Changes between Version 48 and Version 49 of 2016WP/2016SharedActions
- Timestamp:
- 2015-11-12T14:02:57+01:00 (8 years ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
2016WP/2016SharedActions
v48 v49 14 14 * '''HPC actions listed below are the list of required actions. What we can do this year remains to be determined. See Sebastien Masson's document for more details''' : [https://forge.ipsl.jussieu.fr/nemo/attachment/wiki/WorkingGroups/NEMO_HPC/HPC_tasks_Masson.doc] 15 15 * '''From Robustness and tests cases working group, state of the art and 2016 suggestions are here:''' https://forge.ipsl.jussieu.fr/nemo/wiki/WorkingGroups/Robustness-TestCases#InputtotheWorkplan2016 16 17 . 18 19 . 20 21 . 22 23 . 16 24 17 25 '''Summary of Shared Actions:''' … … 54 62 === '''AGRIF''' === 55 63 ==== ''' '''AGRIF-1 / Met Office - Increased vertical resolution with AGRIF ==== 56 ''' Motivation:''' Vertical grid refinement in NEMO would be a useful tool in simulating some aspects of the ocean (e.g. overflows) [[BR]] ''' Status: ''' Started in 2015 but much more work to do [[BR]] ''' Main Tasks: ''' [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' Medium [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : '''Tim Graham64 ''' Motivation:''' Vertical grid refinement in NEMO would be a useful tool in simulating some aspects of the ocean (e.g. overflows) [[BR]] ''' Status: ''' Started in 2015 but much more work to do [[BR]] ''' Main Tasks: ''' [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' Medium [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : ''' Tim Graham 57 65 58 66 ==== ==== … … 60 68 ==== ==== 61 69 ==== AGRIF-2 / CNRS - Assessment of LIM3 with AGRIF ==== 62 ''' Motivation:''' enable the use of AGRIF with LIM3 since LIM3 is the reference sea ice model in NEMO and LIM2 will soon be removed[[BR]] ''' Status: ''' Not yet started in 2015 [[BR]] ''' Main Tasks: ''' AGRIF already works with LIM2, so the strategy here will be to adapt what is done in LIM2 to LIM3. [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline: ''' end of 2016 [[BR]] ''' Priority:''' High [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator :'''Clément Rousset70 ''' Motivation:''' enable the use of AGRIF with LIM3 since LIM3 is the reference sea ice model in NEMO and LIM2 will soon be removed[[BR]] ''' Status: ''' Not yet started in 2015 [[BR]] ''' Main Tasks: ''' AGRIF already works with LIM2, so the strategy here will be to adapt what is done in LIM2 to LIM3. [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline: ''' end of 2016 [[BR]] ''' Priority:''' High [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator :''' Clément Rousset 63 71 64 72 === === 65 73 ==== AGRIF-3 / Mercator Océan - Agrif + vvl (two way nesting) ==== 66 ''' Motivation:''' Enable the use of agrif with temporally varying vertical grid (e.g. with zstar or ztilde coordinates) [[BR]] ''' Status: ''' Started in 2015. One way nesting is ok, two way nesting needs to be finalized. [[BR]] ''' Main Tasks: ''' [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' High [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator :'''Jérôme Chanut74 ''' Motivation:''' Enable the use of agrif with temporally varying vertical grid (e.g. with zstar or ztilde coordinates) [[BR]] ''' Status: ''' Started in 2015. One way nesting is ok, two way nesting needs to be finalized. [[BR]] ''' Main Tasks: ''' [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' High [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator :''' Jérôme Chanut 67 75 68 76 === === … … 74 82 ''' Motivation:''' Grouping communications (lbc_lnk) together is more efficient than several separate lbc_lnk calls. [[BR]] 75 83 76 ''' Status: ''' lbc_lnk_multi in 3.6 [[BR]] ''' Main Tasks: ''' Extend functionality of lbc_lnk_multi to 3D (or 4D variables). Find calls to lbc_lnk that can be replaced with lbc_lnk_multi. [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : '''Miguel Castrillo84 ''' Status: ''' lbc_lnk_multi in 3.6 [[BR]] ''' Main Tasks: ''' Extend functionality of lbc_lnk_multi to 3D (or 4D variables). Find calls to lbc_lnk that can be replaced with lbc_lnk_multi. [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : ''' Miguel Castrillo 77 85 78 86 ==== ''' '''HPC-3 / Barcelona - Suppress useless communications ==== 79 87 ''' Motivation:''' Suppress unnecessary calls to lbc_lnk. 80 88 81 [[BR]] ''' Status: ''' [[BR]] ''' Main Tasks: ''' Use a script to search for and test impact of removing lbc_lnk calls from the model code. Determine whether communications used for diagnostic purposes can be suppressed. [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : '''Tim Graham89 [[BR]] ''' Status: ''' [[BR]] ''' Main Tasks: ''' Use a script to search for and test impact of removing lbc_lnk calls from the model code. Determine whether communications used for diagnostic purposes can be suppressed. [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : ''' Tim Graham 82 90 83 91 ==== ''' '''HPC-4 / CMCC - Scalability with MPI-OPENMP ==== 84 92 ''' Motivation:''' Improving NEMO scalability. 