| T O P I C R E V I E W |
| Dichter |
Posted - Nov 04 2006 : 1:07:45 PM
I was trying to save my FSL model specifications and could not proceed because of some sort of memory issues. Here the error message I received. Any suggestions? Thank you.
error writing "file9": no space left on device
error writing "file9": no space left on device
while executing
"puts $channel "
# FEAT version number
set fmri(version) $fmri(version)
# Analysis level
# 1 : First-level analysis
# 2 : Higher-level analysis
set fm..."
(procedure "feat5:write" line 19)
invoked from within
"feat5:write $w 1 0 0 $fmri(feat_filename) "
(procedure "feat5:setup_model_preview" line 6)
invoked from within
"feat5:setup_model_preview .r"
invoked from within
".wdialog1.btns.b.view invoke"
("uplevel" body line 1)
invoked from within
"uplevel #0 [list $w invoke]"
(procedure "tk::ButtonUp" line 22)
invoked from within
"tk::ButtonUp .wdialog1.btns.b.view"
(command bound to event) |
| 15 L A T E S T R E P L I E S (Newest First) |
| vinod |
Posted - Aug 22 2007 : 1:00:50 PM
I am quite sure it has to do something with your design matrix. It is mostly the case that FSL does not like something about the ordering of events in these particular cases or something. Look at the design_cov.gif files from the output feat folders for these runs, especially the effect required values below and make sure they are not very large numbers. And also make sure that the only white elements are along the disagonal and every other square is black.
The request index = 1 seems to suggest that it not able to find something related to the first element of an array. So, it could just be some empty three-column files or something wrong about the filenames for the inputs or something. |
| Dichter |
Posted - Aug 22 2007 : 09:40:17 AM
A number of us continue to run into this FILM memory error, and it seems to be specific to certain runs only (ie, I get the error repeatedly for a particular run, but not other runs). This is the only job on my machine, and I am running post-stats only, so I doubt this is a truly a memory issue. 3 column format files are fine as well.
The full error is:
An exception has been thrown
Logic error:- detected by Newmat: index error: requested index = 1
MatrixType = Rect # Rows = 0; # Cols = 1
Trace: Histogram::generate; AutoCorrEstimator::spatiallySmooth.
Error: FILM did not complete - it probably ran out of memory.
Thanks for any help,
-Gabriel |
| syam.gadde |
Posted - Jun 11 2007 : 09:54:13 AM
The command-line programs in the report.log file are designed to do the particular analysis and pre-processing you have specified in the .fsf file. It would be much more difficult for anyone to create a template report.log file that is generic enough to accommodate what anyone might want to do. I think from the perspective of creating something that nearly anyone can use, I think the template .fsf file is the right approach (it is also much more compact and easier to read!) But if you already have a successful run using your particular design choices, I agree with Dharol and Chris that you would probably be able to use portions of the report.log file to restart another (failed) analysis (but I haven't tried this myself). |
| vinod |
Posted - Jun 11 2007 : 09:32:38 AM
Yes. It looks like it creates a whole bunch of prefiltered_func* files for MC, ST and so on. Because of a specification in the script, these files get deleted once all preprocessing steps are complete and filtered_func_data.nii.gz file is created. However, if the program crashed, I suspect some of these files should still be found in the directory and we can use the last completed step to restart analysis. So, for example, if Goldman crashed when it was running slice timing for a subject, we should see a file that is named prefiltered_func_data_mcf.nii.gz which can be input to the slice timer step again. And we can run all steps below that in the report.log file at the command line or a bash script.
Petty, given that you can pull out lines from log file and run it on command line, would it be better to script FSL using a template report.log instead of the currently popular way of doing so from template.fsf files? In other words, we save a report.log file as an executable bash script? And then use "sed" to replace subject names and directories? This way, we might even be able to choose which steps we wants to rerun if analysis crashes? I may be missing something.. |
| tankersley |
Posted - Jun 11 2007 : 09:04:03 AM
Hi Vinod.
Logically what you say is sound; What i have not been able to figure out is which output files from crashed analysis to input into the new command line when starting in the middle.
