| 1 | #!/bin/ksh |
|---|
| 2 | # subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0 |
|---|
| 3 | # $Id: subjob 1021 2012-10-02 13:52:49Z raasch $ |
|---|
| 4 | |
|---|
| 5 | # Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS |
|---|
| 6 | # laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden |
|---|
| 7 | # Host zurueckgeschickt werden sollen |
|---|
| 8 | |
|---|
| 9 | |
|---|
| 10 | # letzte Aenderung: |
|---|
| 11 | # 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion |
|---|
| 12 | # 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0) |
|---|
| 13 | # 06/02/98 - Siggi - berte validiert |
|---|
| 14 | # 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea |
|---|
| 15 | # moeglich |
|---|
| 16 | # 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt |
|---|
| 17 | # 25/05/02 - Siggi - Unterstuetzung des LoadLeveler |
|---|
| 18 | # 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie |
|---|
| 19 | # allgemeine Anpassungen fuer ibm-Rechner |
|---|
| 20 | # 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul |
|---|
| 21 | # Ruecktransfer des Jobprotokolls fuer diese |
|---|
| 22 | # Maschine abgeschaltet |
|---|
| 23 | # 31/10/02 - Siggi - berni validiert |
|---|
| 24 | # 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh |
|---|
| 25 | # 08/11/02 - Siggi - quanero validiert |
|---|
| 26 | # 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet |
|---|
| 27 | # 23/01/03 - Siggi - hostname nobel changed to nobela |
|---|
| 28 | # 06/02/03 - Siggi - gregale validated |
|---|
| 29 | # 12/02/03 - Siggi - orkan and maestro validated |
|---|
| 30 | # 21/02/03 - Siggi - all nobel nodes in Seoul validated |
|---|
| 31 | # 12/03/03 - Siggi - nec at DKRZ validated |
|---|
| 32 | # 13/03/03 - Siggi - new nqs resource variable Memory |
|---|
| 33 | # 07/04/03 - Siggi - processor request option -c on nech needs tasks per |
|---|
| 34 | # node |
|---|
| 35 | # 11/04/03 - Siggi - network on ibms has attribute "shared" |
|---|
| 36 | # 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2) |
|---|
| 37 | # cxxl added to ibmh |
|---|
| 38 | # 29/08/03 - Siggi - changes in job queues and communication system on |
|---|
| 39 | # ibms |
|---|
| 40 | # 24/10/03 - Siggi - using alternate hanni address 130.75.4.2 |
|---|
| 41 | # 30/10/03 - Siggi - nech is not supported any more |
|---|
| 42 | # 10/11/03 - Siggi - nech2 renamed to nech |
|---|
| 43 | # 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub |
|---|
| 44 | # 29/03/04 - Siggi - ground not supported any more, gfdl3 validated |
|---|
| 45 | # 31/03/04 - Siggi - new option -N for node usage |
|---|
| 46 | # 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha |
|---|
| 47 | # due to error in ssh installation (otherwise a prompt |
|---|
| 48 | # for the password appears) |
|---|
| 49 | # 23/07/04 - Siggi - changes due to the new berni configuration |
|---|
| 50 | # (federation switch) |
|---|
| 51 | # 01/09/04 - Gerald new job-classes on hanni |
|---|
| 52 | # 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10 |
|---|
| 53 | # 23/11/04 - Siggi - new job class cdata on hanni and berni |
|---|
| 54 | # 03/12/04 - Siggi - notification on ibm switched of in case of |
|---|
| 55 | # delete_dayfile = true, node usage in cdev set to |
|---|
| 56 | # shared |
|---|
| 57 | # 16/02/05 - Gerald hababai validated |
|---|
| 58 | # 29/03/05 - Micha - new job class channi on hanni |
|---|
| 59 | # 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword |
|---|
| 60 | # on ibms |
|---|
| 61 | # 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to |
|---|
| 62 | # p_normal |
|---|
| 63 | # 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip" |
|---|
| 64 | # 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to |
|---|
| 65 | # "sn_all", new job class cexp |
|---|
| 66 | # 08/09/05 - Siggi - IP-address of gfdl3 changed |
|---|
| 67 | # 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on |
|---|
| 68 | # hurricane (with -X0) |
|---|
| 69 | # 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech) |
|---|
| 70 | # added |
|---|
| 71 | # 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105 |
|---|
| 72 | # 09/02/06 - Siggi - ibmy admitted |
|---|
| 73 | # 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha |
|---|
| 74 | # 13/04/06 - Siggi - ostria admitted |
|---|
| 75 | # 18/04/06 - Siggi - new option -O for OpenMP usage |
|---|
| 76 | # 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob |
|---|
| 77 | # 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs |
|---|
| 78 | # 27/09/06 - Siggi - breg/hreg extended with berni/hanni |
|---|
| 79 | # 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni |
|---|
| 80 | # 28/11/06 - Siggi - levanto admitted |
|---|
| 81 | # 13/02/07 - Siggi - hpmuk releated code removed |
|---|
| 82 | # 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx) |
|---|
| 83 | # 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only |
|---|
| 84 | # 27/07/07 - Siggi - autan admitted |
|---|
| 85 | # 03/08/07 - Marcus- lcfimm admitted |
|---|
| 86 | # 08/10/07 - Siggi - further job classes added for hanni (csoloh...) |
|---|
| 87 | # 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's |
|---|
| 88 | # suggestions |
|---|
| 89 | # 19/10/07 - Marcus- add new optional argument -g group_number |
|---|
| 90 | # 19/10/07 - Siggi - a ";" was missing in the last change done by Marcus |
|---|
| 91 | # 30/10/07 - Marcus- further adjustments for queues on lctit |
|---|
| 92 | # 15/05/08 - Siggi - adjustments for lcxt4 (Bergen Center for Computational |
|---|
| 93 | # Science) |
|---|
| 94 | # 14/07/08 - Siggi - adjustments for lcsgih |
|---|
| 95 | # 23/09/08 - Gerald- paesano admitted |
|---|
| 96 | # 02/10/08 - Siggi - PBS adjustments for lcxt4 |
|---|
| 97 | # 02/03/09 - Siggi - Adjustments for new NEC-SX9 at RIAM |
|---|
| 98 | # 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih |
|---|
| 99 | # 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh |
|---|
| 100 | # 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed |
|---|
| 101 | # 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually) |
|---|
| 102 | # 08/07/09 - Siggi - option -e added (email notification on lcsgih/b) |
|---|
| 103 | # 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are |
|---|
| 104 | # now run on the data nodes (feature=data) |
|---|
| 105 | # 25/08/09 - BjornM- adapted for lck |
|---|
| 106 | # 26/08/09 - Marcus- caurus admitted; optional qos feature hiprio on lcsgi |
|---|
| 107 | # 03/09/09 - Siggi - PBS sgi feature directive only used if explicitly |
|---|
| 108 | # set in the config file by the user |
|---|
| 109 | # 16/10/09 - Carolin-adjustments for archiving on SGI-ICE of binary files; |
