Opened 9 years ago

Closed 8 years ago

Last modified 3 years ago

#861 closed Task (duplicate)

Discussion about the performance evaluation of the dynamic memory version

Reported by: epico Owned by: nemo
Priority: low Milestone:
Component: OCE Version: release-3.3
Severity: Keywords:
Cc: Review:
MP ready?:
Progress:

Description

The NEMO version v3.3.1, with the introduction of the dynamical memory allocation, poses some doubt about the computational performance. An analysis of the computational performance of the NEMO ver 3.3.1 is useful to evaluate an eventual loss of performance, to identify the actual bottlenecks and to define strategies to overcome inefficiency.

Commit History (0)

(No commits)

Change History (9)

comment:1 Changed 9 years ago by epico

for the discussion link to wiki:ticket/861?

comment:2 Changed 9 years ago by cbricaud

Some tests have been done by Mondher Chekki <mchekki@…> at Mercator-Ocean, with the GYRE configuration.
He compared NEMO_v3.2 ( no dynamic allocation ) with NEMO_v3.3.1 ( with dynamic allocation ) on diffents computers.
All tests have been done twice.

Run on IBM Blue Gene ( IDRIS, France)


Configuration: GYRE 48 ( ~ ORCA025)

no dynamic allocation:
256 proc: 33 min
512 proc: 18 min

with dynamic allocation:
256 proc: 41 min
512 proc: 23 min

Run on NEC SX9 (Meteo-France, France)


Configuration: GYRE 48 ( ~ ORCA025)

no dynamic allocation:
Elapse Time | Total CPU Time | Max mem / job |

425 s | 2863 s | 120851 MB |

with dynamic allocation:
Elapse Time | Total CPU Time | Max mem / job |

759 s | 5468 s | 51219 MB |

Run on a Dell Cluster with Intel Xeon 3.20 GHz ( Mercator-Ocean)


=⇒ Configuration: GYRE 48 ( ~ ORCA025)
no dynamic allocation:
Resources Used: cput=22:24:55,mem=46335032kb,walltime=00:42:56
with dynamic allocation:
Resources Used: cput=25:28:50,mem=42835392kb,walltime=00:48:44

=⇒ Configuration: GYRE 144 ( ~ ORCA12)
no dynamic allocation:
Resources Used: cput=220:04:25,mem=416895904kb,walltime=02:25:47
with dynamic allocation:
Resources Used: cput=259:10:30,mem=385099856kb,walltime=02:50:45

comment:3 follow-up: Changed 9 years ago by trackstand


How were the domains decomposed for these tests?

comment:4 in reply to: ↑ 3 Changed 9 years ago by cbricaud

Replying to trackstand:


How were the domains decomposed for these tests?

Run on IBM Blue Gene ( IDRIS, France)

Configuration: GYRE 48 ( ~ ORCA025)

no dynamic allocation:
256 proc (16*16): 33 min
512 proc (16*32): 18 min

with dynamic allocation:
256 proc (16*16): 41 min
512 proc (16*32): 23 min

Run on NEC SX9 (Meteo-France, France)

Configuration: GYRE 48 ( ~ ORCA025)
mpi configuration: 2*4

no dynamic allocation:
Elapse Time | Total CPU Time | Max mem / job |

425 s | 2863 s | 120851 MB |

with dynamic allocation:
Elapse Time | Total CPU Time | Max mem / job |

759 s | 5468 s | 51219 MB |

Run on a Dell Cluster with Intel Xeon 3.20 GHz ( Mercator-Ocean)
mpi configuration: 8*4

=⇒ Configuration: GYRE 48 ( ~ ORCA025)
no dynamic allocation:
Resources Used: cput=22:24:55,mem=46335032kb,walltime=00:42:56
with dynamic allocation:
Resources Used: cput=25:28:50,mem=42835392kb,walltime=00:48:44

=⇒ Configuration: GYRE 144 ( ~ ORCA12)
no dynamic allocation:
Resources Used: cput=220:04:25,mem=416895904kb,walltime=02:25:47
with dynamic allocation:
Resources Used: cput=259:10:30,mem=385099856kb,walltime=02:50:45

comment:5 Changed 9 years ago by cbricaud

another test done by Mondher:

Run on IBM power 6 ( ECMWF)
GYRE 48
mpi configuration: 8x16

no dynamic allocation:
Total number of CPUs : 128 proc
Total Elapsed Time : 719.00 sec
Total Memory used : 47506 Mb
Total SBU : 478.963 sbu

with dynamic allocation:
Total number of CPUs : 128 proc
Total Elapsed Time : 938.00 sec
Total Memory used : 54451 Mb
Total SBU : 624.851 sbu

comment:6 Changed 8 years ago by vichi

  • Resolution set to duplicate
  • Status changed from new to closed

comment:7 Changed 4 years ago by nicolasmartin

  • Milestone 2011 Stream 2: Developer Interface deleted

Milestone 2011 Stream 2: Developer Interface deleted

comment:8 Changed 4 years ago by nicolasmartin

  • Keywords evaluation removed

comment:9 Changed 3 years ago by nemo

  • Keywords performance removed
Note: See TracTickets for help on using tickets.