Difference between revisions of "GCHP Timing Tests"
(→Reported Model Runs) |
(→Reported Model Runs) |
||
Line 36: | Line 36: | ||
{| border=1 cellspacing=0 cellpadding=5 | {| border=1 cellspacing=0 cellpadding=5 | ||
|-valign="top" bgcolor="#CCCCCC" | |-valign="top" bgcolor="#CCCCCC" | ||
− | !width=" | + | !width="40px"|ID |
!width="100px"|Organization | !width="100px"|Organization | ||
− | !width=" | + | !width="80px"|Cluster |
− | !width=" | + | !width="80px"|Partition |
− | !width=" | + | !width="80px"|OS |
− | !width=" | + | !width="80px"|Network interconnect |
!width="100px"|Processor | !width="100px"|Processor | ||
!width="100px"|CPU Vendor<br>Model<br>Speed [MHz] | !width="100px"|CPU Vendor<br>Model<br>Speed [MHz] | ||
− | !width=" | + | !width="100px"|Compiler |
− | !width=" | + | !width="100px"|MPI |
− | !width=" | + | !width="100px"|NetCDF |
!width="100px"|Filesystem info | !width="100px"|Filesystem info | ||
Revision as of 17:38, 29 August 2018
Overview
This page features reports of user timing tests done with High Performance GEOS-Chem (GCHP). If you are running GCHP please add your own results below. You can sign up for a wiki account to edit this page or you can send your information to the GEOS-Chem Support Team.
Performing the Timing Tests
Follow instructions on Getting Started with GCHP to install GCHP using GEOS-Chem 12.0.0, acquire a run directory using the GEOS-Chem Unit Tester, and compile source code. We are interested in the results of a 7-day c24 GCHP simulation run on a single node with 6, 12, and 24 cores. Feel free to perform additional tests using any number of nodes and cores and for any internal resolution and record your results as supplementary information on this page.
GCHP timing information is displayed at the end of the GCHP.log log file. The output can be interpreted as follows:
User time = GCHP Total + ExtData Total + HIST Total + COMM_TOTAL
where,
- User time is total run time and is in section with “----“ headers
- GCHP = GIGCchem + Dynamics + GIGCenv (each with their own breakdown sections except GIGCenv)
- GIGCchem = the time for GEOS-Chem “classic” code (e.g. DO_CHEMISTRY)
- ExtData is the input and regridding time
- HIST is the output time
- COMM is the communication time listed until total time
We are most interested in the total time (user time) and the times for ExtData, dynamics (transport), chemistry, emissions, and convection. Please contact the GCST if you have trouble determining what lines to look at for retrieving the times for these components.
Reported Model Runs
If you are running GCHP we encourage you to report performance in this section. Run settings may vary considerably with the following exceptions:
- Jobs start on model date 2016/07/01 00:00 GMT.
- The code was compiled from the standard GCHP run directory (gchp_standard) with the option make compile_clean.
Enter your system information in the top table to the best of your knowledge and your simulation information in the bottom table. Use a unique id to cross-reference between the two.
Enter your system information here:
ID | Organization | Cluster | Partition | OS | Network interconnect | Processor | CPU Vendor Model Speed [MHz] |
Compiler | MPI | NetCDF | Filesystem info |
---|---|---|---|---|---|---|---|---|---|---|---|
000 | Harvard | Odyssey | huce_intel | CentOS 7 | Infiniband | gcc 7.1.0 | openmpi 3.1.1 | netcdf 4.1.3 | LFS | ||
001 | Harvard | Odyssey | huce_intel | CentOS 7 | Infiniband | ifort 17 | openmpi 3.1.1 | netcdf 4.1.3 | LFS | ||
Enter your simulation information here:
User | ID | GEOS-Chem version | # nodes | # cores | Simulation | Duration | Met source | Met res | Grid res | Dyn timestep [min] | Chem timestep [min] | # diagnostics | Diag freq | Diag Duration | Total time (s) | Dyn time (s) | Chem time (s) | Hist time (s) | ExtData time (s) |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Lizzie Lundgren | 000 | 12.0.1 | 3 | 48 | standard | 7 days | GEOS-FP | 0.25x0.3125 | c48 | 10 | 20 | default | 24hr | 24hr | 34848 | 7645 | 16195 | 211 | 2840 |
Graphs of Model Run Times
These will be generated when we have a critical mass of data.