z-series engine, but with
some restrictions in the microcode which will limit what it will run.
It is "not specifically optimized to run Java faster or better". This
is a licensing trick in other words.

Azul are actually interesting: their thing is support for huge GC'd
heaps. This is not particularly language-specific, though they clearly
see their market as the big-JVM people, and they're selling it as such.
I'm pretty sure that their technology is language-neutral inside (and
obviously as commodity systems start growing memories up towards a TB
(they're probably already there actually) this kind of pauseless
support for huge heaps will appear in them). I bet Azul will be gone
in a year or so, of course.

Right, I really will shut up now.

From: joswig on
On 1 Nov., 22:49, Tim Bradshaw <t...(a)cley.com> wrote:
> On 2009-11-01 20:48:51 +0000, "jos...(a)corporate-world.lisp.de"
> <jos...(a)lisp.de> said:
>
> > IBM offers a $125000 processor for mainframes which have JVM support
> > in microcode:
>
> Did you read the Wikipedia article at all?
>
> "Currently a zAAP engine contains the same underlying hardware as any
> other mainframe engine. However, IBM uses microcode to limit code
> execution to Java and XML workloads only. As with Integrated Facility
> for Linux (IFL) and zIIP, zAAPs are not specifically optimized to run
> Java faster or better. IBM does not change the "software model" or MSU
> rating on its mainframes with the installation of zAAPs. Consequently
> IBM's customers do not incur software license charges for their zAAPs,
> reducing their costs of running Java- and XML-oriented applications."
>
> In other words: it is the same HW as a normal z-series engine, but with
> some restrictions in the microcode which will limit what it will run.
> It is "not specifically optimized to run Java faster or better".  This
> is a licensing trick in other words.
>
> Azul are actually interesting: their thing is support for huge GC'd
> heaps.  This is not particularly language-specific, though they clearly
> see their market as the big-JVM people, and they're selling it as such.
>  I'm pretty sure that their technology is language-neutral inside (and
> obviously as commodity systems start growing memories up towards a TB
> (they're probably already there actually) this kind of pauseless
> support for huge heaps will appear in them).  I bet Azul will be gone
> in a year or so, of course.
>
> Right, I really will shut up now.

http://en.wikipedia.org/wiki/AVR32

The AVR32 architecture supports a Java Virtual Machine hardware
implementation.

From: joswig on
On 1 Nov., 22:49, Tim Bradshaw <t...(a)cley.com> wrote:
> On 2009-11-01 20:48:51 +0000, "jos...(a)corporate-world.lisp.de"
> <jos...(a)lisp.de> said:
>
> > IBM offers a $125000 processor for mainframes which have JVM support
> > in microcode:
>
> Did you read the Wikipedia article at all?
>
> "Currently a zAAP engine contains the same underlying hardware as any
> other mainframe engine. However, IBM uses microcode to limit code
> execution to Java and XML workloads only. As with Integrated Facility
> for Linux (IFL) and zIIP, zAAPs are not specifically optimized to run
> Java faster or better. IBM does not change the "software model" or MSU
> rating on its mainframes with the installation of zAAPs. Consequently
> IBM's customers do not incur software license charges for their zAAPs,
> reducing their costs of running Java- and XML-oriented applications."
>
> In other words: it is the same HW as a normal z-series engine, but with
> some restrictions in the microcode which will limit what it will run.
> It is "not specifically optimized to run Java faster or better".  This
> is a licensing trick in other words.
>
> Azul are actually interesting: their thing is support for huge GC'd
> heaps.  This is not particularly language-specific, though they clearly
> see their market as the big-JVM people, and they're selling it as such.
>  I'm pretty sure that their technology is language-neutral inside (and
> obviously as commodity systems start growing memories up towards a TB
> (they're probably already there actually) this kind of pauseless
> support for huge heaps will appear in them).  I bet Azul will be gone
> in a year or so, of course.

A Google Tech Talk: http://www.youtube.com/watch?v=5uljtqyBLxI

Java on a 1000 Cores - Tales of Hardware / Software CoDesign
>
> Right, I really will shut up now.

From: Andrew Reilly on
On Sun, 01 Nov 2009 14:57:19 -0800, joswig(a)corporate-world.lisp.de wrote:

> The AVR32 architecture supports a Java Virtual Machine hardware
> implementation.

These embedded JVM extensions aren't about adding extra functionality
above that normally required to run C or Fortran (mostly) programs.
They're about (power-)optimizing the process of decoding and dispatching
the JVM byte-codes. The interesting thing about Java here isn't the
machine model, but the fact that it has defined an instruction set
encoding.

Cheers,

--
Andrew
From: joswig on
On 2 Nov., 03:27, Andrew Reilly <andrew-newsp...(a)areilly.bpc-
users.org> wrote:
> On Sun, 01 Nov 2009 14:57:19 -0800, jos...(a)corporate-world.lisp.de wrote:
> > The AVR32 architecture supports a Java Virtual Machine hardware
> > implementation.
>
> These embedded JVM extensions aren't about adding extra functionality
> above that normally required to run C or Fortran (mostly) programs.  
> They're about (power-)optimizing the process of decoding and dispatching
> the JVM byte-codes.  The interesting thing about Java here isn't the
> machine model, but the fact that it has defined an instruction set
> encoding.

http://www.atmel.com/dyn/resources/prod_documents/doc32049.pdf

* the processor has a Java mode, where it executes JVM instructions
* some JVM instructions are executed in hardware
* there is an efficient trap mechanism
* it knows about some java data structures
* GC support