From: Curtis Schroeder (graphics++at++curts.mailshell.com)
Date: 01/10/2005 05:30:22
I still use Performer 2.2.13 on Onyx2 machines (IRIX 6.5.13m), so this
might not be entirely relevent, but you may need the remapLibs trick to
increase your shared memory size on your Onyx-class machines. Only by
using remapLibs am I able to run with PFSHAREDSIZE=1572864000 (1.46 GB)
on my 12 & 16 CPU Onyx2 racks. I also make the following calls during
the init of my application.
pfSharedArenaBase( (void *)0x20000000 ); /* 512 MB */
/* Get the Performer shared memory arena pointer */
arena = pfGetSharedArena();
/* Have afree(), arecalloc(), and arealloc() place freed memory at the
beginning of the free list(s) */
amallopt( M_FREEHD, 1, arena );
Curt
info-performer Mailing List wrote:
> From: Steve Satterfield <steve++at++nist.gov>
> Date: Sun, 9 Jan 2005 23:01:20 -0500
> Subject: [info-performer] Re: help with pfMemory and PFSHAREDSIZE
>
>Thanks to Jimmy Wang of SGI, I got a hint of why my application runs on an
>Octane2 and Linux but not on Onyx4. It seems to be related to multi processing
>on the larger machines.
>
>My application uses perfly and a locally developed file loader.
>
>This command will run on Octane2 but not on Onyx4:
>
> setenv PFSHAREDSIZE 1000000000
> setenv PFSHAREDBASE 0x20000000
> perfly demo.seq
>
>However, this command will run on the Onyx4
>
> perfly -m 0 demo.seq
>
>So now my question is why does multiprocessing reduce the apparent size of the
>shared memory? Do the various stages of Performer copy parts or all of the
>scene graph? Is there anything I can to other than restrict the Onxy4 to a
>single process?
>
>Thanks,
>Steve
>
>
-- Curt Schroeder | Lockheed Martin Aeronautics Company - Marietta ===================================================================== Embedded Software Engineer Email: graphics++at++curts.mailshell.com
This archive was generated by hypermail 2b29 : Mon Jan 10 2005 - 05:30:39 PST