= Rebuild Performances = Cette page rassemble des éléments quantitatifs sur les performances de rebuild sur les différents calculateurs : mercure, tx7, brodie, rhodes. Ceci devrait permettre in fine de trouver le mode de fonctionnement optimum. == Couplé IPSLCM4_v2 == * '''Résolution 96x71''' * mercure : A cette résolution, sur 4 processeurs (3 pour LMDZ/ORCHIDEE et 1 pour OPA/OASIS), 1 mois prend environ 7mn sur mercure (temps elapsed) et les rebuild (sur mercure) prennent 1 mn. * brodie : A cette résolution, sur 4 processeurs (3 pour LMDZ/ORCHIDEE et 1 pour OPA/OASIS), 1 mois prend environ 9mn sur brodie (temps elapsed) et les rebuild (sur brodie) prennent 2 mn. || Résolution || Machine || Fichier || Taille des fichiers initiaux || Taille du fichier combiné || Temps elapsed total || || 96x71x19 || '''mercure''' || histmth || 6786404, 6517600, 6786404 || 19419796 || - || || = || = || histday || 13883920, 13329420, 13883920 || 39945024 || 1mn || || = || '''brodie''' || histmth || 6786400, 6517596, 6786400 || 19419792 || - || || = || = || histday || 13883916, 13329416, 13883916 || 39945020 || 2 mn || * '''Résolution 144x96''' * brodie - tests de l'IDRIS. entre 11 mn et 62 mn. Le minimum est obtenu sur $WORKDIR avec {{{ export F_SETBUF=65536 export NC_BLOCKSIZE=67108864 }}} * Tests effectués : {{{ job.o16806:cd $TMPDIR job.o16806:real 39m6.72s job.o24807:export PATH=/workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin:$PATH job.o24807:cd $TMPDIR job.o24807:real 44m22.09s job.o24807:real 44m22.20s job.o26023:export PATH=/workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin:$PATH job.o26023:cd /workdir/rech/psl/rpsl003/REBUILD/batch job.o26023:real 18m11.19s job.o26023:real 18m11.44s job.o26122:export F_SETBUF=32768 job.o26122:export NC_BLOCKSIZE=33554432 job.o26122:export PATH=/workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin:$PATH job.o26122:cd /workdir/rech/psl/rpsl003/REBUILD/batch job.o26122:real 12m22.29s job.o26122:real 12m22.58s job.o26157:export F_SETBUF=65536 job.o26157:export NC_BLOCKSIZE=67108864 job.o26157:export PATH=/workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin:$PATH job.o26157:cd /workdir/rech/psl/rpsl003/REBUILD/batch job.o26157:real 11m42.92s job.o26157:real 11m43.16s job.o26181:export F_SETBUF=65536 job.o26181:export NC_BLOCKSIZE=67108864 job.o26181:export PATH=/workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin:$PATH job.o26181:cd $TMPDIR job.o26181:real 55m39.79s job.o26181:real 55m39.93s job.o26342:export F_SETBUF=65536 job.o26342:export NC_BLOCKSIZE=67108864 job.o26342:export F_HSDIR job.o26342:export PATH=/workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin:$PATH job.o26342:cd $TMPDIR job.o26342:real 47m49.35s job.o26342:real 47m49.63s job.o26440:export F_FILEINF=DETAIL job.o26440:cd $TMPDIR job.o26440:real 1h2m21.06s job.o26440:real 1h2m21.10s job.o27818:cd $TMPDIR job.o27818:real 53m46.54s }}} * brodie - 40 mn sur le fichier histHF en batch et 18 mn en interactif * batch : {{{ rebuild -o histhf.nc HH202_18610401_18610430_HF_histhf_0000.nc HH202_18610401_18610430_HF_histhf_0001.nc HH202_18610401_18610430_HF_histhf_0002.nc HH202_18610401_ 18610430_HF_histhf_0003.nc HH202_18610401_18610430_HF_histhf_0004.nc real 39m6.72s user 0m1.44s sys 0m38.04s }}} {{{ prof /workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin/flio_rbld mon.out + prof /workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin/flio_rbld mon.out %Time Seconds Cumsecs #Calls msec/call Name 49.4 6.84 6.84 50336 0.1360 read 38.9 5.39 12.23 45856 0.1174 write 7.9 1.09 13.32 46551 0.0234 lseek 1.3 0.19 13.51 get_user_chunk_size 0.9 0.12 13.63 fill_NC_var 0.4 0.06 13.69 nc_put_vara_float 0.3 0.04 13.73 _mcount 0.3 0.04 13.77 195320 0.0002 _VEC_memcpy 0.1 0.02 13.79 ncx_putn_float_float 0.1 0.02 13.81 144663 0.0001 strlen 0.1 0.02 13.83 f_copyback 0.1 0.02 13.85 ncx_howmany 0.0 0.00 13.85 32 0.2 brk 0.0 0.00 13.85 33 0.0 stringop.strlowercase_ 0.0 0.00 13.85 1 0. fliocom.fliocrfd_ }}} * interactif : {{{ real 18m15.19s user 0m1.43s sys 4m2.66s }}} {{{ Prof : prof /workdir/rech/psl/rpsl003/IOIPSL_PLUS/modipsl/bin/flio_rbld mon.out %Time Seconds Cumsecs #Calls msec/call Name 47.5 12.98 12.98 45856 0.2831 write 46.6 12.72 25.70 50336 0.2527 read 4.2 1.13 26.84 46551 0.0244 lseek 0.9 0.24 27.08 get_user_chunk_size 0.1 0.04 27.12 fill_NC_var 0.1 0.04 27.16 ncx_putn_float_float 0.1 0.04 27.20 nc_put_vara_float 0.1 0.02 27.22 ncx_howmany 0.1 0.02 27.24 f_cpyin 0.1 0.02 27.26 f_arydsc 0.1 0.02 27.28 _mcount 0.1 0.02 27.30 f_copyback 0.0 0.00 27.30 7 1. close 0.0 0.00 27.31 32 0.2 brk 0.0 0.00 27.31 33 0.0 stringop.strlowercase_ 0.0 0.00 27.31 41 0.0 fliocom.fliodv_rnd_ ... }}} * zahir * batch : {{{ real 0m15.50s user 0m0.38s sys 0m0.12s }}} * mercure : 17 mn dans $TMPDIR, 47 sec dans $LOCALTMPDIR * batch $LOCALTMPDIR : {{{ real 0m46.50s user 0m1.53s sys 0m41.12s prof /workdir/p86maf/IOIPSL_PLUS/modipsl/bin/flio_rbld mon.out + prof /workdir/p86maf/IOIPSL_PLUS/modipsl/bin/flio_rbld mon.out %Time Seconds Cumsecs #Calls msec/call Name 92.0 38.65 38.65 96215 0.4017 syscall 6.0 2.51 41.16 142733 0.0176 lseek 0.6 0.26 41.42 get_user_chunk_size 0.6 0.26 41.68 fill_NC_var 0.2 0.08 41.76 50336 0.0016 read 0.1 0.06 41.82 45879 0.0013 write 0.1 0.04 41.86 _mcount 0.0 0.02 41.88 f_cpyin 0.0 0.02 41.90 ncx_putn_float_float 0.0 0.02 41.92 195473 0.0001 _VEC_memcpy 0.0 0.02 41.94 f_copyback 0.0 0.02 41.96 nc_put_vara_float 0.0 0.02 41.98 396 0.05 fliocom.fliopv_r43_ 0.0 0.00 41.99 sigset 0.0 0.00 41.99 7 1. close 0.0 0.00 42.00 23 0.2 llseek 0.0 0.00 42.00 33 0.0 stringop.strlowercase_ }}} * batch $TMPDIR : {{{ real 17m13.50s user 0m1.62s sys 3m12.15s prof /workdir/p86maf/IOIPSL_PLUS/modipsl/bin/flio_rbld mon.out + prof /workdir/p86maf/IOIPSL_PLUS/modipsl/bin/flio_rbld mon.out %Time Seconds Cumsecs #Calls msec/call Name 82.7 18.29 18.29 96215 0.1901 syscall 13.6 3.02 21.31 142733 0.0211 lseek 1.2 0.26 21.57 fill_NC_var 0.7 0.16 21.73 get_user_chunk_size 0.6 0.14 21.86 50336 