[daip] FRING on large (25Gb) line dataset problem
Eric Greisen
egreisen at nrao.edu
Thu Jan 27 10:36:21 EST 2011
Kazi Rygl wrote:
> Dear Mark Greisen,
>
> I encountered a problem with FRING when I run it on a large (25Gb) VLBI
> line dataset.
> On smaller datasets it works fine. Also the problem is only there with
> the TST version, not in the NEW version. I tried decreasing the solint
> parameter but it didn't help (see suggestions of AIPS output below).
> Setmaxap says:
> >setmaxap(0)
> AIPS 2: SETMAXAP: current max dynamic AP is 528.000 Mbytes
> AIPS 2: SETMAXAP: allowed range is 20 to 1536 Mbytes
>
> Since I'm not sure why FRING crashes (local problems, or due to
> something in the program) I thought I just report it. Last week I could
> run FRING with similar inputs without any problems.
> I paste the output of AIPS below together with my inputs.
>
> Best regards,
> Kazi
>
>
>
>
> FRING2: Task FRING (release of 31DEC11) begins
> AIPS 2: Resumes
> >FRING2: Set integration time to 1.000000 seconds, hope that is ok
> FRING2: UVGET: Using flag table version 1 to edit data
> FRING2: Selecting, editing and calibrating the data
> FRING2: Dividing data by source flux densities
> FRING2: Determining solutions
> FRING2: Writing SN table 3
> FRING2: Time= 0/ 20 58 30, Polarization = 1
> FRING2: QINIT: did a GET of 135168 Kwords, OFF -328032033
> FRING2: DID NOT GET WHAT I NEED FOR MEMORY
> FRING2: USE SETMAXAP TO RAISE THE LIMIT, OR
> FRING2: FRNALL: DYNAMIC MEMORY AP FAILS
> FRING2: FRNALL: MEMORY TOO SMALL FOR SPECIFIED FFT SEARCH
> FRING2: REDUCE DELAY AND/OR RATE WINDOW OR AVERAGE IN
> FRING2: FREQUENCY OR USE A SHORTER SOLINT
> FRING2: Purports to die of UNNATURAL causes
> FRING2: mcn3 31DEC11 TST: Cpu= 18.9 Real= 35 IO= 24974
>
> task 'fring'
> inclass 'linec'
> anten 0
> timera 0 20 57 01 0 20 59 59
> cals 'J2038+5119'
> docal 2
> snver 0
> flagve 1
> bif 0
> eif 0
> aparm 0
> dparm 0
> aparm(1) 2
> aparm(9) 1
> dparm(8) 1
> solint 5
> go;wait
FRING uses a large amount of dynamic memory both in the "pseudo-AP' and
outside it. The pseudo AP is limited by the parameter which you may
change with SETMAXAP. Try HELP SETMAXAP to show the considerations.
Recently, it was discovered that FRING asked for whatever it needs based
on rate and delay windows, SOLINT, data integration time, etc. But
FRING did not test that it was given what it expected and went ahead as
if all was well. When it wanted more than SETMAXAP allowed it was given
the max - 2 Gbytes (0.5 Giga words) in your case. I added tests in
FRING to make sure it got what it needed so that it would not wander off
in dynamic memory writing over memory used for other purposes or even
memory not allowed to the task. ABORTs for such issues come later and
are very confusing.
Try SETMAXAP 1024 to change the allowed upper limit and rerun your FRING.
Eric Greisen
More information about the Daip
mailing list