[daip] Debugging an intermittent IMAGR problem

[Robert Laing]

Dear AIPS group

We have a sporadic problem when running IMAGR on a Linux PIII PC. I
suspect this of being hardware-related (especially as the machine is new,
and similar IMAGR's on our existing PC's have worked fine), but would
value your advice in tracking down the problem, as our support people
cannot find any obvious disk or CPU errors.  The machine in question has a
substantially higher clock speed than the ones we have used previously
(900 MHz PIII as opposed to 350 and 450 MHz PIIs). It has 128 Mbytes of
memory.  At times, we get a crash on 30% of attempts.

As a test, we ran a sequence of 6 identical 2048 x 2048 IMAGR's. The UV
file has concatenated A, B and C configuration data (continuum, ~2E6
visrecs). Tries 1-4 and 6 worked fine. On the fifth execution, we got an
error:

IMAGR1 14:05:26 OUNFWT: Sorting data to make them fit
IMAGR1 14:08:13 UVWAIT: begin finding uniform weights
IMAGR1 14:08:15 UVWAIT SORT ERROR: ROWS   225   226 OUTSIDE  -209     3
IMAGR1 14:08:15 UVWAIT: ERROR WEIGHTING UV DATA
IMAGR1 14:08:15 OUNFWT: ERROR IN UNIFORM WEIGHTING OF UVdata work object
IMAGR1 14:08:15 Deleting UV work file:

In general, the errors are always of the same general type, in the UVWAIT
routine, as above or in GRIDUV at a later stage, e.g.


1    3   17-JAN-2001  18:15:24     IMAGR     ALGSUB: Ipol gridded model
subtraction, chans    1 through    2
   1    3   17-JAN-2001  18:15:50     IMAGR     ALGSUB: Ipol gridded model
subtraction, chans    1 through    2
   1    8   17-JAN-2001  18:16:14     IMAGR     GRIDUV: SORT ERROR: ROWS
57    66 OUTSIDE  -265    48
   1    7   17-JAN-2001  18:16:14     IMAGR     OUVIMG: MAKING IMAGE CLEAN
field number 001
   1    8   17-JAN-2001  18:16:14     IMAGR     OUVIMG: FROM UVdata work
object
   1    8   17-JAN-2001  18:16:14     IMAGR     CLNUV : ERROR CLEANING
CLEAN process object
 
We are running 31DEC99 (a snapshot obtained on 2000 Sept 5) under Red Hat
Linux 6.1.

We are in the process of running the same test on another 2 machines, one
configured identically to the offender and the other a lower-performance
PC which has run very similar IMAGRs many times in the past without
problems.

Any advice you could give in tracking this problem down would be
gratefully appreciated.

Regards

Robert Laing 

Space Science Department        Astrophysics
CLRC                            University of Oxford
Rutherford Appleton Laboratory  Nuclear and Particle Physics Laboratory
Chilton, Didcot,                Keble Road
Oxfordshire OX11 0QX            Oxford OX1 3RH

Tel: (+44) 1235 446401          Tel: (+44) 1865 273429
Sec:            445618                          273303
Fax:            445848                          273390

R.A.Laing at rl.ac.uk              rlaing at astro.ox.ac.uk