Chip Stage Timing

The majority of the processing time at chip stage is consumed by the photometry process (median dtime_photom = 0.64 * dtime_script). Most chips have a similar distribution of dtime_photom values, with medians less than 100 seconds. Ten of the chips show a different distribution, with median dtime_photom values larger than 100s and a large day-to-day variation in distribution shape. The plots below show selected chips, with the entire set of plots attached in pdf format.

The CDF plots show the distribution of dtime_photom values for each chip, separated by the date they were observed. Each curve is labeled with the median, non-parametric dispersion, minimum and maximum dtime_photom.

The second set of plots show for each date the fraction of total dtime_photom spent on jobs with dtime larger than the value on the x-axis. These curves are labeled with the median value (half of the time is spent on jobs with photometry times longer than this value), non-parametric sigma, and the sum of dtime_photom for this date/chip pair. This plot is somewhat skewed by a small number of points, but does show that even on good chips, the ~10% of jobs that run longer than 100s consume ~30% of the total photometry time.

The set of chips that can be considered "bad" (XY04, XY05, XY16, XY27, XY35, XY60, XY64, XY72, XY74, XY75) seem to share the same distribution of n_stars, fwhm, and sky background as the good chips. The photometry time does not correlate with any of these values, either.

	Good chip	Bad chip
CDF of dtime_photom values
Fraction of time spent running photometry