|---|
| 110 | # adding special1q |
|---|
| 111 | # 01/12/09 - BjornM- re-adjustments for lcxt4, added sno (130.75.105.113) |
|---|
| 112 | # 16/10/09 - Carolin-adding permq |
|---|
| 113 | # 01/02/10 - Siggi - adapted for lcxt5m and lckyoto (Fujitsu HX600) |
|---|
| 114 | # 03/02/10 - Siggi - bug in serial jobs removed |
|---|
| 115 | # 26/02/10 - BjornM- re-adjustments for lcxt4 (new modules, email |
|---|
| 116 | # notification) |
|---|
| 117 | # 01/03/10 - Siggi - loading of modules controlled by environment variable |
|---|
| 118 | # module_calls |
|---|
| 119 | # 17/08/10 - BjornM- re-adjustments for lcxt4 (location of qsub) |
|---|
| 120 | # 25/08/10 - BjornM- account geofysisk replaced by guest for lcxt4 |
|---|
| 121 | # 25/08/10 - Siggi - new variable project_account in pbs-statements for |
|---|
| 122 | # lcxt4 |
|---|
| 123 | # 08/12/10 - Siggi - initialization of the module command changed for |
|---|
| 124 | # SGI-ICE/lcsgi |
|---|
| 125 | # adjustments for Kyushu Univ. (lcrte, ibmku) |
|---|
| 126 | # 14/12/10 - Siggi - adjustments for new Tsubame system at Tokyo |
|---|
| 127 | # institute of technology (lctit) |
|---|
| 128 | # 02/02/11 - Siggi - further asjustments for Tsubame concerning openMP |
|---|
| 129 | # 06/03/11 - Siggi - adjustments for ibmkisti |
|---|
| 130 | # 17/03/11 - Siggi - adjustments for openmp usage on ibmkisti |
|---|
| 131 | # 03/04/11 - Micha - added lckordi |
|---|
| 132 | # 17/08/11 - Siggi - hicegate0 added |
|---|
| 133 | # 18/08/11 - Siggi - workaround on lcsgi in order to avoid appends to file |
|---|
| 134 | # last_job_transfer_protocol |
|---|
| 135 | # 21/08/11 - Siggi - inferno admitted |
|---|
| 136 | # 29/11/11 - Siggi - adjustments for lcsgih/lcsgib queues, pingui admitted |
|---|
| 137 | # 21/12/11 - Theres- solano admitted |
|---|
| 138 | # 31/01/12 - Matthias - adjustments for ibmh |
|---|
| 139 | # 02/05/12 - Bjoern - lcxt4 is now lcxe6, adjustments for lcxe6, |
|---|
| 140 | # bugfixes for the parameter file check, initial |
|---|
| 141 | # commands on ssh calls sourced out to .mrun.config |
|---|
| 142 | # 02/10/12 - Siggi - adjustments for lcsb |
|---|
| 143 | |
|---|
| 144 | |
|---|
| 145 | |
|---|
| 146 | # VARIABLENVEREINBARUNGEN + DEFAULTWERTE |
|---|
| 147 | delete_dayfile=false |
|---|
| 148 | email_notification=none |
|---|
| 149 | group_number=none |
|---|
| 150 | locat=normal |
|---|
| 151 | no_default_queue=none |
|---|
| 152 | no_submit=false |
|---|
| 153 | job_catalog="~/job_queue" |
|---|
| 154 | job_name=none |
|---|
| 155 | local_user=$LOGNAME |
|---|
| 156 | node_usage=shared |
|---|
| 157 | numprocs=0 |
|---|
| 158 | punkte="..........................................................." |
|---|
| 159 | submcom=qsub |
|---|
| 160 | queue=default |
|---|
| 161 | remote_host=none |
|---|
| 162 | remote_user="" |
|---|
| 163 | verify=true |
|---|
| 164 | |
|---|
| 165 | typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden |
|---|
| 166 | typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1 |
|---|
| 167 | typeset -L20 spalte1 |
|---|
| 168 | typeset -R40 spalte2 |
|---|
| 169 | typeset -L60 spalte3 |
|---|
| 170 | |
|---|
| 171 | |
|---|
| 172 | |
|---|
| 173 | # FEHLERBEHANDLUNG |
|---|
| 174 | # BEI EXIT: |
|---|
| 175 | trap 'if [[ $locat != normal ]] |
|---|
| 176 | then |
|---|
| 177 | case $locat in |
|---|
| 178 | (option) printf "\n --> available optios can be displayed" |
|---|
| 179 | printf " by typing:" |
|---|
| 180 | printf "\n \"subjob ?\" \n";; |
|---|
| 181 | (ftpcopy|parameter|scp|verify) printf "\n";; |
|---|
| 182 | (*) printf "\n +++ unknown error" |
|---|
| 183 | printf "\n please inform S. Raasch!\n" |
|---|
| 184 | esac |
|---|
| 185 | [[ -f $job_to_send ]] && rm $job_to_send |
|---|
| 186 | printf "\n\n+++ SUBJOB killed \n\n" |
|---|
| 187 | fi' exit |
|---|
| 188 | |
|---|
| 189 | |
|---|
| 190 | # BEI TERMINAL-BREAK: |
|---|
| 191 | trap '[[ -f $job_to_send ]] && rm $job_to_send |
|---|
| 192 | printf "\n\n+++ SUBJOB killed \n\n" |
|---|
| 193 | exit |
|---|
| 194 | ' 2 |
|---|
| 195 | |
|---|
| 196 | |
|---|
| 197 | |
|---|
| 198 | |
|---|
| 199 | # LOKALEN HOSTNAMEN FESTSTELLEN |
|---|
| 200 | local_host=$(hostname) |
|---|
| 201 | |
|---|
| 202 | |
|---|
| 203 | |
|---|
| 204 | # HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST |
|---|
| 205 | # UEBERHAUPT ZULAESSIG IST |
|---|
| 206 | # Note: One of the entries for "lck" or "lckordi" always has to be |
|---|
| 207 | # comment out, because the hostname (node*) is same for both machines |
|---|
| 208 | case $local_host in |
|---|
| 209 | (atmos) local_addres=172.20.25.35; local_host=lcide;; |
|---|
| 210 | (autan) local_addres=130.75.105.57; local_host=lcmuk;; |
|---|
| 211 | (bora) local_addres=130.75.105.103; local_host=lcmuk;; |
|---|
| 212 | (bd1) local_addres=130.73.232.64; local_host=lcsgib;; |
|---|
| 213 | (bd2) local_addres=130.73.232.65; local_host=lcsgib;; |
|---|
| 214 | (bd3) local_addres=130.73.232.66; local_host=lcsgib;; |
|---|
| 215 | (bd4) local_addres=130.73.232.67; local_host=lcsgib;; |
|---|
| 216 | (b01*|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;; |
|---|
| 217 | (bicegate2) local_addres=130.73.232.103; local_host=lcsgib;; |
|---|
| 218 | (blizzard1) local_addres=136.172.40.15; local_host=ibmh;; |
|---|
| 219 | (breva) local_addres=130.75.105.98; local_host=lcmuk;; |
|---|
| 220 | (caurus) local_addres=130.75.105.19; local_host=lcmuk;; |
|---|
| 221 | (climate*) local_addres=165.132.26.68; local_host=lcyon;; |
|---|
| 222 | (compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;; |
|---|
| 223 | (cs*) local_addres=136.172.44.131; local_host=nech;; |
|---|
| 224 | (elephanta) local_addres=130.75.105.6; local_host=lcmuk;; |
|---|
| 225 | (fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;; |
|---|
| 226 | (node*) local_addres=165.132.26.61 local_host=lck;; |
|---|
| 227 | # (node*) local_addres=210.219.61.8 local_host=lckordi;; |
|---|
| 228 | (gaia*) local_addres=150.183.146.24; local_host=ibmkisti;; |
|---|
| 229 | (gallego) local_addres=130.75.105.10; local_host=lcmuk;; |
|---|
| 230 | (gfdl5) local_addres=165.132.26.58; local_host=ibmy;; |
|---|
| 231 | (gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;; |
|---|
| 232 | (gregale) local_addres=130.75.105.109; local_host=lcmuk;; |
|---|
| 233 | (hababai) local_addres=130.75.105.108; local_host=lcmuk;; |
|---|
| 234 | (hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;; |
|---|
| 235 | (hd1) local_addres=130.75.4.104; local_host=lcsgih;; |
|---|
| 236 | (hd2) local_addres=130.75.4.105; local_host=lcsgih;; |
|---|
| 237 | (hd3) local_addres=130.75.4.106; local_host=lcsgih;; |
|---|
| 238 | (hd4) local_addres=130.75.4.107; local_host=lcsgih;; |
