Hi
Present were: Bhavesh, Elo, Erik, Frank, Gwyneth, Ian H., Peter, Roberto, Steven, and Vassili
Status on KNLs Some independently developed optionlists exist, but none are production-ready: - Erik's optionlist on Cori: in branch in simfactory2 - Frank will put his option list into simfactory. - Eloisa has one for Marconi + KNL, and will also put it in simfactory.
Test suites and Jenkins: - Erik: optimizations on "jenkins" use optimization, performance difference ~8% - discussion about - whether to use optimizations or not / whether 8% are ok to loose - whether optimizations "dangerous" or not. - result for now: try to see if testsuites are broken and fix, if possible. - Eloisa will look at boostedpuncture test case - Roland will look at Spec test suites - Frank &| Roland will have a look at the hydro failure
Piraha CST parsing (Stev): - caching parse-tree cuts down time practically to 0 - files holding cache (one per ccl file) will be within configurations - plan: new ticket, branch, eventually inclusion - parallelism problematic with Perl, but may not be needed anyway
Gwyneth asked: - Constraints: what/how to use to calculate? - ML_AMD_Contstraints: uses ADM constraints, independent of evolution system - McLachlan: uses BSSN variables, no need to calculate ADM variables result: It does (and should) not really matter. Preferences for one or the other exist, and may depend on particular simulation specifics, but both should get the results to within numerical accuracy. - What thorn to use for which output type (basic/scalar/ascii) (Frank: It would probably be good to have this as overview on a page or so.) - Question about comment in bbh gallery example par file: "SummationByParts::sbp_upwind_deriv": not required for that parameter file would drop order of derivatives near boundary in some cases, but derivatives are not used here - ADMMass: how to measure - using initial data output (but that might differ from the one on the finite comp. grid) - ADMMass thorn: surface integral or - ADMMass thorn: volume integral (but needs thorn not in the toolkit to excise BHs) or - QuasiLocalMeasures: surface integral - time refinement factors on coarse grids - Stricter requirements on time step on very coarse grids than regular cfl factor may lead to instabilities (gauge evolution in particular). - Result: don't need to use if no instability seen.
Frank