85 93 86 [[BR]] ''' Status: ''' continuation of 2015 dev CMCC-1 [[BR]] ''' Main Tasks: ''' 1. the activity started at the end of 2013 with the analysis of three OpenMP approaches (implemented using “mini-apps”) on several multicore architectures; the second step was the implementation of the hybrid version of the GYRE configuration using the best OpenMP approach. Results in terms of scalability were not so good. Thus, the porting and the analysis of the hybrid version (using “mini-apps”) on many core architectures (e.g. Xeon Phi) is needed, before extending the hybrid OpenMP/MPI approach to the entire code, in order to justify the effort of the development. This analysis is carrying out in 2015. The extension of the hybrid implementation on the entire code will depend on the results of this analysis. 2. analysis of scalability improvement using MPI3 new communications (e.g. collective neighbours communications), instead of point to point communications [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' HIGH [[BR]] ''' Depends on:''' results of the same action in 2015 [[BR]] ''' Principal Investigator : '''Silvia Mocavero94 [[BR]] ''' Status: ''' continuation of 2015 dev CMCC-1 [[BR]] ''' Main Tasks: ''' 1. the activity started at the end of 2013 with the analysis of three OpenMP approaches (implemented using “mini-apps”) on several multicore architectures; the second step was the implementation of the hybrid version of the GYRE configuration using the best OpenMP approach. Results in terms of scalability were not so good. Thus, the porting and the analysis of the hybrid version (using “mini-apps”) on many core architectures (e.g. Xeon Phi) is needed, before extending the hybrid OpenMP/MPI approach to the entire code, in order to justify the effort of the development. This analysis is carrying out in 2015. The extension of the hybrid implementation on the entire code will depend on the results of this analysis. 2. analysis of scalability improvement using MPI3 new communications (e.g. collective neighbours communications), instead of point to point communications [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' HIGH [[BR]] ''' Depends on:''' results of the same action in 2015 [[BR]] ''' Principal Investigator : ''' Silvia Mocavero 87 95 88 96 ==== ''' '''HPC-5 / NERC - Implement larger haloes ==== 89 97 ''' Motivation:''' Improving NEMO scalability. 90 98 91 [[BR]] ''' Status: ''' New implementation - old idea [[BR]] ''' Main Tasks: ''' Investigate the benefits of using larger halo to reduce the communication frequency and hence improve performance and scalability. This is likely to be most influential in the barotropic time-splitting where it should be possible to emulate earlier solutions for the iterative solvers. Extension to the rest of the code will be considered subject to an impact assessment. [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' HIGH [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : '''Andrew Coward99 [[BR]] ''' Status: ''' New implementation - old idea [[BR]] ''' Main Tasks: ''' Investigate the benefits of using larger halo to reduce the communication frequency and hence improve performance and scalability. This is likely to be most influential in the barotropic time-splitting where it should be possible to emulate earlier solutions for the iterative solvers. Extension to the rest of the code will be considered subject to an impact assessment. [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' HIGH [[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : ''' Andrew Coward 92 100 93 101 === === … … 96 104 '''Motivation:''' Introduce a vertical physics manager that allows to save MPP communications 97 105 98 ''' Status: ''' Already started[[BR]] ''' Main Tasks: ''' [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' High[[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : '''Gurvan Madec106 ''' Status: ''' Already started[[BR]] ''' Main Tasks: ''' [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' High[[BR]] ''' Depends on:''' [[BR]] ''' Principal Investigator : ''' Gurvan Madec 99 107 100 108 ==== SIMPLIF-2 / CNRS - Simplification of configurations, introduction of user-defined module ==== … … 151 159 ''' Motivation: ''' Following investigative work by Nicolas Ducousso it became clear that the EEN scheme in NEMO did not perform as expected in idealised test cases. Although a fix was implemented that improved the performance in idealised test cases the performance in realistic simulations (e.g. ORCA025) is worse. It is not clear that the new or old formulation is correct for the case of sloping bathymetry and partial steps. 152 160 153 [[BR]] ''' Status: ''' [[BR]] ''' Main Tasks: ''' [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' [[BR]] ''' Depends on:'''[[BR]]161 [[BR]] ''' Status: ''' [[BR]] ''' Main Tasks: ''' [[BR]] ''' Science Reviewer:''' [[BR]] ''' System Reviewer:''' [[BR]] ''' Deadline:''' [[BR]] ''' Priority:''' [[BR]] ''' Depends on:''' [[BR]] 154 162 155 163 '''ROBUSTNESS-3 / CNRS - Academic test cases in the system'''
