OPENMPE Archives

January 2003

OPENMPE@RAVEN.UTC.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Gavin Scott <[log in to unmask]>
Reply To:
Gavin Scott <[log in to unmask]>
Date:
Tue, 28 Jan 2003 10:32:08 -0800
Content-Type:
text/plain
Parts/Attachments:
text/plain (45 lines)
Matthew writes:
> From the latest DSPP newsletter:
>
> "Aries is the PA-emulator that allows PA machine code to execute on an
> Itanium Processor Family (IPF) based system and the following short
> article discusses some of the issues surrounding it's use in running
> ISV's PA-based applications without need to re-build the application
> under IPF.

Aries is the component used to allow PA-RISC application binaries to run
under HP-UX on IPF-based systems.  It is nothing new, having been around
since the genesis of the Hp-UX IPF development process.  When CSY did their
investigation into producing an Native Mode MPE for IPF, they investigated
what it would take to extend Aries to be able to be the basis for a new
"Compatibility Mode" that would allow running PA-RISC NM programs on top of
an IPF-based MPE system.  I believe that investigation determined that this
was possible but would require plenty of work as Aries does not support all
of the PA-RISC features that MPE needs.

Also Aries as it exists today is unable to run a complete operating system,
as it does not "virtualize" enough of the system state and does not support
emulation of all the system-level functions.  So you could not use it as a
platform emulator to run PA-RISC MPE as-is.

Aries is a combination emulator and load-time code translator.  I believe
its primary mode of operation is to translate segments of PA-RISC code into
native IPF instructions at he time the program is loaded or the code segment
is referenced, though it can also fall back to pure emulation for
untranslatable code.

It's my recollection that the original performance goal was to be able to
run translated PA-RISC code at up to half of the speed of native IPF code.
Also note that their 2-8x performance slow down measurements are usually
referenced to the PA-RISC code running on some particular PA-RISC box, *not*
to native code running on the same IPF system!  So your performance may
vary.  I would not want to run a CPU-intensive PA-RISC application on an IPF
box without migrating it to true Native Mode.

Just as with the x86 to IPF transition, the IPF systems do not yet offer
enough of a performance advantage to overcome the penalty for running legacy
(non-native) applications, and there are not enough native applications yet,
so buying an IPF system still requires that you know what you're doing.

G.

ATOM RSS1 RSS2