Event generation after reading integration results very inefficient

[Frank Siegert]

Event generation takes significantly longer if previously stored integration results are read-in instead of integrated anew. I am attaching a run card and two corresponding log files, one for event generation directly after integration (4829 s / 5000 events) and one for event generation after reading in the results (14024 s / 5000 events).

The setup is:

• Z+0,1,2,3,4 jets, I can't see a similarly drastic difference for Z+0,1,2,3 jets and if I remember correctly also not for exclusive Z+4 jets (but that one I am trying to reproduce currently).
• Unweighted events (I'm currently running a test whether the same happens for weighted events)
• Enhance factors for the higher multis (that's why the events take so long in first place)
• Comix with multi-threading (in both runs)

It might be interesting to note the development of the time estimate in the read-in log, as there seems to be some kind of optimisation happening in the beginning, which doesn't seem to be present in the run with integration?