|---|
| 239 | (hicegate0) local_addres=130.75.4.101; local_host=lcsgih;; |
|---|
| 240 | (h01*|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;; |
|---|
| 241 | (hicegate2) local_addres=130.75.4.103; local_host=lcsgih;; |
|---|
| 242 | (hx*) local_addres=133.3.51.11; local_host=lckyoto;; |
|---|
| 243 | (inferno) local_addres=130.75.105.5; local_host=lcmuk;; |
|---|
| 244 | (irifi) local_addres=130.75.105.104; local_host=lcmuk;; |
|---|
| 245 | (sno) local_addres=130.75.105.113; local_host=lcmuk;; |
|---|
| 246 | (levanto) local_addres=130.75.105.45; local_host=lcmuk;; |
|---|
| 247 | (maestro) local_addres=130.75.105.2; local_host=lcmuk;; |
|---|
| 248 | (meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;; |
|---|
| 249 | (hexagon*) local_addres=129.177.20.113; local_host=lcxe6;; |
|---|
| 250 | (nobel*) local_addres=150.183.5.101; local_host=ibms;; |
|---|
| 251 | (orkan) local_addres=130.75.105.3; local_host=lcmuk;; |
|---|
| 252 | (ostria) local_addres=130.75.105.106; local_host=lcmuk;; |
|---|
| 253 | (paesano) local_addres=130.75.105.46; local_host=lcmuk;; |
|---|
| 254 | (pingui) local_addres=134.106.74.118; local_host=lcfor;; |
|---|
| 255 | (quanero) local_addres=130.75.105.107; local_host=lcmuk;; |
|---|
| 256 | (rte*) local_addres=133.5.185.60; local_host=lcrte;; |
|---|
| 257 | (scirocco) local_addres=172.20.25.41; local_host=lcmuk;; |
|---|
| 258 | (solano) local_addres=130.75.105.110; local_host=lcmuk;; |
|---|
| 259 | (sun1|sun2) local_addres=130.75.6.1; local_host=unics;; |
|---|
| 260 | (sx-*) local_addres=172.16.1.131; local_host=necriam;; |
|---|
| 261 | (r1*) local_addres=130.75.4.102; local_host=lcsgih;; |
|---|
| 262 | (r2*) local_addres=130.73.232.102; local_host=lcsgib;; |
|---|
| 263 | (t2a*) local_addres=10.1.6.165; local_host=lctit;; |
|---|
| 264 | (urban*) local_addres=147.46.30.151 local_host=lcsb;; |
|---|
| 265 | (vorias) local_addres=172.20.25.43; local_host=lcmuk;; |
|---|
| 266 | (*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;; |
|---|
| 267 | (*) printf "\n +++ \"$local_host\" unknown"; |
|---|
| 268 | printf "\n please inform S. Raasch!"; |
|---|
| 269 | locat=parameter; exit;; |
|---|
| 270 | esac |
|---|
| 271 | |
|---|
| 272 | |
|---|
| 273 | |
|---|
| 274 | # REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN |
|---|
| 275 | remote_host=$local_host |
|---|
| 276 | |
|---|
| 277 | |
|---|
| 278 | |
|---|
| 279 | |
|---|
| 280 | # PROZEDUROPTIONEN EINLESEN |
|---|
| 281 | while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option |
|---|
| 282 | do |
|---|
| 283 | case $option in |
|---|
| 284 | (c) job_catalog=$OPTARG;; |
|---|
| 285 | (d) delete_dayfile=true;; |
|---|
| 286 | (D) no_submit=true;; |
|---|
| 287 | (e) email_notification=$OPTARG;; |
|---|
| 288 | (g) group_number=$OPTARG;; |
|---|
| 289 | (h) remote_host=$OPTARG;; |
|---|
| 290 | (m) memory=$OPTARG;; |
|---|
| 291 | (n) job_name=$OPTARG;; |
|---|
| 292 | (N) node_usage=$OPTARG;; |
|---|
| 293 | (O) threads_per_task=$OPTARG;; |
|---|
| 294 | (q) no_default_queue=$OPTARG;; |
|---|
| 295 | (t) cputime=$OPTARG;; |
|---|
| 296 | (T) tasks_per_node=$OPTARG;; |
|---|
| 297 | (u) remote_user=$OPTARG;; |
|---|
| 298 | (v) verify=false;; |
|---|
| 299 | (X) numprocs=$OPTARG;; |
|---|
| 300 | (\?) printf "\n +++ Option $OPTARG unknown \n"; |
|---|
| 301 | locat=option; exit;; |
|---|
| 302 | esac |
|---|
| 303 | done |
|---|
| 304 | |
|---|
| 305 | |
|---|
| 306 | # JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN |
|---|
| 307 | shift OPTIND-1; file_to_send=$1 |
|---|
| 308 | |
|---|
| 309 | |
|---|
| 310 | # KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN |
|---|
| 311 | if [ "$1" = "?" ] |
|---|
| 312 | then |
|---|
| 313 | (printf "\n *** subjob can be called as follows:\n" |
|---|
| 314 | printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n" |
|---|
| 315 | printf "\n Description of available options:\n" |
|---|
| 316 | printf "\n Option Description Default-Value" |
|---|
| 317 | printf "\n -c job-input- and output-catalog ~/job_queue" |
|---|
| 318 | printf "\n -d no job-protocol will be created ---" |
|---|
| 319 | printf "\n -D only the job-file will be created ---" |
|---|
| 320 | printf "\n -h execution host, available hosts: $remote_host" |
|---|
| 321 | printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...," |
|---|
| 322 | printf "\n lctit, nech, necriam, unics" |
|---|
| 323 | printf "\n -m memory demand per process in MByte ---" |
|---|
| 324 | printf "\n -n jobname <jobdatei>" |
|---|
| 325 | printf "\n -O threads per task (for OpenMP usage) 1" |
|---|
| 326 | printf "\n -q job-queue to be used default" |
|---|
| 327 | printf "\n -t allowed cpu-time in seconds ---" |
|---|
| 328 | printf "\n -T tasks per node (on parallel hosts) ---" |
|---|
| 329 | printf "\n -u username on execution host from .netrc" |
|---|
| 330 | printf "\n -v no prompt for confirmation ---" |
|---|
| 331 | printf "\n -X # of processors (on parallel hosts) 1" |
|---|
| 332 | printf "\n " |
|---|
| 333 | printf "\n The only possible positional parameter is <jobfile>:" |
|---|
| 334 | printf "\n The complete NQS-job must be provided here." |
|---|
| 335 | printf "\n <jobfile>=? creates this outline\n\n") | more |
|---|
| 336 | exit |
|---|
| 337 | fi |
|---|
| 338 | |
|---|
| 339 | |
|---|
| 340 | |
|---|
| 341 | # PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT |
|---|
| 342 | if [[ "$file_to_send" = "" ]] |
|---|
| 343 | then |
|---|
| 344 | printf "\n +++ job-file missing" |
|---|
| 345 | locat=parameter; exit |
|---|
| 346 | else |
|---|
| 347 | if [[ -f $file_to_send ]] |
|---|
| 348 | then |
|---|
| 349 | true |
|---|
| 350 | else |
|---|
| 351 | printf "\n +++ job-file: " |
|---|
| 352 | printf "\n $file_to_send" |
|---|
| 353 | printf "\n does not exist" |
|---|
| 354 | locat=parameter; exit |
|---|
| 355 | fi |
|---|
| 356 | fi |
|---|
| 357 | |
|---|
| 358 | |
|---|
| 359 | |
|---|
| 360 | # FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME |
|---|
| 361 | # GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE |
|---|
| 362 | if [[ $job_name = none ]] |
|---|
| 363 | then |
|---|
| 364 | job_name=$file_to_send |
|---|
| 365 | fi |
|---|
| 366 | if [[ $(echo $job_name | grep -c "/") != 0 ]] |
|---|
| 367 | then |
|---|
| 368 | printf "\n +++ job-file name: " |
|---|
| 369 | printf "\n $job_name" |
|---|
| 370 | printf "\n must not contain \"/\"-characters" |
|---|
| 371 | locat=parameter; exit |
|---|
| 372 | fi |
|---|
| 373 | |
|---|
| 374 | |
|---|
| 375 | |
|---|
| 376 | |
|---|
| 377 | # HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST |
|---|
| 378 | # ODER WENN HOST NICHT ANGEGEBEN WURDE |
|---|
