CSS slows down simple 2->2 jet production by large amount

[Jan Winter]

As suggested by FrankS, I put in my findings so that we can at least discuss and are aware of this issue:

While working on predictions for Lambda_b production @ 7 TeV LHC with Sherpa, I found that a simple 2->2 jet production setup goes from generating 1000000 weighted ME-level events in 3 1/2 minutes (on a Intel(R) Pentium(R) D CPU 3.40GHz) to 3 1/2 hours when inlcuding the parton showering. This is with no merging whatsoever, the run card is attached as well as the output from monitoring with valgrind --tool=callgrind (use kcachegrind <..> to visualize) and a little picture showing that the CSS routines dealing with overestimating the shower kernel take up most of the computing time.

Here is an overview of approximate computing times for the considered setup (jj->bbar,massive bs) and 1000000 weighted events: only MEs: 3:30 mins incl showers: 3:30 hours incl fragmentation: 4:50 hours incl hadron decays: 7:40 hours

[Stefan Hoeche]

unless anybody vounteers to speed up the shower, i would opt for closing this ticket. looks a bit awkward to have that as an open issue on the bug tracker; after all, this is probably the price to pay for initial state evolution.