[evlatests] Date Rate Record results

[Rick Perley]

    A week ago, Ken ran a short (35 minute) observation on 3C84, testing 
our performance boundaries.
    I append his request to me:

Rick,

I started trying to find the edges of performance space today.
Here is an example that is worth looking at enough to see that
the data is reasonable, i.e., all there and not zeros.

Recirculation by 32 was used, eight subband pairs covering 1 GHz,
and one second integrations.  This resulted in about 730 MB/sec
to the CBE and 360 MB/sec to Lustre.  So far as I could tell
both the CBE and Lustre could keep up.  The script ran for about
35 minutes from 2130 to 2205 UT this afternoon.  The CBE reported
no zeros.

Look for:     C_recirc_nocal_8sbx32_1.0

It is going to be a lot of data, so perhaps you should let it
fill overnight.

-----------------------------------------------------------

  The data were extracted from the archive and read into AIPS via BDF2AIPS.
The database size is about 700 GB -- it took 14 times longer to fill into AIPS
than the observation duration  (i.e. about 8 hours for 35 minutes observing).   
  There are a total of 4pol * 16 subbands * 2048 channels/pol = 131072 total
channels produced.  26 antennas were in the array.   With a factor of 32
recirculation, the spectral resolution is 2000/32 = 62.5 kHz.  

  BDF2AIPS reported that 4.8% of the data were removed due to pure zeros.  
I think this is within the normal range.  (Gurus should chime in here, if 
I'm wrong).  

  Needless to say, I couldn't look at ALL the data.  :)  Short listings showed
good quality data, at least at the beginning.  I ran a BPASS on a scan for which
all antennas were on source.  The resulting bandpass shapes are just marvelous.  
Ken chose the frequency selections to miss RFI.  All antennas were fringing 
strongly.  

  I looked at the switched power -- it's as good (for all antennas and subbands) 
as any I've seen.  (There were no band changes, so the problems associated with
that are avoided).  
 
  I'm attempting a 'TYAPL' to apply the switched power -- I want to see if the
resulting amplitudes are where they should be.  Should take all night, at least ...