| 379 | if [[ $remote_host = none ]] |
|---|
| 380 | then |
|---|
| 381 | printf "\n +++ host missing" |
|---|
| 382 | locat=option; exit |
|---|
| 383 | else |
|---|
| 384 | case $remote_host in |
|---|
| 385 | (ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;; |
|---|
| 386 | (ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;; |
|---|
| 387 | (ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;; |
|---|
| 388 | (ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;; |
|---|
| 389 | (ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;; |
|---|
| 390 | (ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;; |
|---|
| 391 | (lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;; |
|---|
| 392 | (lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;; |
|---|
| 393 | (lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;; |
|---|
| 394 | (lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;; |
|---|
| 395 | (lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;; |
|---|
| 396 | (lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;; |
|---|
| 397 | (lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;; |
|---|
| 398 | (lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;; |
|---|
| 399 | (lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;; |
|---|
| 400 | (lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;; |
|---|
| 401 | (lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;; |
|---|
| 402 | (nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";; |
|---|
| 403 | (necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";; |
|---|
| 404 | (vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;; |
|---|
| 405 | (unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;; |
|---|
| 406 | (*) printf "\n +++ hostname \"$remote_host\" not allowed"; |
|---|
| 407 | locat=parameter; exit;; |
|---|
| 408 | esac |
|---|
| 409 | fi |
|---|
| 410 | |
|---|
| 411 | |
|---|
| 412 | # EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST |
|---|
| 413 | if [[ $no_default_queue != none ]] |
|---|
| 414 | then |
|---|
| 415 | error=false |
|---|
| 416 | ndq=$no_default_queue |
|---|
| 417 | case $remote_host in |
|---|
| 418 | (ibm) case $ndq in |
|---|
| 419 | (p690_express|p690_standard|p690_long) error=false;; |
|---|
| 420 | (*) error=true;; |
|---|
| 421 | esac;; |
|---|
| 422 | (ibmh) case $ndq in |
|---|
| 423 | (no_class) error=false;; |
|---|
| 424 | (*) error=true;; |
|---|
| 425 | esac;; |
|---|
| 426 | (ibmkisti) case $ndq in |
|---|
| 427 | (class.32plus|class.1-2|class.2-32) error=false;; |
|---|
| 428 | (*) error=true;; |
|---|
| 429 | esac;; |
|---|
| 430 | (ibmku) case $ndq in |
|---|
| 431 | (sdbg1|sdbg2|sdbg4|s4|s16|s32|s32-s) error=false;; |
|---|
| 432 | (*) error=true;; |
|---|
| 433 | esac;; |
|---|
| 434 | (ibms) case $ndq in |
|---|
| 435 | (express|normal|p_express|p_normal|p_normal_1.3|p_normal_1.7|grand) error=false;; |
|---|
| 436 | (*) error=true;; |
|---|
| 437 | esac;; |
|---|
| 438 | (ibmy) case $ndq in |
|---|
| 439 | (parallel) error=false;; |
|---|
| 440 | (*) error=true;; |
|---|
| 441 | esac;; |
|---|
| 442 | (lckyoto) case $ndq in |
|---|
| 443 | (eh|ph) error=false;; |
|---|
| 444 | (*) error=true;; |
|---|
| 445 | esac;; |
|---|
| 446 | (lcsgib|lcsgih) case $ndq in |
|---|
| 447 | (testq|serialq|smallq|mediumq|bigq|workq|dataq|permq|special1q) error=false;; |
|---|
| 448 | (*) error=true;; |
|---|
| 449 | esac;; |
|---|
| 450 | (lctit) case $ndq in |
|---|
| 451 | (G|L128|L256|L512H|S|S96|V) error=false;; |
|---|
| 452 | (*) error=true;; |
|---|
| 453 | esac;; |
|---|
| 454 | (t3eb) case $ndq in |
|---|
| 455 | (berte|p50|p100|p392|forfree|p25himem) error=false;; |
|---|
| 456 | (*) error=true;; |
|---|
| 457 | esac;; |
|---|
| 458 | (necriam) case $ndq in |
|---|
| 459 | (SP|SS|P6) error=false;; |
|---|
| 460 | (*) error=true;; |
|---|
| 461 | esac;; |
|---|
| 462 | (t3eh) case $ndq in |
|---|
| 463 | (para_t3e|em|k|l|lm|comp_t3e|c|p|ht) error=false;; |
|---|
| 464 | (*) error=true;; |
|---|
| 465 | esac;; |
|---|
| 466 | (t3ej2|t3ej5) case $ndq in |
|---|
| 467 | (low|normal|high) error=false;; |
|---|
| 468 | (*) error=true;; |
|---|
| 469 | esac;; |
|---|
| 470 | (t3es) case $ndq in |
|---|
| 471 | (batch|serial-4|pe4|p48|pe16|pe32|pe64|pe128) error=false;; |
|---|
| 472 | (*) error=true;; |
|---|
| 473 | esac;; |
|---|
| 474 | (unics) case $ndq in |
|---|
| 475 | (unics|ht) error=false;; |
|---|
| 476 | (*) error=true;; |
|---|
| 477 | esac;; |
|---|
| 478 | esac |
|---|
| 479 | if [[ $error = true ]] |
|---|
| 480 | then |
|---|
| 481 | printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed" |
|---|
| 482 | locat=parameter; exit |
|---|
| 483 | else |
|---|
| 484 | queue=$no_default_queue |
|---|
| 485 | fi |
|---|
| 486 | fi |
|---|
| 487 | |
|---|
| 488 | |
|---|
| 489 | |
|---|
| 490 | # PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN |
|---|
| 491 | # AUFTEILEN |
|---|
| 492 | done=false |
|---|
| 493 | while [[ $done = false ]] |
|---|
| 494 | do |
|---|
| 495 | if (( $cputime <= 0 )) |
|---|
| 496 | then |
|---|
| 497 | printf "\n +++ wrong cpu-time or cpu-time missing" |
|---|
| 498 | printf "\n >>> Please type cpu-time in seconds as INTEGER:" |
|---|
| 499 | printf "\n >>> " |
|---|
| 500 | read cputime 1>/dev/null 2>&1 |
|---|
| 501 | else |
|---|
| 502 | done=true |
|---|
| 503 | fi |
|---|
| 504 | done |
|---|
| 505 | if [[ $remote_host = nech ]] |
|---|
| 506 | then |
|---|
| 507 | if (( tasks_per_node != 0 )) |
|---|
| 508 | then |
|---|
| 509 | (( cputime = cputime * tasks_per_node )) |
|---|
| 510 | elif [[ $numprocs != 0 ]] |
|---|
| 511 | then |
|---|
| 512 | (( cputime = cputime * numprocs )) |
|---|
| 513 | fi |
|---|
| 514 | fi |
|---|
| 515 | (( stunden = cputime / 3600 )) |
|---|
| 516 | (( resttime = cputime - stunden * 3600 )) |
|---|
| 517 | (( minuten = resttime / 60 )) |
|---|
| 518 | (( sekunden = resttime - minuten * 60 )) |
|---|
| 519 | timestring=${stunden}:${minuten}:${sekunden} |
|---|
| 520 | |
|---|
| 521 | |
|---|
| 522 | |
|---|
| 523 | # PRUEFEN DER KERNSPEICHERANFORDERUNG |
|---|
| 524 | done=false |
|---|
| 525 | while [[ $done = false ]] |
|---|
| 526 | do |
|---|
| 527 | if (( memory <= 0 )) |
|---|
| 528 | then |
|---|
| 529 | printf "\n +++ wrong memory demand or memory demand missing" |
|---|
| 530 | printf "\n >>> Please type memory in MByte per process as INTEGER:" |
|---|
| 531 | printf "\n >>> " |
|---|
| 532 | read memory 1>/dev/null 2>&1 |
|---|
| 533 | else |
|---|
| 534 | done=true |
|---|
| 535 | fi |
|---|
| 536 | done |
|---|
| 537 | |
|---|
| 538 | if [[ $remote_host = nech || $remote_host = necriam ]] |
|---|
| 539 | then |
|---|
| 540 | if (( tasks_per_node != 0 )) |
