OK, here some refinements for systems running a kernel 2.6:
- swap files should have the same speed as swap partitions
- you can have up to 32 swap files/partitions
- with a current mkswap (in util-linux), each file/partition can have up
to 64GB. Without that, it's 2GB.
- with kernel 2.6, processes can grow up to any size (physical ram + swap
is the limit)
- file sizes on block-oriented devices can be in the terabyte range (32bit)
Hope that helps,
Cornelius
On Mon, 11 Oct 2004 13:37:06 +0200, Cornelius Werner
<[log in to unmask]> wrote:
> Besides, swap FILES are extremely slow as opposed to true swap
> PARTITIONS.
> And multiple swap partitions will be reasonably fast only if placed on
> different (physical) hard drives (like in a RAID system). Remember to
> give
> each swap partition entry in your /etc/fstab the same pri=xx statement.
>
> Best regards,
> Cornelius
>
>
> On Mon, 11 Oct 2004 11:26:30 +0100, Stephen Smith <[log in to unmask]>
> wrote:
>
>> Sure - that all makes sense, thanks. Though note - I was referring not
>> to
>> maximum swap size, but there's a separate point, which is that I think
>> that on some OS's there is a limit to the max size that a single process
>> can have, which I think may be 2GB on linux - I think there's ways to
>> change that.
>>
>> Cheers.
>>
>>
>>
>> On Mon, 11 Oct 2004, Denis Brown wrote:
>>
>>> For what it's worth...
>>>
>>> Bear in mind that you can set up multiple swap partitions or swap files
>>> (each 2GB in size) and use them. If I recall reading correctly, up to
>>> eight partitions/files can be used to give the equivalent of 16GB under
>>> a
>>> 2.4 series kernel. The 2.6 kernel may have expanded that. I have
>>> never
>>> had to set up systems that support more than a single swap partition so
>>> I
>>> cannot speak from experience.
>>>
>>> Best regards,
>>> Denis
>>>
>>>
>>> On Mon, 11 Oct 2004, Stephen Smith wrote:
>>>
>>> > Hi. This is a very large dataset - you'll need quite a lot more swap
>>> > probably - try 4GB (though you may have a problem with the 2GB
>>> process
>>> > limit on linux if you need more than that - though I think you can
>>> > change the kernel params to increase that?).
>>> >
>>> > Cheers.
>>> >
>>> >
>>> > On Fri, 8 Oct 2004, Axel Thielscher wrote:
>>> >
>>> > > Hi - I try to analyze an EP image series created with AFNI using
>>> the
>>> > > GLM-module of FSL. I used 3dAFNItoANALYZE provided by AFNI to
>>> transform the
>>> > > EP BRIK to an 4D-Analyze-File. The size of this file is either
>>> around 630 MB
>>> > > (voxel values stored as integer) or 1.3GB (voxel values stored as
>>> floating
>>> > > point) and it contains 1280 volumes of one session.
>>> > >
>>> > > The analysis with FEAT doesn't work with either of the two files.
>>> More
>>> > > specifically, the command
>>> > > /usr/local/fsl/bin/film_gls -rn stats -noest filtered_func_data
>>> design.mat
>>> > > 11.12 3201
>>> > > fails and is apparently unable to read to file.
>>> > >
>>> > > My computer: Red Hat Linux 9, Intel P-4, 1GB RAM, 2GB swap space
>>> > >
>>> > > Does this problem stem from insufficient swap space? The analysis
>>> with AFNI
>>> > > works fine, but I would be happy to use the more advanced features
>>> such as
>>> > > prewhitening etc. provided by FSL. How much memory do you recommend
>>> for
>>> > > input file sizes as quoted above? Can this failure be caused by a
>>> different
>>> > > problem?
>>> > >
>>> > > Thanks,
>>> > > Axel
>>> > >
>>> >
>>> > Stephen M. Smith DPhil
>>> > Associate Director, FMRIB and Analysis Research Coordinator
>>> >
>>> > Oxford University Centre for Functional MRI of the Brain
>>> > John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
>>> > +44 (0) 1865 222726 (fax 222717)
>>> >
>>> > [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>>> >
>>>
>>
>> --
>> Stephen M. Smith DPhil
>> Associate Director, FMRIB and Analysis Research Coordinator
>>
>> Oxford University Centre for Functional MRI of the Brain
>> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
>> +44 (0) 1865 222726 (fax 222717)
>>
>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>>
>
>
>
> --
> Cornelius Werner, AiP
> Institut fuer Medizin (IME)
> AG Kognitive Neurologie
> Forschungszentrum Juelich
> 52425 Juelich
> Germany
>
> Tel. +49-(0)2461-61-8609
>
--
Cornelius Werner
Institut fuer Medizin (IME)
AG Kognitive Neurologie
Forschungszentrum Juelich
52425 Juelich
Germany
Tel. +49-(0)2461-61-8609
|