[daip] FRING memory requirements

[Ingyin Zaw]

On Thu, 7 May 2009, Eric Greisen wrote:

> Ingyin Zaw wrote:
>>  Hello,
>>
>>  I checked setapmax. It was set to 528 Mbytes. I changed it to the maximum
>>  allowed:
>> >  setmaxap(0)
>>  AIPS 1: SETMAXAP: current max dynamic AP is  6144.000 Mbytes
>>  AIPS 1: SETMAXAP: allowed range is  20 to  6144 Mbytes
>>
>>  I still get the same memory error as below.
>>
>>  The fring parameters I've set (which are not defaults) are:
>>  calsour '2145+067'
>>  refant 6
>>  solint 0
>>  aparm 3 0 0 0 2 0
>>  dparm 3 1 1 4.1 0
>>
>>  (docalib -1; flagver 0; doband -1).
>>
>>  Thanks,
>>  Ingyin
>>
>>
>>
>>  On Wed, 6 May 2009, Eric Greisen wrote:
>> 
>> >  Ingyin Zaw wrote:
>> > >   Hello,
>> > > 
>> > >  I'm trying to run FRING with aparm(5)=2 to solve for multiband and 
>> > >  single
>> > >  band delays on a file with 256 channels and 8 IFs (experiment BK114A). 
>> > >  I
>> > >   get the following error.
>> > > 
>> > > FRING1:   Task FRING  (release of 31DEC09) begins
>> > > FRING1:   You are using a non-standard program
>> > > FRING1:   Doing no flagging this time
>> > > FRING1:   Selecting the data
>> > > FRING1:   Dividing data by source flux densities
>> > > FRING1:   Determining solutions
>> > > FRING1:   Doing Least Squares fits for multi- and single-band delays
>> > > FRING1:   Writing SN table    1
>> > > FRING1:   Time=   0/ 02 28 14, Polarization = 1
>> > > FRING1:   QINIT: did a GET  of      5120 Kwords, OFF   11963214967809
>> > > FRING1:   FRNSRC: MEMORY TOO SMALL FOR SPECIFIED FFT SEARCH
>> > > FRING1:   REDUCE DELAY AND/OR RATE WINDOW OR AVERAGE IN
>> > > FRING1:   FREQUENCY OR USE A SHORTER SOLINT
>> > > FRING1:   Purports to die of UNNATURAL causes
>> > > FRING1:   rglinux9 31DEC09 TST: Cpu=      0.9  Real=      1 
>> > >  IO=        86
>> > >   AIPS 1: Resumes
>> > >   AIPS 1: RETURN CODE      2 RECEIVED: STOPPING
>> > > 
>> > >   I have set dparm 3 1 1 4.1 0. It seems strange that it fails for such 
>> > >   a
>> > >   small search window. It fails even when the search window is set to 
>> > >   0.1
>> > >   nsec and 0.1 mHz. I'm not applying any calibrations or bandpass.
>> > > 
>> > >   I'm running 31 DEC 09, 64 bit AIPS on a machine with following
>> > >   specifications:
>> > >   Linux rglinux9.cfa.harvard.edu 2.6.18-128.1.1.el5 #1 SMP Mon Jan 26
>> > >   13:58:24 EST 2009 x86_64 x86_64 x86_64 GNU/Linux
>> > > 
>> > >   I have seen the same problem on 31 DEC 09, 32 bit AIPS as well on an
>> > >   experiment with 512 channels and 8 IFs (BB242D).
>> > > 
>> > >  I have made a FITS file of only the source I used for FRING for 
>> > >  testing.
>> > >   It's available by ftp from cfa-ftp.harvard.edu,
>> > >   User = anonymous, password = <youremailaddress>.
>> > >   The file is located at: incoming/lincoln/BK114A_2145.FITS
>> > > 
>> > >  Please let me know what I can do to be able to run FRING with 
>> > >  aparm(5)=2.
>> > 
>> >  Needless to say FRING should not do this - I note that it only asked for 
>> >  20 Megabytes of AP memory.  It is supposed to ask for what it actually 
>> >  needs if you allow it to do so.  Have you run SETMAXAP with a small 
>> >  limit?  I would recommend set your max AP to 0.5 or 1 Gbyte on your 
>> >  64-bit computer.
>> > 
>> >  If that does not work, send me your inputs and we will fetch the file 
>> >  and look at the problem.
>> > 
>> >  Eric Greisen
>> > 
>> > 
> You have asked for an FFT on the frequency axis of length 131072 or perhaps 
> it is 524288 - I do not know the details enough to be sure.
> AIPS now does FFTs up to 32768 and we have no plans to allow larger in the 
> near term.  I will repair the error message which now is only about allowed 
> size of FFTs (it used to mean what it says) and repair the limit in FRING 
> which is currently lower than it has to be.  But that will not allow you to 
> do what you are trying to do.  There are aips memos which describe what you 
> should do to accomplish what you need to accomplish.
I tried to reproduce the 131072 number and come up short. I have an 
experiment (BK114A, 256 channels, 8 IFs) with 500 MHz total bandwidth with
~ 60 kHz channels which would seem, at most, to have a frequency axis of 
length 16384. Even for 512 channels and 8 IFs, I get a frequency axis of 
length 32768, which is within AIPS capabilities. Could you point out where 
my calculations are different from yours.

> See the aips web page:
>   http://www.aips.nrao.edu/
> and read AIPS Memo 110.
>
> You will need to run FRING an IF at a time effectively and then solve for 
> multi-band delay with MBDLY.
Thanks for pointing me to the memo. It says that FRING with aparm(5)=2 is 
better than aparm(5)=0 + MBDLY when sources are weaker. We observe at 22 
GHz and our calibrators are relatively weaker, so we'd like the option of 
running with aparm(5)=2, if possible.

> Eric Greisen
>
Thanks,
Ingyin