[MINC-users] mincaverage error
Robert D. Vincent
robert.d.vincent at mcgill.ca
Thu Jun 11 11:09:48 EDT 2015
The fix I proposed allows mincaverage to correctly calculate the result for
38 files each of ~400 M voxels in size (740 voxels in each dimension).
-bert
On Thu, Jun 11, 2015 at 10:38 AM, Vladimir S. FONOV <
vladimir.fonov at gmail.com> wrote:
> Hello,
>
>
> well, this is interesting. I can imagine that we will have to do something
> about it to make MINC even remotely attractive to the masses.
>
> One of the attractive features was ability to handle very large datasets
> that wouldn't fit into memory. Otherwise - it's very difficult to explain
> why even bother with MINC these days.
>
>
> On 15-06-11 10:07 AM, Robert D. Vincent wrote:
>
>> Hi all,
>>
>> I have a partial fix for this issue. The voxel_loop code enforces a
>> maximum
>> number of open files based on the constant MI_MAX_NUM_ICV, which is
>> currently set to 32. The way the code is written, if it cannot open all of
>> the files, it opens them one at a time. This causes some serious overhead
>> that seems to lead to an out-of-resource condition in the HDF5 library.
>>
>> I will continue to investigate whether we can fix the resource leak, but
>> in
>> the short run I will increase the number allowable open files to something
>> more realistic, like 1000. This can be done by changing the value of
>> MI_MAX_NUM_ICV in minc.h.
>>
>> -bert
>>
>> On Thu, Jun 11, 2015 at 10:06 AM, Robert D. Vincent <bert at phalarope.com>
>> wrote:
>>
>> Hi all,
>>>
>>> I have a partial fix for this issue. The voxel_loop code enforces a
>>> maximum number of open files based on the constant MI_MAX_NUM_ICV, which
>>> is
>>> currently set to 32. The way the code is written, if it cannot open all
>>> of
>>> the files, it opens them one at a time. This causes some serious overhead
>>> that seems to lead to an out-of-resource condition in the HDF5 library.
>>>
>>> I will continue to investigate whether we can fix the resource leak, but
>>> in the short run I will increase the number allowable open files to
>>> something more realistic, like 1000. This can be done by changing the
>>> value
>>> of MI_MAX_NUM_ICV in minc.h.
>>>
>>> -bert
>>>
>>> On Tue, Jun 9, 2015 at 4:49 AM, Andrew Janke <a.janke at gmail.com> wrote:
>>>
>>> that's the idea. it is in mine...
>>>>
>>>> a
>>>> On 09/06/2015 6:48 PM, "Simon Eskildsen" <eskild at gmail.com> wrote:
>>>>
>>>> Does it mean I can safely replace mincaverage in xfmavg with
>>>>> mincbigaverage? Why hasn't already?
>>>>>
>>>>> On Tue, Jun 9, 2015 at 10:37 AM, Andrew Janke <a.janke at gmail.com>
>>>>>
>>>> wrote:
>>>>
>>>>>
>>>>> supports most options yes. Will be identical apart from very small
>>>>>>
>>>>> rounding
>>>>>
>>>>>> errors.
>>>>>>
>>>>>> take note of the options for defining the number to average at once.
>>>>>>
>>>>>> a
>>>>>> On 09/06/2015 6:25 PM, "Simon Eskildsen" <eskild at gmail.com> wrote:
>>>>>>
>>>>>> Thanks. I'm trying this now. I assume same behaviour as mincaverage,
>>>>>>>
>>>>>> right?
>>>>>>
>>>>>>>
>>>>>>> On Tue, Jun 9, 2015 at 10:12 AM, Andrew Janke <a.janke at gmail.com>
>>>>>>>
>>>>>> wrote:
>>>>>
>>>>>>
>>>>>>> If it's memory, try mincbigaverage. google knows where it is.
>>>>>>>>
>>>>>>>> a
>>>>>>>> On 09/06/2015 5:15 PM, "Simon Eskildsen" <eskild at gmail.com>
>>>>>>>>
>>>>>>> wrote:
>>>>
>>>>>
>>>>>>>> Hi Robert,
>>>>>>>>>
>>>>>>>>> I'm averaging 38 deformation fields. Here's the info on one of
>>>>>>>>>
>>>>>>>> them:
>>>>>
>>>>>>
>>>>>>>>> $ mincinfo 21/stx2_0004_20131113_091044_t1.mnc.21.i_grid_0.mnc
>>>>>>>>> file: 21/stx2_0004_20131113_091044_t1.mnc.21.i_grid_0.mnc
>>>>>>>>> image: signed__ float -10.166246414184570312 to
>>>>>>>>>
>>>>>>>> 11.006509780883789062
>>>>>
>>>>>> image dimensions: xspace yspace zspace vector_dimension
>>>>>>>>> dimension name length step start
>>>>>>>>> -------------- ------ ---- -----
>>>>>>>>> xspace 388 0.5 -96.5
>>>>>>>>> yspace 460 0.5 -132.5
>>>>>>>>> zspace 388 0.5 -78.5
>>>>>>>>> vector_dimension 3 unknown unknown
>>>>>>>>>
>>>>>>>>> The files are quite big (~800 Mb each). Could this be a memory
>>>>>>>>>
>>>>>>>> problem?
