[daip] IMAGR errors on Linux box

[Eric Greisen]

Lynn D. Matthews writes:
 > I have begun to experience problems with my Linux machine running Fedora
 > Core that I am fairly convinced are caused by a hardware problem.
 > However, I am writing with the hope that the symptoms exhibited by AIPS
 > might yield some clue as to the origin of the problem, since so far I've
 > not been able to trigger them outside of AIPS.
 > 
 > The most dramatic problems occur when running IMAGR. Each time I do a
 > "tget" and rerun the program, I get different results and different
 > messages, including things like:
 > 
 > IMAGR1: UVWAIT: Sum of weights in  2.083E+32 and out Infinity
 > 
 > IMAGR1: Field 1 Beam min = NaN Jy, max = NaN Jy
 > 
 > IMAGR1: BEAM HISTOGRAM PATHOLOGICAL - DEALING WITH IT
 > 
 > IMAGR1: Field   1 min =  NaN        Jy,max =  NaN        Jy
 > IMAGR1: Restoring Clean components
 > IMAGR1: Checking image max/min
 > IMAGR1: Field    1 final Clean flux   131.076      Jy
 > 
 > Sometimes there will be no obvious complaints, but I might end up with
 > image headers like this:
 > 
 > AIPS 1: Minimum=-2.78580113E+38      Maximum= 2.78580133E+38 JY/BEAM
 > 
 > Any idea whether this is likely to indicate a memory problem? A disk
 > problem? Checks we've done so far on both have been unable to pinpoint the
 > source.
 > 
 > uname -a shows:
 > Linux wrest 2.6.19-1.2895.fc6xen #1 SMP Wed Jan 10 19:47:12 EST 2007 i686
 > i686 i386 GNU/Linux
 > 

This reminds me of a page fault bug that we had in RedHat about 7
years ago.  The program would go on computing before the new page was
rolled all the way in.  RedHat congratulated themselves that they
found the problem before anyone had encountered it - but we were at
the time studying issues with IMAGR and FRING.  AIPS beats on
computers hard enough that timing problems in disk I/O and in paging
will be exposed.  Are any of the disks mounted over NFS?

I will forward this to a few other folks who may remember other
problems or know of Fedora issues.  Note that the error I am
remembering was a software issue in the o/S rather than a pure
hardware issue.

Good luck,

Eric Greisen