From: Allan Schaffer (allan++at++sgi.com)
Date: 03/24/2004 08:22:35
Forwarding.. the filters trapped this by mistake.
-------- Original Message --------
Subject: Re: [info-performer] arena size in Linux
Date: Wed, 24 Mar 2004 10:38:17 +0100
From: Bram Stolk <bram++at++sara.nl>
Organization: SARA
To: dr++at++vircinity.com
CC: info-performer++at++sgi.com
References: <40608CFC.9F80498A++at++vircinity.com>
On Tue, 23 Mar 2004 20:16:12 +0100
Daniela Rainer <dr++at++vircinity.com> wrote:
> Hi all,
>
> on a Linux PC (Intel) the maximum arena size we can allocate with
> pfSharedArenaSize() is about 950 MB.If we try larger sizes, we get the
> following error:
>
> PF Warning/Internal: perror: Success
> PF Warning/SysErr: Could not mmap file /usr/tmp/pfClock.5343.pfdpool
> for arena of size 50224
> PF Warning/Internal: perror: Resource temporarily unavailable
> PF Warning/SysErr(11): Could not mmap file /usr/tmp/pfClock.5343.pfdpool
> for arena of size 49152
>
> As this 950 MB is not enough for the application, I wonder if it is
> possible to get it larger. Is the max. adress space on a intel computer
> with Linux 2 GB or 4 GB? Why is the arena size limited to about 1 GB?
Daniela,
I think that there is a very hard limit at 2Gb, because I suspect
that unsigned ints are being used somewhere, either in linux, or in
performer.
See:
$ PFSHAREDSIZE=2200000000 perfly -m0 /usr/share/Performer/data/rocket_tux.pfb
PF Warning/Internal: perror: Cannot allocate memory
PF Warning/SysErr(12): Could not mmap file
/usr/tmp/pfShmemFile.XXXXXX for arena of size -2147482576
As you can see, the 2.2 * 10^9 bytes I requested is shown as a negative
nr (-2147482576) by the PF message. Is this only in the printf, or also in
the ammount requested via mmap?
My guess is that mmap uses unsigned:
posix_types.h:typedef unsigned int __kernel_size_t;
So the unsigned/signed problem may be in performer.
Note that:
$ PFSHAREDSIZE=2000000000 perfly -m0 /usr/share/Performer/data/rocket_tux.pfb
Works for me.
This is on:
dual P4 xeon
kernel 2.4.24
1GB ram
1GB swap
As you get less than 1Gb shared mem, I suggest increasing your swap space.
Check with:
$ free
total used free shared buffers cached
Mem: 1031140 961228 69912 0 393636 157928
-/+ buffers/cache: 409664 621476
Swap: 1052216 0 1052216
Bram
> Thanks for any explaination and Best Regards
> Daniela
> --
> -------------------------------------------------------------------------
> Daniela Rainer | email: dr++at++vircinity.com
> Vircinity GmbH | http://www.vircinity.com
> Nobelstrasse 15, 70550 Stuttgart | Tel: +49 (0) 711 6864534 22
> GERMANY | Fax: +49 (0) 711 6864534 99
>
-- ------------------------------------------------------------------------------ Bram Stolk, VR Specialist. SARA Academic Computing Services Amsterdam, PO Box 94613, 1090 GP AMSTERDAM email: bram++at++nospam.sara.nl Phone +31-20-5923059 Fax +31-20-6683167"For the costs of subsidized agriculture in the EU, we can have all 56 million European cows fly around the world. First Class." - J. Norberg ------------------------------------------------------------------------------
This archive was generated by hypermail 2b29 : Wed Mar 24 2004 - 08:22:27 PST