Dear Zhang,
The x86 architecture uses a 32-bit int for many things.
Therefore, on older kernels on the x86 architecture, file size is
limited to the values you can store in a 32-bit signed int.
Older glibc and all the programs linked against it inherit this
32-bit signed int limitation, leading to the 2G problem. It never was a
problem on 64-bit architectures like Alpha and Sparc.
Older versions of Matlab still use (g)libc5 to keep compatibility
with all the other Unix-architectures which aren't updated as often
as Linuxboxes may be. Versions up to 12 got some arcane
dependencies. You can help yourself if you use a recent
kernel (2.4.21) and a newer version of gcc (above 3.0).
For the repartitioning:
You would be better off in using the native Linux-commands instead
of that Partition-Magic. Try fdisk and 'man mkswap' for turning swapon.
There's also a excellent commandline-tool called parted (does the same as
PM for free) which is quite intuitive. And it does the thing in your systems
manner.
Greetings
Markus Schatzl
University of Regensburg, Germany
> Dear Zhang,
>
> According to MathWorks, Matlab is a 32-bit application. Thus, a file
> is limited at 2-4 GB.
> Our SuSE 32-bit Intel Linux and 64-bit SUN systems swap spaces are set
> at 10-20 GB (2-4 GB RAM).
> Per Matlab window on both systems, we can have only one biggest 2-GB
> matrix and many smaller matrices.
>
> Cheers,
> Witaya Sungkarat
>
> >Dear Chris,
> >
> >Thank you for your reply.
> >
> >On 27 Aug 2003, Christopher Joseph Bailey wrote:
> >
> >
> >
> >>> Dear Zhang,
> >>>
> >>> Sounds to me that you should consider chopping up that dataset in
> >>
> >>
> >some
> >
> >
> >>> way (what, I can't say). You should also upgrade your system to at
> >>
> >>
> >least
> >
> >
> >>> 1-2GB RAM, because swapping is, in general, a rather bad thing (as it
> >>> slows things down /considerably/).>
> >>
> >>
> >
> >yes indeed, at the end we used SPM to reslice the images and thrinked the
> whole
> >dataset to the size somehow bearable for the PC system, with a price of
> lower image
> >resolution which normally should be avoided, anyway we bought a Sun
> >Workstation later (that was very sad).
> >
> >
> >
> >
> >>> On your swap-problem, I can give an answer. Linux (the kernel, I
> >>
> >>
> >guess)
> >
> >
> >>> has a 2GB limitation for swap devices so expanding swap partitions to
> >>> 10GB won't help. However, you can have /multiple/ swap filesystems,
> >>
> >>
> >2GB
> >
> >
> >>> each. Use PM to repartition your 10GB swap to 5x2GB and try "swapon"
> >>
> >>
> >on
> >
> >
> >>> them separately.
> >>
> >>
> >
> >This is really very helpful and informative, I read someone provided a
> >kernel hack in order to overcome the 2G swap limit, and I am not afraid
> >of recompling the kernel with this patch, but your method looks lot
> easier
> >and safer.
> >
> >Many thanks!
> >
> >Zhang
> >
> >
> >
> >>> I'm no Windows expert (that would be very sad), so can't help you on
> >>> that side.
> >>>
> >>> Good luck!
> >>>
> >>> Best, -Chris
> >>>
> >>>
> >>> --
> >>> Christopher Bailey
> >>> Research Assistant (MSc)
> >>> Center for Functionally Integrative Neuroscience (CFIN)
> >>> [log in to unmask]
> >>> phn: +45 89 49 43 78
> >>> fax: +45 89 49 44 00
> >>>
> >>> Street address:
> >>> Bygning 30, Troejborgvej 72
> >>> Aarhus Kommunehospital, DK-8000 Aarhus C
> >>> Denmark
> >>>
> >>>
> >>> On Tue, 2003-08-26 at 13:28, Hao Zhang wrote:
> >>
> >>
> >>>> > Dear SPMers,
> >>>> >
> >>>> > Sorry this might be bit off the topic,while I'd really like to know
> >>>
> >>>
> >if
> >
> >
> >>>> > someone has gone though the similar situation and the if there is
> >>>
> >>>
> >any
> >
> >
> >>>> > solution for it.
> >>>> > I once obtained a dataset of 5 Gigabyte in size, when my colleague
> >>>
> >>>
> >and I
> >
> >
> >>>> > tried to launch the second-level group analysis, there came the
> >>>> > BIG problem: The 5G dataset seemed to have driven PC to the
> >>>
> >>>
> >performance
> >
> >
> >>>> > limit , the machine is a P4 with 560M RAM, Win2K/Linux dual-boot,
> >>>> > SPM/Matlab compained
> >>>> > out of memory at the very early stage of loading the big group
> >>>
> >>>
> >contrast
> >
> >
> >>>> > files under Win2K, enlarging the virtual memory did not help at
> >>>
> >>>
> >all, the
> >
> >
> >>>> > the situation under Linux is slightly better than under Win2k, but
> >>>
> >>>
> >the task
> >
> >
> >>>> > bar stoped at 60-70% something.
> >>>> > Having looked at the kernel mailing list archive, I started to
> >>>
> >>>
> >realize
> >
> >
> >>>> > there is a 2G limit for the swap partition for Linux(I did set the
> >>>
> >>>
> >swap
> >
> >
> >>>> > partition to the maximum), while there are people saying it is just
> >>>
> >>>
> >the
> >
> >
> >>>> > limitation of swapon utility, but after I used ParitionMagic to set
> >>>
> >>>
> >the
> >
> >
> >>>> > Linux swap partition to 10G , the system still report only 2G of
> >>>
> >>>
> >swap file
> >
> >
> >>>> > system, and I had no clue where to fine-tune the swapon options. I
> >>>
> >>>
> >have
> >
> >
> >>>> > been using Linux quite confidently and thought it can always
> >>>
> >>>
> >replace the
> >
> >
> >>>> > expensive Unix system in every aspect, is there anyone who met such
> >>>
> >>>
> >problem
> >
> >
> >> > bfore and found some solution for it? Many thanks!
> >> >
> >> > best regards
> >> >
> >> > zhang
> >> >
> >> >
> >> >
> >> >
> >> >
> >> > *************************************************
> >> > Zhang Hao, Dr. med. M.Sc
> >> > International Institute for Research in Paraplegia
> >> > Balgrist University Hospital
> >> > Forchstrasse 340
> >> > CH-8008 Zuerich
> >> > Tel.: ++41 1 386 3721
> >> > Fax: ++41 1 386 3731
> >> > email: [log in to unmask]
> >> > *************************************************
> >
>
--
COMPUTERBILD 15/03: Premium-e-mail-Dienste im Test
--------------------------------------------------
1. GMX TopMail - Platz 1 und Testsieger!
2. GMX ProMail - Platz 2 und Preis-Qualitätssieger!
3. Arcor - 4. web.de - 5. T-Online - 6. freenet.de - 7. daybyday - 8. e-Post
|