>>>>>>
>>>>>>>
>>>>>>>>> I used mincaverage without any flags/options. Just a list of
>>>>>>>>>
>>>>>>>> files
>>>>
>>>>> in
>>>>>
>>>>>> and a
>>>>>>>>
>>>>>>>>> file out.
>>>>>>>>>
>>>>>>>>> Simon
>>>>>>>>>
>>>>>>>>> On Mon, Jun 8, 2015 at 3:46 PM, Robert D. Vincent <
>>>>>>>>> robert.d.vincent at mcgill.ca> wrote:
>>>>>>>>>
>>>>>>>>> Hi Simon,
>>>>>>>>>>
>>>>>>>>>> It looks like HDF5 might be running out of resource ids, which
>>>>>>>>>>
>>>>>>>>> seems
>>>>>>
>>>>>>> unlikely, but there could be a resource leak somewhere.
>>>>>>>>>>
>>>>>>>>>> Can you give me the details of the failing case (e.g. the
>>>>>>>>>>
>>>>>>>>> files
>>>>
>>>>> you're
>>>>>>>
>>>>>>>> using)?
>>>>>>>>>>
>>>>>>>>>> -bert
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On Mon, Jun 8, 2015 at 9:33 AM, Simon Eskildsen <
>>>>>>>>>>
>>>>>>>>> eskild at gmail.com>
>>>>>
>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Hi all,
>>>>>>>>>>>
>>>>>>>>>>> Averaging a bunch of deformation fields using mincaverage, I
>>>>>>>>>>>
>>>>>>>>>> got
>>>>>
>>>>>> this
>>>>>>>
>>>>>>>> error:
>>>>>>>>>>>
>>>>>>>>>>> HDF5-DIAG: Error detected in HDF5 (1.8.14) thread 0:
>>>>>>>>>>> #000: /users/fristed/git/minc-toolkit/build/HDF5/src/H5A.c
>>>>>>>>>>>
>>>>>>>>>> line
>>>>>
>>>>>> 679
>>>>>>>
>>>>>>>> in
>>>>>>>>>
>>>>>>>>>> H5Aget_space(): can't get space ID of attribute
>>>>>>>>>>> major: Invalid arguments to routine
>>>>>>>>>>> minor: Can't get value
>>>>>>>>>>> #001:
>>>>>>>>>>>
>>>>>>>>>> /users/fristed/git/minc-toolkit/build/HDF5/src/H5Aint.c
>>>>
>>>>> line
>>>>>>>
>>>>>>>> 719
>>>>>>>>>
>>>>>>>>>> in
>>>>>>>>>>
>>>>>>>>>>> H5A_get_space(): unable to register dataspace atom
>>>>>>>>>>> major: Object atom
>>>>>>>>>>> minor: Unable to register new atom
>>>>>>>>>>> #002: /users/fristed/git/minc-toolkit/build/HDF5/src/H5I.c
>>>>>>>>>>>
>>>>>>>>>> line
>>>>>
>>>>>> 902
>>>>>>>
>>>>>>>> in
>>>>>>>>>
>>>>>>>>>> H5I_register(): no IDs available in type
>>>>>>>>>>> major: Object atom
>>>>>>>>>>> minor: Out of IDs for group
>>>>>>>>>>> HDF5-DIAG: Error detected in HDF5 (1.8.14) thread 0:
>>>>>>>>>>> #000: /users/fristed/git/minc-toolkit/build/HDF5/src/H5S.c
>>>>>>>>>>>
>>>>>>>>>> line
>>>>>
>>>>>> 1383
>>>>>>>>
>>>>>>>>> in
>>>>>>>>>
>>>>>>>>>> H5Screate_simple(): unable to register dataspace ID
>>>>>>>>>>> major: Object atom
>>>>>>>>>>> minor: Unable to register new atom
>>>>>>>>>>> #001: /users/fristed/git/minc-toolkit/build/HDF5/src/H5I.c
>>>>>>>>>>>
>>>>>>>>>> line
>>>>>
>>>>>> 902
>>>>>>>
>>>>>>>> in
>>>>>>>>>
>>>>>>>>>> H5I_register(): no IDs available in type
>>>>>>>>>>> major: Object atom
>>>>>>>>>>> minor: Out of IDs for group
>>>>>>>>>>> (from miicv_get): Should not happen!
>>>>>>>>>>>
>>>>>>>>>>> Setting -debug flag does not provide further information.
>>>>>>>>>>>
>>>>>>>>>>> Ideas for debugging this are very welcome.
>>>>>>>>>>>
>>>>>>>>>>
>
> --
> Best regards,
>
> Vladimir S. FONOV ~ vladimir.fonov <at> gmail.com
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>
More information about the MINC-users
mailing list