I have implemented the changes and sent a test run to the cluster before I went to Hawaii (started on Nov13). It seems that calculation is not finished by Nov17, indicating some sort of problem.

The problem was that I have filled my user directory on tigress. I have cleaned my user directory (which had some old reduced data that should be in rerun directories, and many intermediate files from creation of poststamp images which could be deleted) and rerouted my outputs to 

scratch/gpfs

which is a) actually proper way to do it and b) has more space.

I have noticed that the code is slower than I expected, i.e., when I analyze 8 spots it is around 50% slower than analyzing 8 individual images. I am making a small profiling run to see what is happening. 

The slowdown on della was happening because I did not execute 

export OMP_NUM_THREADS=1

before running my code (see https://jira.lsstcorp.org/browse/DM-4714)

I implemented this code some time ago already. I will shortly describe the figures I attached here and which I described during one of the telecons.

You can see that the code has been correctly manipulated so that the parameters describing focal ratio degradation are same for all of the spots in the fiber.

`Improvment` measure that I use to determine how well the algorithm is doing. In general, the `improvment` shows the same features as in previous versions of the fitting procedure when I was fitting individual spots.

this figures shows an example of a single defocused spot. The quality is comparable to previous versions - but you can see some problems in modelling the illumination (e.g., struts). I will elaborate more on this in the follow-up tickets.