Closed AHorneffer closed 7 years ago
I guess I can answer my own question: It is important that no facetselfcal
operation reads from the input data files while a facetsub
operation is going on.
And because the predict_and_difference_models
step reads from the input data files, this would be a problem...
Do I see it correctly that even in groups with several directions in it, the
facetsub
operations are only started after allfacetselfcal
operations have finished? Couldn't the firstfacetsub
operation start as soon as the firstfacetselfcal
operation finished successfully?I can also see that it probably won't be trivial to implement that, but would there be any other problems except having to re-write a good portion of Factor?
P.S. I usually run Factor on a single workstation, so it is not important to me. P.P.S. In my recent run the
facetselfcal
took between 6 and 23 hours per facet and thefacetsub
took about 1 hour per facet. So the gain wouldn't be huge, but still significant.