|---|
| 541 | then |
|---|
| 542 | (( Memory = memory * tasks_per_node / 1000 )) |
|---|
| 543 | elif [[ $numprocs != 0 ]] |
|---|
| 544 | then |
|---|
| 545 | (( Memory = memory * numprocs / 1000 )) |
|---|
| 546 | else |
|---|
| 547 | (( Memory = memory / 1000 )) |
|---|
| 548 | fi |
|---|
| 549 | elif [[ $remote_host = lctit ]] |
|---|
| 550 | then |
|---|
| 551 | (( Memory = memory * tasks_per_node / 1000 )) |
|---|
| 552 | fi |
|---|
| 553 | |
|---|
| 554 | |
|---|
| 555 | # SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG |
|---|
| 556 | if [[ $(echo $remote_host | cut -c1-3) = ibm ]] |
|---|
| 557 | then |
|---|
| 558 | (( memory = memory * threads_per_task )) |
|---|
| 559 | fi |
|---|
| 560 | |
|---|
| 561 | |
|---|
| 562 | # BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN |
|---|
| 563 | if (( tasks_per_node != 0 )) |
|---|
| 564 | then |
|---|
| 565 | (( nodes = numprocs / ( tasks_per_node * threads_per_task ) )) |
|---|
| 566 | fi |
|---|
| 567 | |
|---|
| 568 | # Calculate number of processes per node |
|---|
| 569 | (( processes_per_node = tasks_per_node * threads_per_task )) |
|---|
| 570 | |
|---|
| 571 | # Calculate number of MPI tasks |
|---|
| 572 | (( mpi_tasks = numprocs / threads_per_task )) |
|---|
| 573 | |
|---|
| 574 | |
|---|
| 575 | # HEADER-AUSGABE |
|---|
| 576 | if [[ $verify = true ]] |
|---|
| 577 | then |
|---|
| 578 | printf "\n\n" |
|---|
| 579 | printf "#--------------------------------------------------------------# \n" |
|---|
| 580 | spalte1=SUBJOB;spalte2=$(date) |
|---|
| 581 | printf "| $spalte1$spalte2 | \n" |
|---|
| 582 | printf "| | \n" |
|---|
| 583 | printf "| values of parameters/options: | \n" |
|---|
| 584 | spalte1=local_host$punkte; spalte2=$punkte$local_host |
|---|
| 585 | printf "| $spalte1$spalte2 | \n" |
|---|
| 586 | spalte1=remote_host$punkte; spalte2=$punkte$remote_host |
|---|
| 587 | printf "| $spalte1$spalte2 | \n" |
|---|
| 588 | spalte1=queue$punkte; spalte2=$punkte$queue |
|---|
| 589 | printf "| $spalte1$spalte2 | \n" |
|---|
| 590 | spalte1=memory$punkte; spalte2="$punkte$memory mb" |
|---|
| 591 | printf "| $spalte1$spalte2 | \n" |
|---|
| 592 | spalte1=cputime$punkte; spalte2="$punkte$cputime sec" |
|---|
| 593 | printf "| $spalte1$spalte2 | \n" |
|---|
| 594 | spalte1=job_name$punkte; spalte2="$punkte$job_name" |
|---|
| 595 | printf "| $spalte1$spalte2 | \n" |
|---|
| 596 | printf "#--------------------------------------------------------------# \n\n" |
|---|
| 597 | |
|---|
| 598 | |
|---|
| 599 | # KONTROLLABFRAGE, OB ALLES O.K. |
|---|
| 600 | antwort="dummy" |
|---|
| 601 | while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]] |
|---|
| 602 | do |
|---|
| 603 | read antwort?" >>> continue (y/n) ? " |
|---|
| 604 | done |
|---|
| 605 | if [[ $antwort = n || $antwort = N ]] |
|---|
| 606 | then |
|---|
| 607 | locat=verify; exit |
|---|
| 608 | fi |
|---|
| 609 | printf "\n" |
|---|
| 610 | fi |
|---|
| 611 | |
|---|
| 612 | |
|---|
| 613 | |
|---|
| 614 | # ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN |
|---|
| 615 | kennung=$RANDOM |
|---|
| 616 | job_on_remhost=${job_name}_${kennung}_$local_host |
|---|
| 617 | job_to_send=job_to_send_$kennung |
|---|
| 618 | if [[ $delete_dayfile = false ]] |
|---|
| 619 | then |
|---|
| 620 | remote_dayfile=${local_host}_${job_name}_result_$kennung |
|---|
| 621 | local_dayfile=${remote_host}_${job_name} |
|---|
| 622 | else |
|---|
| 623 | remote_dayfile=/dev/null |
|---|
| 624 | fi |
|---|
| 625 | |
|---|
| 626 | |
|---|
| 627 | |
|---|
| 628 | # Generate the batch job scripts (qsub/msub/LoadLeveler) |
|---|
| 629 | if [[ $(echo $remote_host | cut -c1-3) = ibm && $numprocs != 0 ]] |
|---|
| 630 | then |
|---|
| 631 | |
|---|
| 632 | # General LoadLeveler settings |
|---|
| 633 | execute_in_shell="#!/bin/ksh" |
|---|
| 634 | use_shell="# @ shell = /bin/ksh" |
|---|
| 635 | consumable_memory="ConsumableMemory($memory mb)" |
|---|
| 636 | class="# @ class = $queue" |
|---|
| 637 | environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes" |
|---|
| 638 | network_to_use="# @ network.mpi = sn_all,shared,us" |
|---|
| 639 | data_limit="# @ data_limit = 1.76gb" |
|---|
| 640 | image_size="# @ image_size = 50" |
|---|
| 641 | wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring" |
|---|
| 642 | |
|---|
| 643 | if [[ $email_notification = none ]] |
|---|
| 644 | then |
|---|
| 645 | notify_user="" |
|---|
| 646 | else |
|---|
| 647 | notify_user="# @ notify_user = $email_notification" |
|---|
| 648 | if [[ $delete_dayfile = true ]] |
|---|
| 649 | then |
|---|
| 650 | notification='# @ notification = never' |
|---|
| 651 | fi |
|---|
| 652 | fi |
|---|
| 653 | |
|---|
| 654 | if [[ $remote_host = ibmh ]] |
|---|
| 655 | then |
|---|
| 656 | data_limit="" |
|---|
| 657 | network_to_use="" |
|---|
| 658 | class="" |
|---|
| 659 | environment="" |
|---|
| 660 | rset="# @ rset = RSET_MCM_AFFINITY" |
|---|
| 661 | task_affinity="# @ task_affinity = core(1)" |
|---|
| 662 | elif [[ $remote_host = ibmkisti ]] |
|---|
| 663 | then |
|---|
| 664 | network_to_use="# @ network.MPI = sn_all,shared,US" |
|---|
| 665 | wall_clock_limit="# @ wall_clock_limit = $timestring" |
|---|
| 666 | if [[ $threads_per_task = 1 ]] |
|---|
| 667 | then |
|---|
| 668 | rset="# @ rset = RSET_MCM_AFFINITY" |
|---|
| 669 | mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute" |
|---|
| 670 | fi |
|---|
| 671 | environment="" |
|---|
| 672 | use_shell="" |
|---|
| 673 | data_limit="" |
|---|
| 674 | image_size="" |
|---|
| 675 | elif [[ $remote_host = ibmku ]] |
|---|
| 676 | then |
|---|
| 677 | execute_in_shell="#!/usr/bin/ksh" |
|---|
| 678 | use_shell="# @ shell = /usr/bin/ksh" |
|---|
| 679 | consumable_memory="" |
|---|
| 680 | environment="" |
|---|
| 681 | network_to_use="# @ network.mpi = sn_all,shared,us" |
|---|
| 682 | data_limit="" |
|---|
| 683 | image_size="" |
|---|
| 684 | elif [[ $remote_host = ibms ]] |
|---|
| 685 | then |
|---|
| 686 | network_to_use="# @ network.mpi = csss,shared,us" |
|---|
| 687 | elif [[ $remote_host = ibmy ]] |
|---|
| 688 | then |
|---|
| 689 | consumable_memory="" |
|---|
| 690 | network_to_use="" |
|---|
| 691 | fi |
|---|
| 692 | |
|---|
| 693 | cat > $job_to_send << %%END%% |
|---|
| 694 | $execute_in_shell |
|---|
| 695 | $use_shell |
|---|
| 696 | |
|---|
| 697 | # @ job_type = parallel |
|---|
| 698 | # @ job_name = $job_name |
|---|
| 699 | # @ resources = ConsumableCpus($threads_per_task) $consumable_memory |
|---|
| 700 | # @ output = $remote_dayfile |
|---|
| 701 | # @ error = $remote_dayfile |
|---|
| 702 | $wall_clock_limit |
|---|
| 703 | $image_size |
|---|
| 704 | $class |
|---|
| 705 | $environment |
|---|
| 706 | $network_to_use |