If your analysis got through basic preprocessing steps (i.e., if you ended up with the file filtered_func_data.nii.gz, it's pretty easy to start the analysis from this point.
Instructions here http://fourier.biac.duke.edu/wiki/doku.php/huettel:rerunning_fsl |
| petty |
Posted - Jun 11 2007 : 08:57:06 AM
yes, if you pull out the lines that you are interested in from a log file, you should be able to directly run those from the command line. |
| vinod |
Posted - Jun 11 2007 : 07:20:49 AM
quote:
I'm not aware of any way to restart FSL in the middle.
Given that FSL is basically executing instructions one by one at command line and saving all these commands to report.log file, is it possible to just save a smaller subset of these commands and run them as a shell script? This would mean restarting FSL in the middle? Let me know if this would work? |
| tankersley |
Posted - Dec 01 2006 : 12:44:00 PM
FYI, the error
Unable to open design.mat
Error: FILM did not complete - it probably ran out of memory
was due to some of the 3 column format files having commas that they weren't supposed to have.
Dharol
|
| tankersley |
Posted - Nov 30 2006 : 12:31:30 PM
The FILM error has been happening reliably lately. About two weeks ago it started consistently crashing at this place and giving me this error; before that, it was completing a good percentage of the time (I think then it really was probably memory problems).
FEAT is supposed to create
design.mat, design.con, design.frf, design.fsf, design.gif, design.ppm, and design.trg
Now when it's crashing and gives me this error message, it has only created design.con and design.fsf.
I took a look at the design.fsf for a run that completed and one that did not, and there do not appear to be any differences; the log report doesn't have any extra info either . . .?
Thanks,
Dharol |
| syam.gadde |
Posted - Nov 29 2006 : 3:45:10 PM
Does the FILM error happen consistently for certain analyses, or does it run successfully if you keep re-running it? Disk space should not be a problem anymore since /data/users has a lot of space. It's also hard to imagine that you are running out of memory, but perhaps that's what's happening -- does your data have a large number of timepoints?
I'm not aware of any way to restart FSL in the middle.
The report is complaining about not finding design.mat. Does that file exist in the .feat directories in which you see this error? Perhaps that's why FILM is exiting with an error... |
| tankersley |
Posted - Nov 29 2006 : 2:54:20 PM
I'm a bit unclear on the latest updates to FSL and their implications, but am still having memory problems.
I cleaned out my users/tmp directory and am only running 3 FSL jobs, but am still geting the
Unable to open design.mat
Error: FILM did not complete - it probably ran out of memory
/mnt/katz/data/ExecInfo.09/Data/FSL/31385/run5
Error: FILM did not complete - it probably ran out of memory
Is there some new thing I'm not doing, or should we stick to running 1 job at a time?
Also, is there an easy way to pick up analysis wherever it stalled instead of starting over?
Thanks,
Dharol
|
| syam.gadde |
Posted - Nov 16 2006 : 1:57:54 PM
Dharol,
This is due to /tmp filling up, not your data directories.
I have installed a new version of FSL that now respects the TMPDIR variable but have not yet made it easy to get TMPDIR set. I will post again soon how to get this working. In the meantime I have removed some files from /tmp. |
| tankersley |
Posted - Nov 16 2006 : 1:10:01 PM
I have been trying to run 3rd level FEAT analysis on Reward.01 data for over a week and always get the message
no space left on disk
I have deleted some files and continue to get this message. Is there an alternative place to GOLGI to run these analyses or some other fix?
Dharol |
| Dichter |
Posted - Nov 06 2006 : 11:08:19 AM
Sorry - disk problems. |
| syam.gadde |
Posted - Nov 06 2006 : 11:07:28 AM
quote:
Originally posted by Dichter
Thanks a lot, Syam. No this is a new study, so it was not exceeding an experiment quota.
FYI, we were also unable to run bputil on Friday because of memory problems.
Memory problems? Or disk problems? I don't think bputil should encounter memory problems unless your runs have a large number of timepoints. |
