Text archives Help
- From: tom fogal <tfogal@alumni.unh.edu>
- To: iv3d-users@sci.utah.edu
- Subject: [IV3D-USERS] Re: Importing large raw datasets
- Date: Mon, 18 Jan 2010 10:49:24 -0700
Hi Louis,
Louis Borgeat <louis.borgeat@nrc-cnrc.gc.ca> writes:
>
I have been trying to import a raw volume dataset with imageVis3D.
>
The dataset is a cube of about 10K^3 x 8 bit = ~1TB.
>
The import task was killed after about 10 days due to a power outage,
>
and judging from the variations in temp files sizes, was far from
>
completed. The machine used has large&fast disk arrays, and plenty
>
of resources. Is this normal behavior for such a large dataset, or
>
should this have been faster ?
Unfortunately, as you're seeing, there are some known performance
issues with the conversion process. It requires at least two O(n)
passes over the data, and the bricking stage in particular is not at
all cache friendly.
>
What is the best/fastest way to import TB-sized datasets into
>
ImageVis3D ?
We could invest some time into fixing the issues I know about (which
would be a Good Thing anyway), but realistically bricking a terabyte is
going to take an absurd amount of time no matter how you slice it. A
better approach is to have data that ImageVis3D can read without any
conversion.
There are two paths to get there: have your data source write out a
different format that fits better with IV3D, and modify ImageVis3D's IO
subsystem to understand your format better. Attacking the problem on
both fronts is likely to be a good approach. While it doesn't directly
cover this, the PDF I sent to Jim Kress (on this list) over the weekend
would probably be enlightening.
The big questions are: 1) do you have someone in your lab that can
write C++ code, and 2) do you have control over your output format?
Cheers,
-tom
Archive powered by MHonArc 2.6.16.