|---|
| 707 | $data_limit |
|---|
| 708 | $rset |
|---|
| 709 | $mcm_affinity_options |
|---|
| 710 | $task_affinity |
|---|
| 711 | $notification |
|---|
| 712 | $notify_user |
|---|
| 713 | |
|---|
| 714 | %%END%% |
|---|
| 715 | |
|---|
| 716 | if (( nodes > 0 )) |
|---|
| 717 | then |
|---|
| 718 | |
|---|
| 719 | if [[ $remote_host != ibmkisti ]] |
|---|
| 720 | then |
|---|
| 721 | |
|---|
| 722 | cat >> $job_to_send << %%END%% |
|---|
| 723 | # @ node = $nodes |
|---|
| 724 | # @ tasks_per_node = $processes_per_node |
|---|
| 725 | # @ node_usage = $node_usage |
|---|
| 726 | # @ queue |
|---|
| 727 | |
|---|
| 728 | %%END%% |
|---|
| 729 | |
|---|
| 730 | else |
|---|
| 731 | |
|---|
| 732 | cat >> $job_to_send << %%END%% |
|---|
| 733 | # @ total_tasks = $mpi_tasks |
|---|
| 734 | # @ blocking = unlimited |
|---|
| 735 | # @ queue |
|---|
| 736 | |
|---|
| 737 | %%END%% |
|---|
| 738 | |
|---|
| 739 | fi |
|---|
| 740 | |
|---|
| 741 | else |
|---|
| 742 | |
|---|
| 743 | if [[ $remote_host != ibmy ]] |
|---|
| 744 | then |
|---|
| 745 | |
|---|
| 746 | cat >> $job_to_send << %%END%% |
|---|
| 747 | # @ blocking = unlimited |
|---|
| 748 | # @ total_tasks = $numprocs |
|---|
| 749 | # @ node_usage = $node_usage |
|---|
| 750 | # @ queue |
|---|
| 751 | |
|---|
| 752 | %%END%% |
|---|
| 753 | |
|---|
| 754 | else |
|---|
| 755 | |
|---|
| 756 | cat >> $job_to_send << %%END%% |
|---|
| 757 | # @ node = 1 |
|---|
| 758 | # @ total_tasks = $numprocs |
|---|
| 759 | # @ queue |
|---|
| 760 | |
|---|
| 761 | %%END%% |
|---|
| 762 | |
|---|
| 763 | fi |
|---|
| 764 | |
|---|
| 765 | fi |
|---|
| 766 | |
|---|
| 767 | # workaround because of silly job filter on ibmkisti |
|---|
| 768 | if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]] |
|---|
| 769 | then |
|---|
| 770 | echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send |
|---|
| 771 | fi |
|---|
| 772 | |
|---|
| 773 | elif [[ $(echo $remote_host | cut -c1-3) = ibm && $numprocs = 0 ]] |
|---|
| 774 | then |
|---|
| 775 | |
|---|
| 776 | cat > $job_to_send << %%END%% |
|---|
| 777 | #!/bin/ksh |
|---|
| 778 | |
|---|
| 779 | # @ job_type = serial |
|---|
| 780 | # @ node_usage = $node_usage |
|---|
| 781 | # @ job_name = palm |
|---|
| 782 | # @ wall_clock_limit = ${timestring},$timestring |
|---|
| 783 | # @ resources = ConsumableCpus(1) ConsumableMemory(1 gb) |
|---|
| 784 | # @ output = $remote_dayfile |
|---|
| 785 | # @ error = $remote_dayfile |
|---|
| 786 | $class |
|---|
| 787 | $notification |
|---|
| 788 | |
|---|
| 789 | # @ queue |
|---|
| 790 | |
|---|
| 791 | %%END%% |
|---|
| 792 | |
|---|
| 793 | elif [[ $remote_host = lcfimm ]] |
|---|
| 794 | then |
|---|
| 795 | |
|---|
| 796 | if [[ $numprocs != 0 ]] |
|---|
| 797 | then |
|---|
| 798 | cat > $job_to_send << %%END%% |
|---|
| 799 | #!/bin/ksh |
|---|
| 800 | #PBS -N $job_name |
|---|
| 801 | #PBS -A $project_account |
|---|
| 802 | #PBS -l walltime=$timestring |
|---|
| 803 | #PBS -l nodes=${nodes}:ppn=$processes_per_node |
|---|
| 804 | #PBS -l pmem=${memory}mb |
|---|
| 805 | #PBS -m abe |
|---|
| 806 | #PBS -o $remote_dayfile |
|---|
| 807 | #PBS -j oe |
|---|
| 808 | mpd & |
|---|
| 809 | |
|---|
| 810 | %%END%% |
|---|
| 811 | |
|---|
| 812 | else |
|---|
| 813 | cat > $job_to_send << %%END%% |
|---|
| 814 | #!/bin/ksh |
|---|
| 815 | #PBS -N $job_name |
|---|
| 816 | #PBS -A $project_account |
|---|
| 817 | #PBS -l walltime=$timestring |
|---|
| 818 | #PBS -l ncpus=1 |
|---|
| 819 | #PBS -l pmem=${memory}mb |
|---|
| 820 | #PBS -m abe |
|---|
| 821 | #PBS -o $remote_dayfile |
|---|
| 822 | #PBS -j oe |
|---|
| 823 | |
|---|
| 824 | %%END%% |
|---|
| 825 | |
|---|
| 826 | fi |
|---|
| 827 | |
|---|
| 828 | elif [[ $remote_host = lck || $remote_host = lckordi || $remote_host = lcsb ]] |
|---|
| 829 | then |
|---|
| 830 | |
|---|
| 831 | if [[ $numprocs != 0 ]] |
|---|
| 832 | then |
|---|
| 833 | cat > $job_to_send << %%END%% |
|---|
| 834 | #!/bin/ksh |
|---|
| 835 | #PBS -N $job_name |
|---|
| 836 | #PBS -l walltime=$timestring |
|---|
| 837 | #PBS -l ncpus=$numprocs |
|---|
| 838 | #PBS -l pmem=${memory}mb |
|---|
| 839 | #PBS -o $remote_dayfile |
|---|
| 840 | #PBS -l nodes=$nodes:ppn=${processes_per_node} |
|---|
| 841 | #PBS -j oe |
|---|
| 842 | |
|---|
| 843 | mpd & |
|---|
| 844 | |
|---|
| 845 | %%END%% |
|---|
| 846 | |
|---|
| 847 | else |
|---|
| 848 | cat > $job_to_send << %%END%% |
|---|
| 849 | #!/bin/ksh |
|---|
| 850 | #PBS -N $job_name |
|---|
| 851 | #PBS -l walltime=$timestring |
|---|
| 852 | #PBS -l ncpus=1 |
|---|
| 853 | #PBS -l pmem=${memory}mb |
|---|
| 854 | #PBS -o $remote_dayfile |
|---|
| 855 | #PBS -j oe |
|---|
| 856 | |
|---|
| 857 | %%END%% |
|---|
| 858 | |
|---|
| 859 | fi |
|---|
| 860 | |
|---|
| 861 | elif [[ $remote_host = lcyon ]] |
|---|
| 862 | then |
|---|
| 863 | |
|---|
| 864 | if [[ $numprocs != 0 ]] |
|---|
| 865 | then |
|---|
| 866 | cat > $job_to_send << %%END%% |
|---|
| 867 | #!/bin/ksh |
|---|
| 868 | #PBS -N $job_name |
|---|
| 869 | #PBS -l walltime=$timestring |
|---|
| 870 | #PBS -l ncpus=$numprocs |
|---|
| 871 | #PBS -l pmem=${memory}mb |
|---|
| 872 | #PBS -o $remote_dayfile |
|---|
| 873 | #PBS -j oe |
|---|
| 874 | |
|---|
| 875 | %%END%% |
|---|
| 876 | |
|---|
| 877 | else |
|---|
| 878 | cat > $job_to_send << %%END%% |
|---|
| 879 | #!/bin/ksh |
|---|
| 880 | #PBS -N $job_name |
|---|
| 881 | #PBS -l walltime=$timestring |
|---|
| 882 | #PBS -l ncpus=1 |
|---|
| 883 | #PBS -l pmem=${memory}mb |
|---|
| 884 | #PBS -o $remote_dayfile |
|---|
| 885 | #PBS -j oe |
|---|
| 886 | |
|---|
| 887 | %%END%% |
|---|
| 888 | |
|---|
| 889 | fi |
|---|
| 890 | |
|---|
| 891 | elif [[ $remote_host = lcsgih || $remote_host = lcsgib ]] |
|---|
| 892 | then |
|---|
| 893 | |
|---|
| 894 | if [[ "$sgi_feature" != "" ]] |
|---|
| 895 | then |
|---|
| 896 | feature_directive="#PBS -l feature=$sgi_feature" |
|---|
| 897 | else |
|---|
| 898 | feature_directive="" |
|---|
| 899 | fi |
|---|
| 900 | |
|---|
| 901 | if [[ $queue = dataq || $queue = permq ]] |
|---|
| 902 | then |
|---|
| 903 | feature_directive="#PBS -l feature=data" |
|---|
| 904 | fi |
|---|
| 905 | |
|---|
| 906 | if [[ $queue = testq || $queue = mediumq || $queue = bigq || $queue = workq || $queue = dataq || $queue = permq || $queue = serialq || $queue = special1q ]] |
|---|
| 907 | then |
|---|
| 908 | queue_directive="#PBS -q $queue" |
|---|
| 909 | else |
|---|
| 910 | queue_directive="" |
|---|
| 911 | fi |
|---|
| 912 | |
|---|
| 913 | if [[ $email_notification = none ]] |
|---|
| 914 | then |
|---|
| 915 | email_directive="" |
|---|
| 916 | else |
|---|
| 917 | email_directive="#PBS -M $email_notification" |
|---|
| 918 | fi |
|---|
