Re: crossing magic 1 GB limit

New Message Reply Date view Thread view Subject view Author view

From: Don Tidrow (dtidrow++at++nvl.army.mil)
Date: 04/11/2001 10:52:02


The biggest thing I've found is that the default address where the Performer
libraries are loaded by default prevent you from getting more than 1G of arena
space. By moving the default load address down below the start of normal
program space, I've managed to get a 1.5G arena. Much more than that would
probably threaten the normal heap and stack space.

Ran Yakir published a script that helped in this - check his message from
Monday.

On 11-Apr-2001 Matthew Hall wrote:
> Dirk Scheffter wrote:
>>
>> Hi there,
>>
>> a customer of our institute has just reported that our application dies
>> exactly if it reaches a use of 1 GB data space.
> ...
>> However, what does this mean? Do I have to increase shared arena
>> size? Do I have to use lib64 instead of lib32? (Is this dangerous?)
>> (Or does it depends? Do I need more information from our customer?)
>>
>
> I have found in the past (with a non-performer app even) that
> one must switch to -64 (and hence the lib64 libraries) when you
> surpass the 1 gig limit. At least when shared memory is involved.
> I noticed no problems recompiling with -64 and 64 bit libs.
>
Only if you don't depend on 3rd-party libraries that are only available in
n32 format - if there is a way to use n32 libraries in a 64-bit program, I'm
all ears (would be immensely useful to me)

> It would seem that 32-bit pointers could access 4G of data (2^32),
> but perhaps there is some overhead or other intricacies involved.
>
On MIPS cpu's, only the lower 2G of the 32-bit address space is available for
user-mode data (code, stack, heap, assorted .so's, etc.) - the upper 2G is
reserved for kernel-mode stuff.

> -matt
> -----------------------------------------------------------------------
> List Archives, FAQ, FTP: http://www.sgi.com/software/performer/
> Open Development Project: http://oss.sgi.com/projects/performer/
> Submissions: info-performer++at++sgi.com
> Admin. requests: info-performer-request++at++sgi.com

-- 
"If Java had true garbage collection, most programs would delete
themselves upon execution." - Robert Sewell

-------------------------------------------------- | Don Tidrow | | Visual Simulation Developer | | US Army TACOM (currently working with NVESD) | | ph: (703)704-1361 fax: (703)704-1753 | --------------------------------------------------


End of MIME message


New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Wed Apr 11 2001 - 10:52:07 PDT

This message has been cleansed for anti-spam protection. Replace '++at++' in any mail addresses with the '@' symbol.