Speeding up GEOS-Chem

From Geos-chem
Jump to: navigation, search

Previous | Next | Guide to GEOS-Chem performance

  1. Parallelizing GEOS-Chem
  2. GEOS-Chem 7-day timing tests
  3. GEOS-Chem scalability
  4. GEOS-Chem 1-month benchmark timing results
  5. Profiling GEOS-Chem with the TAU performance system
  6. Speeding up GEOS-Chem


On this page we will describe some recommended methods for speeding up GEOS-Chem simulations.

Overview

GEOS-Chem performance is continuously monitored by the GEOS-Chem Support Team and user community via official benchmarks and unofficial timing tests. It has been shown that running GEOS-Chem with the recommended timesteps from Philip et al. (2016) can increase run times by approximately a factor of 2. To speed up GEOS-Chem simulations, users may choose to use any of the following options.

Run with coarser timesteps for faster turnaround and lower accuracy

Using these transport timesteps (as was done prior to v11-01) should speed up your simulation by a factor of 2:

  • 30 min = 1800 sec (4° x 5°)
  • 15 min = 900 sec (2° x 2.5°)

Select one of the mechanisms in which detailed stratospheric chemistry is omitted

These mechanisms are: Tropchem, complexSOA, complexSOA_SVPOA.

Turn off diagnostics that you don't need

Several diagnostics are turned on by default in [[GEOS-Chem_input_files#The_input.geos_fileinput.geos]] (for bpch diagnostics) and HISTORY.rc (for netCDF diagnostics). Please check those files and turn off any diagnostics that you do not need to reduce time spent in file I/O.

Make sure debug flags are turned off

If you previously compiled GEOS-Chem with debug compiler flags turned on, then you should do make realclean and recompile prior to submitting your production runs. The debug options are known to slow down simulations.

Validation

The table below demonstrates the speedups that can be obtained by changing the chemistry mechanisms and/or timesteps:

Run name,
timesteps,
and submitter
Machine or Node
and Compiler
CPU vendor CPU model Speed [MHz] # of
CPUs
CPU time Wall time CPU / Wall
ratio
% of ideal
Standard chemistry mechanism
v11-01-standard
C20T10
Bob Yantosca
regal16.rc.fas.harvard.edu
ifort 11.1
GenuineIntel Intel(R) Xeon(R) CPU E5-2660 @ 2.20GHz 2199.915 8 62554.07 s
17:22:34
9355.80 s
02:35:59
6.6861 83.58
v11-01-standard
C60T30
Melissa Sulprizio
regal17.rc.fas.harvard.edu
ifort 11.1
GenuineIntel Intel(R) Xeon(R) CPU E5-2660 @ 2.20GHz 2199.993 8 20770.70 s
08:16:11
4560.04 s
01:16:03
2.05X faster than v11-01-standard C20T10
6.5286 81.61
Tropchem chemistry mechanism
v11-01-tropchem
C20T10
Melissa Sulprizio
regal16.rc.fas.harvard.edu
ifort 11.1
GenuineIntel Intel(R) Xeon(R) CPU E5-2660 @ 2.20GHz 2199.993 8 26156.34 s
07:15:56
3881.55 s
01:04:48
2.41X faster than v11-01-standard C20T10
6.7386 84.23
v11-01-tropchem
C60T30
Melissa Sulprizio
regal16.rc.fas.harvard.edu
ifort 11.1
GenuineIntel Intel(R) Xeon(R) CPU E5-2660 @ 2.20GHz 2200.993 8 11802.31 s
03:16:42
1923.35 s
00:32:12
2.02X faster than v11-01-tropchem C20T10
6.1363 76.70

As you can see from the results above:

  1. Switching from the C20T10 configuration to the coarser C60T30 configuration results in a factor-of-2 speedup.
    • This is attributed to calling transport and chemistry less frequently.
  2. Switching from v11-01-standard simulation to v11-01-tropchem results in an additional factor-of-2 speedup.
    • This is because the standard simulation contains the UCX option, which includes a detailed stratospheric chemistry mechanism.
    • The detailed stratospheric chemistry is of course omitted in the tropchem simulation.



Previous | Guide to GEOS-Chem performance