| 919 | |
|---|
| 920 | if [[ $numprocs != 0 ]] |
|---|
| 921 | then |
|---|
| 922 | cat > $job_to_send << %%END%% |
|---|
| 923 | #!/bin/bash |
|---|
| 924 | #PBS -N $job_name |
|---|
| 925 | #PBS -l walltime=$timestring |
|---|
| 926 | #PBS -l nodes=$nodes:ppn=${processes_per_node} |
|---|
| 927 | #PBS -l naccesspolicy=$node_usage |
|---|
| 928 | #PBS -o $remote_dayfile |
|---|
| 929 | #PBS -j oe |
|---|
| 930 | $feature_directive |
|---|
| 931 | $queue_directive |
|---|
| 932 | $email_directive |
|---|
| 933 | |
|---|
| 934 | eval \`/sw/swdist/bin/modulesinit\` |
|---|
| 935 | #. /usr/share/modules/init/bash |
|---|
| 936 | $init_cmds |
|---|
| 937 | $module_calls |
|---|
| 938 | |
|---|
| 939 | echo ld_library_path=\$LD_LIBRARY_PATH |
|---|
| 940 | |
|---|
| 941 | %%END%% |
|---|
| 942 | |
|---|
| 943 | else |
|---|
| 944 | cat > $job_to_send << %%END%% |
|---|
| 945 | #PBS -S /bin/bash |
|---|
| 946 | #PBS -N $job_name |
|---|
| 947 | #PBS -l walltime=$timestring |
|---|
| 948 | #PBS -l ncpus=1 |
|---|
| 949 | #PBS -o $remote_dayfile |
|---|
| 950 | #PBS -j oe |
|---|
| 951 | $queue_directive |
|---|
| 952 | $email_directive |
|---|
| 953 | |
|---|
| 954 | eval \`/sw/swdist/bin/modulesinit\` |
|---|
| 955 | #. /usr/share/modules/init/bash |
|---|
| 956 | $init_cmds |
|---|
| 957 | $module_calls |
|---|
| 958 | |
|---|
| 959 | %%END%% |
|---|
| 960 | |
|---|
| 961 | fi |
|---|
| 962 | |
|---|
| 963 | elif [[ $remote_host = lcxe6 ]] |
|---|
| 964 | then |
|---|
| 965 | |
|---|
| 966 | if [[ $numprocs != 0 ]] |
|---|
| 967 | then |
|---|
| 968 | cat > $job_to_send << %%END%% |
|---|
| 969 | #!/bin/ksh |
|---|
| 970 | #PBS -S /bin/ksh |
|---|
| 971 | #PBS -N $job_name |
|---|
| 972 | #PBS -A $project_account |
|---|
| 973 | #PBS -j oe |
|---|
| 974 | #PBS -l walltime=$timestring |
|---|
| 975 | #PBS -l mppwidth=${numprocs} |
|---|
| 976 | #PBS -l mppnppn=${processes_per_node} |
|---|
| 977 | #PBS -m abe |
|---|
| 978 | #PBS -o $remote_dayfile |
|---|
| 979 | $email_directive |
|---|
| 980 | |
|---|
| 981 | $init_cmds |
|---|
| 982 | $module_calls |
|---|
| 983 | |
|---|
| 984 | %%END%% |
|---|
| 985 | |
|---|
| 986 | else |
|---|
| 987 | cat > $job_to_send << %%END%% |
|---|
| 988 | #!/bin/ksh |
|---|
| 989 | #PBS -S /bin/ksh |
|---|
| 990 | #PBS -N $job_name |
|---|
| 991 | #PBS -A $project_account |
|---|
| 992 | #PBS -j oe |
|---|
| 993 | #PBS -l walltime=$timestring |
|---|
| 994 | #PBS -l ncpus=1 |
|---|
| 995 | #PBS -l pmem=${memory}mb |
|---|
| 996 | #PBS -m abe |
|---|
| 997 | $email_directive |
|---|
| 998 | #PBS -o $remote_dayfile |
|---|
| 999 | |
|---|
| 1000 | $init_cmds |
|---|
| 1001 | $module_calls |
|---|
| 1002 | |
|---|
| 1003 | %%END%% |
|---|
| 1004 | |
|---|
| 1005 | fi |
|---|
| 1006 | |
|---|
| 1007 | elif [[ $remote_host = lckyoto ]] |
|---|
| 1008 | then |
|---|
| 1009 | |
|---|
| 1010 | cat > $job_to_send << %%END%% |
|---|
| 1011 | #!/bin/ksh |
|---|
| 1012 | # @\$-o $remote_dayfile |
|---|
| 1013 | # @\$-eo -oi |
|---|
| 1014 | # @\$-lP 16 |
|---|
| 1015 | # @\$-lp 1 |
|---|
| 1016 | # @\$-lm 28gb -llm unlimited -ls unlimited |
|---|
| 1017 | # @\$-q $queue |
|---|
| 1018 | # @\$-Pvn abs_pack |
|---|
| 1019 | ##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack |
|---|
| 1020 | #. /thin/local/etc/setprofile/intel-11.0.sh |
|---|
| 1021 | #. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh |
|---|
| 1022 | . ~/.myprofile |
|---|
| 1023 | #. /home2/t/t51254/palm/current_version/myprofile |
|---|
| 1024 | #. /thin/apps/pgi/mpi.sh |
|---|
| 1025 | # |
|---|
| 1026 | env |
|---|
| 1027 | # |
|---|
| 1028 | set -x |
|---|
| 1029 | |
|---|
| 1030 | %%END%% |
|---|
| 1031 | |
|---|
| 1032 | elif [[ $remote_host = lcxt5m ]] |
|---|
| 1033 | then |
|---|
| 1034 | |
|---|
| 1035 | if [[ $numprocs != 0 ]] |
|---|
| 1036 | then |
|---|
| 1037 | cat > $job_to_send << %%END%% |
|---|
| 1038 | #!/bin/ksh |
|---|
| 1039 | #PBS -S /bin/ksh |
|---|
| 1040 | #PBS -N $job_name |
|---|
| 1041 | #PBS -j oe |
|---|
| 1042 | #PBS -l walltime=$timestring |
|---|
| 1043 | #PBS -l mppwidth=${numprocs} |
|---|
| 1044 | #PBS -l mppnppn=${processes_per_node} |
|---|
| 1045 | #PBS -m abe |
|---|
| 1046 | #PBS -o $remote_dayfile |
|---|
| 1047 | |
|---|
| 1048 | $init_cmds |
|---|
| 1049 | $module_calls |
|---|
| 1050 | |
|---|
| 1051 | %%END%% |
|---|
| 1052 | |
|---|
| 1053 | else |
|---|
| 1054 | cat > $job_to_send << %%END%% |
|---|
| 1055 | #!/bin/ksh |
|---|
| 1056 | #PBS -S /bin/ksh |
|---|
| 1057 | #PBS -N $job_name |
|---|
| 1058 | #PBS -j oe |
|---|
| 1059 | #PBS -l walltime=$timestring |
|---|
| 1060 | #PBS -l ncpus=1 |
|---|
| 1061 | #PBS -l pmem=${memory}mb |
|---|
| 1062 | #PBS -m abe |
|---|
| 1063 | #PBS -o $remote_dayfile |
|---|
| 1064 | |
|---|
| 1065 | $init_cmds |
|---|
| 1066 | $module_calls |
|---|
| 1067 | |
|---|
| 1068 | %%END%% |
|---|
| 1069 | |
|---|
| 1070 | fi |
|---|
| 1071 | |
|---|
| 1072 | elif [[ $remote_host = nech ]] |
|---|
| 1073 | then |
|---|
| 1074 | |
|---|
| 1075 | if (( nodes > 1 )) |
|---|
| 1076 | then |
|---|
| 1077 | cat > $job_to_send << %%END%% |
|---|
| 1078 | #!/bin/ksh |
|---|
| 1079 | #PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime |
|---|
| 1080 | #PBS -l ${qsubmem}=${Memory}gb |
|---|
| 1081 | #PBS -b $nodes |
|---|
| 1082 | #PBS -o $remote_dayfile |
|---|
| 1083 | #PBS -N palm |
|---|
| 1084 | #PBS -j o |
|---|
| 1085 | #PBS -T mpisx |
|---|
| 1086 | |
|---|
| 1087 | %%END%% |
|---|
| 1088 | |
|---|
| 1089 | elif [[ $numprocs != 0 ]] |
|---|
| 1090 | then |
|---|
| 1091 | cat > $job_to_send << %%END%% |
|---|
| 1092 | #!/bin/ksh |
|---|
| 1093 | #PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime |
|---|
| 1094 | #PBS -l ${qsubmem}=${Memory}gb |
|---|
| 1095 | #PBS -o $remote_dayfile |
|---|
| 1096 | #PBS -N palm |
|---|
| 1097 | #PBS -j o |
|---|
| 1098 | |
|---|
| 1099 | %%END%% |
|---|
| 1100 | |
|---|
| 1101 | else |
|---|
| 1102 | cat > $job_to_send << %%END%% |
|---|
| 1103 | #!/bin/ksh |
|---|
| 1104 | #PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime |
|---|
| 1105 | #PBS -o $remote_dayfile |
|---|
| 1106 | #PBS -j o |
|---|
| 1107 | |
|---|
| 1108 | %%END%% |
|---|
| 1109 | |
|---|
| 1110 | fi |
|---|
| 1111 | |
|---|
| 1112 | elif [[ $remote_host = necriam ]] |
|---|
| 1113 | then |
|---|
| 1114 | |
|---|
| 1115 | if (( nodes > 1 )) |
|---|
| 1116 | then |
|---|
| 1117 | cat > $job_to_send << %%END%% |
|---|
| 1118 | #!/bin/ksh |
|---|
| 1119 | #PBS -b $nodes |
|---|
| 1120 | #PBS -o $remote_dayfile |
|---|
| 1121 | #PBS -N $job_name |
|---|
| 1122 | #PBS -j o |
|---|
| 1123 | #PBS -v MPIPROGINV=YES |
|---|
| 1124 | |
|---|
| 1125 | %%END%% |
|---|
| 1126 | |
|---|
| 1127 | elif [[ $numprocs != 0 ]] |
|---|
