LAP Scheduling (2011-08-22 draft)

The colored stripes on this image represent the 2-hour bands in RA that the LAP processing will be broken into. In order to accomplish this, once we are in a steady state, we will need to complete a single filter in a given band within six days. This will require ~30-40 projection cells to be queued (and presumably completed) each day. We are currently (2011-08-22) working on the band from 16-18 hours (240-270 degrees) although we have not yet ramped up to a satisfactory processing rate.

The reason that we will do each filter across an entire band instead of completing five-filter data on individual projection cells is that we are attempting to make the best use of exposure sharing to minimize the processing time. This also reduces the number of exposures to keep on disk at any given time, as we are cleaning processed data as we go according to the following process:

• Block of runs are queued for a set of projection cells (manually at this point, automatically at some point in the near future)
• Exposures are processed through to the final stacking stage.
• When all final stacks have completed for a run, the exposures that comprise that run are marked as inactive.
• A cleanup task operating on completed runs identifies exposures that are marked inactive for all runs they contribute to, and sets the data (chip/warp/diff/destreak/quickstack) from that exposure to be cleaned. This stage is also running manually, although there is code in place to have the cleanup be automatic.

We have not yet defined the number of LAP runs to keep active during the reprocessing, as this is mainly an issue of the disk usage available to devote to LAP. If this number is set high enough, it may be possible to run all five filters simultaneously and still have acceptable sharing of exposures.

I have appended an example list of commands that will queue LAP across the entire sky.