From: Antonin Bouchez
Sent: Friday, January 07, 2005 4:05 PM
To: David Le Mignant; Randy Campbell
Cc: Al Conrad
Subject: NIRC2 overhead - a new paradigm!

Guys,

Here is a more complete set of results of timing NIRC2 integrations (time
between successive frames in a "goi N", in seconds):

subc   1024         512          256
samp
2      17.8200      12.7600      11.9100
4      19.1700      12.9200      11.9000
8      19.0200      13.0400      11.4900

And here is what I believe is going on.  The total time spent integrating
and reading out the detector (not including the time to transfer data
around) should be: tread = tint + (0.5 * samp * 0.181 * (subc/1024)^2)

If I subtract this from each measured time/frame, assume that there is a
fixed-length overhead A associated with gathering and writing the FITS
header data, plus extra overhead per pixel B due to opening the file
and writing the data, I get: A = 1.24 seconds per frame B = 6.68e-6
seconds per pixel

The final formula for total time per frame is:
time/frame = tint + (0.5 * samp * 0.181 * (subc/1024)^2) + A + (B * subc^2)

The time/frame predicted by this model are:
      18.4207      13.0352      11.7029
      18.6017      13.0804      11.7146
      18.9637      13.1709      11.7379

There is clearly some variation in the data not predicted by the model,
maybe due to network traffic affecting the file write time.

Antonin