Previous page      Top page      Next page

Measuring speed-up

For every aspect change, except maybe ones that impact the whole model, I think we'll want to use Dr Hook to measure change.

Profile of our AMIP job

The numbers show total time, which is the time in each routine + the time for all the routines called by a given routine.

UM_SHELL (6,031s)
U_MODEL_4A (5,916s)
ATM_STEP_4A (5,841s)
Slow physics: ATMOS_PHYSICS1 (636s) Fast physics: ATMOS_ PHYSICS2 (356s) Pressure solver: EG_SL_ HELM- HOLTZ (342s) Advection & transport: SL_TRACER1_4A (324s), EG_CORRECT_ TRACERS_ PRIESTLEY (228s), TR_SET_ PHYS_4A (212s) & EG_SL_ FULL_WIND (137s).
Total of 901s
 ⇓  UKCA: UKCA_MAIN1 (1,880s, non-diagnostics ~650s)
Radiation: RAD_CTL (169s) Micro- physics: MICROPHYS_ CTL (146s) Gravity wave drag: NI_ GWD_ CTL (131s) Diagnostics: STASH (2,114s) Chemistry: UKCA_ CHEMISTRY_ CTL (207s) Photolysis: UKCA_ FASTJX (129s) Aerosol: UKCA_ AERO_ CTL (111s) Emissions: UKCA_ EMISS_ CTL (87s) Cloud nuclei: UKCA_ ACTIVATE (40s)
... ... Aerosol solver: UKCA_ COAG- WITHNUCL (36s)
UKCA_ RADAER_ BRAND_ AVERAGE (45s)     Chemistry solver: ASAD_ SPMJPDRIV (158s)