Segclassrecontruct issues

[Guendalina Marini]

Hello Carsten!

I'm Guendalina and I'm very new to helical reconstruction so I'm trying the SPRING tutorial in parallel with my dataset. I'm using the very last version of SPRING installed on my 24-core-machine from the package you provide in the website. Everything seems to work fine until I get to the 3D expert panel and I start segclassreconstruct. Here I face two different problems for the two different datasets.

1. I cannot use any parallelization (MPI option with CPU number) in the tutorial dataset (tobacco virus). The reconstruction works only without MPI (one CPU only)...and of course is painful slow. If I try to force parallelization with any number of CPUs the process either crashes at 10% without any error (I see CPUs dying using the command "top" in the terminal) or it gets into a loop at the same 10% (the "top" command shows CPUs working) without giving any result nor error. I attachment you can see that the folder contains no Vol1.hdf file but has a parfile.par that I don't usually get during the reconstruction without MPI.
2. In the analysis of my dataset I try to keep the default parameters or use some of the parameters of the tutorial because I know they should work but in this case segclassrecontruct fails immediately (1%) with an error (also in the report.log attached): 

RuntimeError: ImageWriteException at /home/eman64/EMAN2/src/eman2/libEM/hdfio2.cpp:578: error with '/mnt/projects/Gln1-HR/segmentclass_24_Sep_2015_13_36_29_19877/AveragesHeight600.hdf': 'Image does not exist' caught

The hdf file does exist:

$ ls -l /mnt/projects/Gln1-HR/segmentclass_24_Sep_2015_13_36_29_19877/AveragesHeight600.hdf

-rw-r--r-- 1 marini piginolab 3526000 Sep 24 13:41 /mnt/projects/Gln1-HR/segmentclass_24_Sep_2015_13_36_29_19877/AveragesHeight600.hdf

What we don't have is the path: /home/eman64/EMAN2/src/eman2/libEM/hdfio2.cpp:578

Can you kindly help me? I'm stuck in both analysis and I really don't know how to overcome these issues!

Thank you very much in advance...looking forward to your response,

Gwen

~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~

Guendalina Marini, PhD student

Pigino Lab, MPI-CBG

Pfotenhauerstr. 108

01307 Dresden

mar...@mpi-cbg.de

Lab: +49 351 210 2527

[Carsten Sachse]

Dear Guendalina,

Thank you for your interest in our SPRING package. I will answer the questions below.

On Thursday, October 1, 2015 at 3:41:39 PM UTC+2, Guendalina Marini wrote:

Hello Carsten!

I'm Guendalina and I'm very new to helical reconstruction so I'm trying the SPRING tutorial in parallel with my dataset. I'm using the very last version of SPRING installed on my 24-core-machine from the package you provide in the website. Everything seems to work fine until I get to the 3D expert panel and I start segclassreconstruct. Here I face two different problems for the two different datasets.

1. I cannot use any parallelization (MPI option with CPU number) in the tutorial dataset (tobacco virus). The reconstruction works only without MPI (one CPU only)...and of course is painful slow. If I try to force parallelization with any number of CPUs the process either crashes at 10% without any error (I see CPUs dying using the command "top" in the terminal) or it gets into a loop at the same 10% (the "top" command shows CPUs working) without giving any result nor error. I attachment you can see that the folder contains no Vol1.hdf file but has a parfile.par that I don't usually get during the reconstruction without MPI.

According to your provided parameter file the MPI works fine. You may not see at as the activity of the SPRING GUI but the seglassreconstruct_mpi process. Presumably, the 24 cores are too few for your requested grid size. People have encountered this before. I recommend making the size of the grid significantly smaller to see progress faster. For example you can simply change: 

Range of helical rise or pitch search in Angstrom = (23.0, 23.0)

Range of helical rotation in degrees or number of units per turn search = (15.0, 19.0)

Increments of helical symmetry steps in Angstrom or degrees = (0.0, 0.01)

This way you will scan along the pitch of 23 Å, which will still give you what you want. 

1. In the analysis of my dataset I try to keep the default parameters or use some of the parameters of the tutorial because I know they should work but in this case segclassrecontruct fails immediately (1%) with an error (also in the report.log attached): 

RuntimeError: ImageWriteException at /home/eman64/EMAN2/src/eman2/libEM/hdfio2.cpp:578: error with '/mnt/projects/Gln1-HR/segmentclass_24_Sep_2015_13_36_29_19877/AveragesHeight600.hdf': 'Image does not exist' caught

The hdf file does exist:

$ ls -l /mnt/projects/Gln1-HR/segmentclass_24_Sep_2015_13_36_29_19877/AveragesHeight600.hdf

-rw-r--r-- 1 marini piginolab 3526000 Sep 24 13:41 /mnt/projects/Gln1-HR/segmentclass_24_Sep_2015_13_36_29_19877/AveragesHeight600.hdf

What we don't have is the path: /home/eman64/EMAN2/src/eman2/libEM/hdfio2.cpp:578

Yes, the file exists - are you trying to reconstruct class number 3? Is it possible you don’t have this image on the stack? Remember, the stack starts with 0, i.e. an image stack with 3 classes needs to addressed by 0, 1, 2. Is this the issue?

Best wishes,

Carsten