Re: AW: Corrupt dcd files

From: Marc Baaden (baaden_at_smplinux.de)
Date: Sat May 14 2011 - 16:23:30 CDT

Hi Norman,

Thanks for your interest and suggestions. I had some off-list exchange
with Jim Phillips, and it seems the corrupt trajectories cannot be
recovered easily.

>> what do you consider as a corrupt dcd file, what are the problems you
>> are faced with while working with the file?

I consider as corrupt dcd files those that cannot be read by tools such
as VMD or catdcd. Usually only a fraction of the frames are read from
the beginning until an error occurs. The error that is returned is:

"dcdplugin) read_dcdstep: corruption or unrecognized file structure"

with both VMD and catdcd.

>> Maybe you could try to write a little script in scilab or matlab for
>> example, because there are functions to read binary files and the
>> file description of binary dcd is surely shown somewhere in the net.
>> You could also try to use the flipdcd binary which is shipped with
>> namd, maybe the tool is able to recognize such things, or you can use
>> the source to change it to your needs.

Exactly, only I was hoping someboby may have written such a script
already. With Gromacs for example there is an equivalent script, and it
skips over corrupt frames until it can recover intact frames again. Now
in my particular case it was simpler to re-run the simulations than try
to develop a new tool from scratch.

>> Another question, what kind of filesystem do you use to save your dcd
>> and outfiles while namd is running? Is it a nfs source or a local
>> disk?

It was on a local filesystem. But actually I have the feeling there is
a general issue with the node the simulations were running on. 4
independent runs, all on the same node, produced all corrupt files! I
have meanwhile re-run 3 of the 4 simulations on another node and don't
have any problem. I also use a slightly newer NAMD version as suggested
by Jim Phillips, which should provide better error checking.

Best regards,
  Marc

-- 
 Dr. Marc Baaden  - Institut de Biologie Physico-Chimique, Paris
 mailto:baaden_at_smplinux.de      -      http://www.baaden.ibpc.fr
 FAX: +33 15841 5026  -  Tel: +33 15841 5176  ou  +33 609 843217

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:20:16 CST