Use of FLUKA on LXBATCH - proper methodology for installation

Installation
1

Dear colleagues,
I am executing FLUKA simulations at CERN using the LXBATCH system. I am using the very nice lxflu-master tool that was developed by Fernando Perreira, Chris Theis, Alessio Mereghetti, Roberto Versaci, Markus Widorski and others.

To run FLUKA, I installed it on afs, in my work area (/afs/cern.ch/user/a/acelenta/work/apps/fluka/fluka4-3.3). However, this is creating some issues with data access. Some of the jobs that I execute exit with the error:

At line 568 of file lowneu/xsread.f (unit = 9, file = '/afs/cern.ch/user/a/acelenta/work/apps/fluka/fluka4-3.3/data/neuxsc.bin')
Fortran runtime error: Unformatted file structure has been corrupted

Error termination. Backtrace:
#0  0x1480e34237a2 in ???
#1  0x1480e3424269 in ???
#2  0x1480e3424e6f in ???
#3  0x1480e364edce in ???
#4  0x1480e364a0f1 in ???
#5  0x58ed7a in xsread_
	at lowneu/xsread.f:568
#6  0x581e56 in lowset_
	at lowneu/lowset.f:379
#7  0x4fc78a in flinit_
	at cascade/flinit.f:170
#8  0x46128f in flukam_
	at main/flukam.f:2781
#9  0x43d1b0 in fluka
	at main/fluka.f:77
#10  0x43d1b0 in main
	at /shared/src/usflmd.inc:15

I suppose this is due to some disk I/O error while launching multiple FLUKA runs that access this file simultaneously. May I ask you if you ever encountered this error? Is it correct to have FLUKA installed in afs? Maybe a common FLUKA installation path in cvmfs exists?

Thanks,
Bests,
Andrea

3

Hi Andrea,

there is no centralised installation of FLUKA on cvmfs due to license reasons.

I guess there is no problem in your installation, as it runs fine if you submit a few jobs only ?

Indeed we encounter I/O errors from time to time on EOS or AFS (usually less on AFS) when we run large number of jobs, but not systematically and not reproducibly. I had been several times in contact with the CERN IT support, but they were unable to solve the problem or to help in the analysis. It is unclear to me how read access to theses files in particular should lead to the corruption errors. That’s for the filesystem specialists to tell, I suppose. But as said, I was unable to get an answer on that so far. I don’t know if FLUKA does something fancy that AFS/EOS is not able to cope with. My conclusion was that it is not linked to FLUKA, but the file system.

You may try and open a ticket with the AFS support and hopefully you are more successful.

Cheers,

Markus

4

Hi Markus,
thanks for the reply. Indeed, no errors are observed if few jobs are submitted, pointing to FLUKA-unrelated issue, possibly connected to I/O on eos or afs. My point was mostly regarding the existence of a common installation on cvmfs, but from your answer this does not exist.

I’ll contact CERN support to check what is going on, and eventually obtain from them a suggestion. If relevant, I’ll come back to the forum with an update.

Cheers,
Andrea

5

Dear colleagues,
I report here the solution from CERN support in case it may be relevant for others.

The solution was to move FLUKA installation to EOS, while keeping lxflu-master tool on AFS.

I tested this, executing 1k jobs. All of them were successful.

6

Thank you, Andrea, for following up with. Just to note: we had been using EOS for the FLUKA installation for some time - and moved back at some point to AFS, as it was more stable and produced less I/O errors. But maybe that has changed now.