13 | | IPSL-CM5A2 is an '''the IPSL coupled climate model''' including atmosphere, land, ocean, sea ice and carbon cycle. The goal of this model is to reproduce IPSLCM5A-LR version, with reduced computing times to allow running long simulations (i.e thousands of years) within reasonnable time. Main caracteritics of this model are ORCA2-LIM2-PISCES x LMD 96x95x39, old LMDZ physics, ORCHIDEE with Choisnel hydrology (2 layers). This model uses recent versions of each components as well as technical developments (hybrid parallelisation MPI-OpenMP, XIOS 2.0 as input-output library, OASIS3-MCT parallel coupler) that allow to reach 55 simulated years per day on about 300 Curie computing cores. This model is available on Ada (IDRIS) and Curie (TGCC) supercomputers and benefits from last developments of libIGCM runnnin enviornment. |
14 | | bénéficie des derniers développements de l'environnement de production libIGCM. |
| 13 | IPSL-CM5A2 is an '''the IPSL coupled climate model''' including atmosphere, land, ocean, sea ice and carbon cycle. The goal of this model is to reproduce IPSLCM5A-LR version, with reduced computing times to allow running long simulations (i.e thousands of years) within reasonnable time. Main caracteritics of this model are '''ORCA2-LIM2-PISCES x LMD 96x95x39''', old LMDZ physics, ORCHIDEE with Choisnel hydrology (2 layers). This model uses recent versions of each component as well as technical developments (hybrid parallelization MPI-OpenMP, XIOS 2.0 as input-output library, OASIS3-MCT parallel coupler) that allow to reach '''55 simulated years per day''' on about 300 Curie computing cores. This model is available on Ada (IDRIS) and Curie (TGCC) supercomputers and benefits from last developments of libIGCM runnning environment. |
110 | | Default configuration on '''160 cores''' allows you to run '''38 simulated years per day'''. Because of load-balancing (difference between ocean computing time and atmosphere computing time), not all configurations (in terms of number of process/threads) are efficient. If you want to run a configuration with less cores, ask Arnaud Caubel what would be the optimum configuration. The configuration is regularly used and evaluated on this machine: http://webservices.ipsl.jussieu.fr/trusting/ |
| 108 | The basic configuration (default configuration) runs on Curie TGCC machine and uses 302 computing cores : 1 process for XIOS, 45 processes for NEMO, and 32 MPI processes and 8 OpenMP thread for LMDZ. This configuration allows you to run 56 simulated years per day. In order to use less computing ressources, it is possible to run on 160 cores (38 simulated years per day) as follows : 1 process for XIOS, 31 processes for NEMO, and 32 MPI processes and 4 OpenMP thread for LMDZ. To do that, you have to : |
| 109 | * Modify config.card as follows : |
| 110 | {{{ |
| 111 | #======================================================================== |
| 112 | #D-- Executable - |
| 113 | [Executable] |
| 114 | #D- For each component, Real name of executable, Name of executable for oasis |
| 115 | ATM= (gcm.e, lmdz.x, 32MPI, 4OMP) |
| 116 | SRF= ("" ,"" ) |
| 117 | SBG= ("" ,"" ) |
| 118 | OCE= (opa, opa.xx, 31MPI) |
| 119 | ICE= ("" ,"" ) |
| 120 | MBG= ("" ,"" ) |
| 121 | CPL= ("", "" ) |
| 122 | IOS= (xios_server.exe, xios.x, 1MPI) |
| 123 | }}} |
| 124 | |
| 125 | * create the new job : |
| 126 | {{{ |
| 127 | ../../libIGCM/ins_job |
| 128 | }}} |
| 129 | * Modify PARAM/namelist_ORCA2_cfg |
| 130 | {{{ |
| 131 | jpni = 1 ! jpni number of processors following i (set automatically if < 1) |
| 132 | jpnj = 31 ! jpnj number of processors following j (set automatically if < 1) |
| 133 | jpnij = 31 ! jpnij number of local domains (set automatically if < 1) |
| 134 | }}} |
| 135 | * submit the job : ccc_msub (or llsubmit) Job_MYJOBNAME |
| 136 | |
| 137 | |
| 138 | Because of load-balancing (difference between ocean computing time and atmosphere computing time), not all configurations (in terms of number of process/threads) are efficient. |
| 139 | If you want to run a configuration with less cores, ask Arnaud Caubel what would be the optimum configuration. |