0.0028 read 0.3 0.06 21.92 45879 0.0013 write 0.2 0.04 21.97 _mcount 0.2 0.04 22.01 ncx_howmany 0.1 0.02 22.03 11021 0.0018 _real_malloc 0.1 0.02 22.05 fy_chtrim 0.1 0.02 22.07 ncx_putn_float_float 0.1 0.02 22.09 4256 0.005 _VEC_memset 0.1 0.02 22.11 195473 0.0001 _VEC_memcpy 0.0 0.00 22.11 33 0.0 stringop.strlowercase_ 0.0 0.00 22.11 1 0. fliocom.fliocrfd_ 0.0 0.00 22.11 41 0.0 fliocom.fliodv_rnd_ }}} * '''Résolution 144x142''' * brodie : A cette résolution, sur 6 processeurs (5 pour LMDZ/ORCHIDEE et 1 pour OPA/OASIS), 1 mois prend 48 mn sur brodie (temps elapsed) et les rebuild 90 mn (/tmpdir)! Prend 45 mn sur /workdir. * mercure : A cette résolution, sur 6 processeurs, l'ensemble des rebuild (histmt, histrac, histda histhf) prend 3 mn. || Résolution || Machine || Fichier || Taille des fichiers initiaux || Taille du fichier combiné || Temps elapsed || || 144x142x19 || '''brodie''' || histmth || 12162628, 11759424, 11759424, 11759424, 12162628 || 57724284 || 1 mn || || = || = || histday || 24974064, 24142316, 24142316, 24142316, 24974064 || 118961192 || 7 mn-10 mn || || = || = || histrac || 6459092, 6244240, 6244240, 6244240, 6459092|| 30736972 || 1 mn || || = || = || histhf || 149316592, 144339948, 144339948, 144339948, 149316592 || 711676968 || 79 mn (33 mn en /workdir) || || = || = || sechiba_out || 2787544, 2695380, 2695380, 2695380, 2787544 || 13201680 || 1 mn || || 144x142x19 || '''mercure''' || histmth + histday +histrac +histhf || || > 930 Mo || 3 mn || * Petit fichier histmth: * essai en interactif sur brodie01 : {{{ brodie01 - /workdir/rech/psl/rpsl003/REBUILD/test : time rebuild -o histmth.nc histmth_*.nc real 0m2.29s user 0m0.88s sys 0m0.46s }}} * essai en interactif sur zahir/$WORKDIR : {{{ zahir001-rpsl003 : time rebuild -o histmth.nc histmth_* real 0m5.97s user 0m0.81s sys 0m1.64s }}} * essai en interactif sur rhodes/$WORKDIR : {{{ rhodes - /workdir/rech/psl/rpsl003/REBUILD : time rebuild -o histmth.nc histmth_*.nc real 0m56.71s user 0m10.34s sys 0m3.81s }}} == Tests sur mercure == Voici plusieurs tests réalisés sur mercure. [[br]] A chaque fois il s'agit de reconstruire un fichier de données inca_avgr.nc de taille 772M coupé en 6 (6x 140M) [[br]] * essai en interactif sur le scratchdir/ : {{{ elapsed time (s) : 1.3337 e+03 cpu time (s) : 7.2045 e+02 }}} * essai en batch sur le $LOCALTMPDIR {{{ Real Time (sec) : 67.379782 User Time (sec) : 4.167928 Sys Time (sec) : 62.203794 }}} * essai en batch sur le scratchdir/ {{{ Real Time (sec) : 1069.691442 User Time (sec) : 4.132475 Sys Time (sec) : 216.605864 }}} == Tests sur platine == Voici plusieurs tests réalisés sur platine. [[br]] A chaque fois il s'agit de reconstruire un fichier de données inca_avgr.nc de taille 772M coupé en 6 (6x 140M) [[br]] * essai en interactif/batch sur le scratchdir : {{{ Real Time (sec) : 683.84 User Time (sec) : 8.08 Sys Time (sec) : 202.37 }}} * essai en interactif/batch sur le home : {{{ Real Time (sec) : 77.33 User Time (sec) : 6.74 Sys Time (sec) : 6.02 }}}