Changes between Version 9 and Version 10 of 2009WP/2009Stream1
- Timestamp:
- 2009-10-30T18:23:43+01:00 (14 years ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
2009WP/2009Stream1
v9 v10 45 45 # io servers | # ocean procs | buffersize | success? | runtime (s) | comments 46 46 ---------------------------------------------------------------------------------------------------- 47 0 | 256 | - | yes | 2255 | 47 0 | 256 | - | yes | 2255 | Base run without IO servers 48 48 64 | 256 | 128 | yes | 2318 | 49 | | | | | 49 50 32 | 256 | 128 | no | - | " Plus de requete disponible !!!!" 50 51 32 | 256 | 512 | yes | 2654 | 52 | | | | | 51 53 16 | 256 | 512 | yes | 3782 | 54 16 | 256 | 640 | yes | 3406 | run with corrected colour assignment (null effect in this case) 55 16 | 256 | 640 | yes | 2819 | ditto but with only 2 IO servers on each node 56 16 | 256 | 640 | yes | 2532 | ditto but with only 1 IO server on each node <---best 52 57 }}} 53 58 Manufacturers, such as CRAY, are suggesting using the square root of the number of processing elements 54 59 as a guide figure for the number of I/O dedicated processes on modern MPP architectures. With this in 55 mind, further investigations are requiredinto the degradation in performance experienced with 1660 mind, further investigations have been carried out into the degradation in performance experienced with 16 56 61 IO servers. Variations in configuration parameters (such as global_mpi_buffer_size), the placement of IO 57 processes within multi-core processors and environment settings (such as MPICH_UNEX_BUFFER_SIZE) will 58 be tested in an attempt to reduce overheads. 62 processes within multi-core processors and environment settings (such as MPICH_UNEX_BUFFER_SIZE) were 63 tested in an attempt to reduce overheads. Limiting to 1 IO server per quad-core processor and using a buffer 64 size of 1024 produced the best performance. Such findings are likely to be machine-architecture specific. 59 65 60 66 For the higher resolution models, the ability to discard land-only regions is essential. Preliminary … … 73 79 ----------------------------------------------------------------------------------------------------- 74 80 32 | 221 | 256 | no | - | " Plus de requete disponible !!!!" 75 32 | 221 | 384 | yes | 2781 | Completed but output domains irregularly sized and overlap 76 32 | 221 | 1024 | no | - | Out Of Memory error ("OOM killer terminated this process") 77 32 | 221 | 512 | yes | 2873 | Also increased MPICH_UNEX_BUFFER_SIZE from 60MB to 128MB 78 32 | 221 | 512 | yes | 2472 | First run with corrected colour assignment 79 32 | 221 | 512 | yes | 2407 | run with final version of the corrected colour assignment 80 16 | 221 | 512 | yes | 3593 | run with final version of the corrected colour assignment 81 16 | 221 | 512 | yes | 2617 | ditto but with only 1 IO server on each node 81 32 | 221 | 384 | yes | 2781 | Completed but output domains irregularly sized and overlap 82 32 | 221 | 1024 | no | - | Out Of Memory error ("OOM killer terminated this process") 83 32 | 221 | 512 | yes | 2873 | Also increased MPICH_UNEX_BUFFER_SIZE from 60MB to 128MB 84 32 | 221 | 512 | yes | 2472 | First run with corrected colour assignment 85 32 | 221 | 512 | yes | 2407 | run with final version of the corrected colour assignment 86 | | | | | 87 16 | 221 | 512 | yes | 3593 | run with final version of the corrected colour assignment 88 16 | 221 | 512 | yes | 2617 | ditto but with only 1 IO server on each node 89 16 | 221 | 640 | yes | 2579 | ditto but with increased buffer size 90 16 | 221 | 768 | yes | 2506 | ditto but with increased buffer size 91 16 | 221 | 896 | yes | 2432 | ditto but with increased buffer size 92 16 | 221 | 960 | yes | 2433 | ditto but with increased buffer size 93 16 | 221 | 1024 | yes | 2376 | ditto but with increased buffer size <---- best 94 16 | 221 | 1152 | yes | 2416 | ditto but with increased buffer size 95 16 | 221 | 1280 | yes | 2434 | ditto but with increased buffer size 82 96 }}} 83 97 '''Status:''' Onging (NOCS-Southanpton)