[MINC-users] Determining memory usage for minctracc calls

[Matthijs van Eede]

Hi Andrew,

Knowing that you have attempted to solve this but failed/gave up, spells trouble :-) 

So far we used valgrind, something that worked well for getting a memory profile for mincANTS and mincblur (which uses a file size dependent amount of tmp space, something that threw us off while running on a system where tmp space was allocated as part of the machine's memory). But it looks like the only way to get this right, is to dig into the actual code. If we end up getting to the bottom of this, we'll report back!

Cheers,
Matthijs

----- Original Message -----
From: "Andrew Janke" <a.janke at gmail.com>
To: "MINC users mailing list" <minc-users at bic.mni.mcgill.ca>
Sent: Monday, June 8, 2015 3:50:58 PM
Subject: Re: [MINC-users] Determining memory usage for minctracc calls

Hi Matthijs,

I too have attempted to work this out in the past but failed! There is
something strange going on in there though as once you go past an
input file of about 1GB things go awry. It's something that has been
on my todo list for a long time so I'm very happy if you fix it for
me! Last time I went hunting for this things were going strange in the
grid sub-sampling.

Remember also that minctracc will convert to various filetypes
(typically short) when using the -mi objective function. A while back
there was also a change from rescaled short deformation grids to
double to get around some range issues.


a





On 9 June 2015 at 05:45, Matthijs van Eede <matthijs at mouseimaging.ca> wrote:
> Hi all,
>
> We are trying to determine/estimate how much memory a minctracc call will use depending on the file resolution, number of voxels and the step size used in the minctracc call. So far we have not been able to fit the memory data we've gathered in a way that can predict memory usage for a wide range of input resolutions.
>
> One of the things we realized is that minctracc uses (at least one) memory saving strategy: if a mask is provided for the registration, the bounding box for the region of interest is determined, and only that volume is used for the actual registration. Does any know of any other memory saving strategies that minctracc uses? Or is there anyone who might have an intuition on how the memory usage should vary on file size and step size / resolution?
>
> Thanks in advance,
> Matthijs
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
_______________________________________________
MINC-users at bic.mni.mcgill.ca
http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users