| 1128 | then |
|---|
| 1129 | cat > $job_to_send << %%END%% |
|---|
| 1130 | #!/bin/ksh |
|---|
| 1131 | #PBS -q ${queue} |
|---|
| 1132 | #PBS -o $remote_dayfile |
|---|
| 1133 | #PBS -N $job_name |
|---|
| 1134 | #PBS -j o |
|---|
| 1135 | #PBS -v MPIPROGINV=YES |
|---|
| 1136 | |
|---|
| 1137 | %%END%% |
|---|
| 1138 | |
|---|
| 1139 | else |
|---|
| 1140 | cat > $job_to_send << %%END%% |
|---|
| 1141 | #!/bin/ksh |
|---|
| 1142 | #PBS -o $remote_dayfile |
|---|
| 1143 | #PBS -j o |
|---|
| 1144 | |
|---|
| 1145 | %%END%% |
|---|
| 1146 | |
|---|
| 1147 | fi |
|---|
| 1148 | |
|---|
| 1149 | elif [[ $remote_host = lctit ]] |
|---|
| 1150 | then |
|---|
| 1151 | cat > $job_to_send << %%END%% |
|---|
| 1152 | #!/bin/ksh |
|---|
| 1153 | $init_cmds |
|---|
| 1154 | $module_calls |
|---|
| 1155 | |
|---|
| 1156 | %%END%% |
|---|
| 1157 | |
|---|
| 1158 | # OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN |
|---|
| 1159 | if [[ $tasks_per_node != $processes_per_node ]] |
|---|
| 1160 | then |
|---|
| 1161 | submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -o $remote_dayfile -j oe -q $queue " |
|---|
| 1162 | else |
|---|
| 1163 | submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -q $queue " |
|---|
| 1164 | fi |
|---|
| 1165 | |
|---|
| 1166 | else |
|---|
| 1167 | |
|---|
| 1168 | cat > $job_to_send << %%END%% |
|---|
| 1169 | # @\$-q ${queue} |
|---|
| 1170 | # @\$-l${qsubtime} $timestring |
|---|
| 1171 | # @\$-l${qsubmem} ${memory}mb |
|---|
| 1172 | # @\$-o $remote_dayfile |
|---|
| 1173 | # @\$-eo |
|---|
| 1174 | |
|---|
| 1175 | %%END%% |
|---|
| 1176 | |
|---|
| 1177 | fi |
|---|
| 1178 | |
|---|
| 1179 | |
|---|
| 1180 | # BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP |
|---|
| 1181 | # BEI EXIT VERANLASSEN |
|---|
| 1182 | # VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT |
|---|
| 1183 | # AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH |
|---|
| 1184 | if [[ $delete_dayfile = false && $remote_host != $local_host ]] |
|---|
| 1185 | then |
|---|
| 1186 | echo "set +vx" >> $job_to_send |
|---|
| 1187 | echo "trap '" >> $job_to_send |
|---|
| 1188 | echo "set +vx" >> $job_to_send |
|---|
| 1189 | if [[ $(echo $remote_host | cut -c1-3) = ibm || $(echo $remote_host | cut -c1-5) = lcsgi || $(echo $remote_host | cut -c1-3) = nec ]] |
|---|
| 1190 | then |
|---|
| 1191 | if [[ $remote_host = ibmh ]] |
|---|
| 1192 | then |
|---|
| 1193 | return_queue=c1 |
|---|
| 1194 | elif [[ $remote_host = ibmkisti ]] |
|---|
| 1195 | then |
|---|
| 1196 | return_queue=class.1-2 |
|---|
| 1197 | elif [[ $remote_host = ibmku ]] |
|---|
| 1198 | then |
|---|
| 1199 | return_queue=sdbg2 |
|---|
| 1200 | elif [[ $remote_host = ibms ]] |
|---|
| 1201 | then |
|---|
| 1202 | return_queue=p_normal |
|---|
| 1203 | elif [[ $remote_host = ibmy ]] |
|---|
| 1204 | then |
|---|
| 1205 | return_queue=serial |
|---|
| 1206 | elif [[ $remote_host = lcsgih || $remote_host = lcsgib ]] |
|---|
| 1207 | then |
|---|
| 1208 | return_queue=serialq |
|---|
| 1209 | elif [[ $remote_host = necriam ]] |
|---|
| 1210 | then |
|---|
| 1211 | return_queue=SP |
|---|
| 1212 | else |
|---|
| 1213 | return_queue=unknown |
|---|
| 1214 | fi |
|---|
| 1215 | |
|---|
| 1216 | if [[ $(echo $remote_host | cut -c1-3) = ibm ]] |
|---|
| 1217 | then |
|---|
| 1218 | |
|---|
| 1219 | if [[ $remote_host = ibmku ]] |
|---|
| 1220 | then |
|---|
| 1221 | echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1222 | echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1223 | else |
|---|
| 1224 | echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1225 | fi |
|---|
| 1226 | echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1227 | echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1228 | echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1229 | echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send |
|---|
| 1230 | echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1231 | echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1232 | if [[ $host != "ibmh" ]] |
|---|
| 1233 | then |
|---|
| 1234 | echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1235 | fi |
|---|
| 1236 | echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1237 | echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1238 | |
|---|
| 1239 | echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1240 | echo "echo \" \" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1241 | |
|---|
| 1242 | echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1243 | echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1244 | if [[ $remote_host = ibmku ]] |
|---|
| 1245 | then |
|---|
| 1246 | echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1247 | fi |
|---|
| 1248 | echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send |
|---|
| 1249 | |
|---|
| 1250 | elif [[ $remote_host = nech ]] |
|---|
| 1251 | then |
|---|
| 1252 | echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send |
|---|
| 1253 | echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send |
|---|
| 1254 | echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send |
|---|
| 1255 | echo "#PBS -o last_job_transfer_protocol" >> $job_to_send |
|---|
| 1256 | echo "#PBS -j o" >> $job_to_send |
|---|
| 1257 | echo " " >> $job_to_send |
|---|
| 1258 | echo "set -x" >> $job_to_send |
|---|
| 1259 | echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send |
|---|
| 1260 | echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send |
|---|
| 1261 | echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send |
|---|
| 1262 | echo "%%END%%" >> $job_to_send |
|---|
| 1263 | |
|---|
| 1264 | elif [[ $remote_host = necriam ]] |
|---|
| 1265 | then |
|---|
| 1266 | echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send |
|---|
| 1267 | echo "#PBS -q $return_queue" >> $job_to_send |
|---|
| 1268 | echo "#PBS -o last_job_transfer_protocol" >> $job_to_send |
|---|
| 1269 | echo "#PBS -j o" >> $job_to_send |
|---|
| 1270 | echo " " >> $job_to_send |
|---|
| 1271 | echo "set -x" >> $job_to_send |
|---|
| 1272 | echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send |
|---|
| 1273 | echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send |
|---|
| 1274 | echo "%%END%%" >> $job_to_send |
|---|
| 1275 | |
|---|
| 1276 | elif [[ $(echo $remote_host | cut -c1-5) = lcsgi ]] |
|---|
| 1277 | then |
|---|
| 1278 | echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send |
|---|
| 1279 | echo "#!/bin/bash" >> $job_to_send |
|---|
| 1280 | echo "#PBS -N job_protocol_transfer" >> $job_to_send |
|---|
| 1281 | echo "#PBS -l walltime=00:30:00" >> $job_to_send |
|---|
| 1282 | echo "#PBS -l nodes=1:ppn=1" >> $job_to_send |
|---|
| 1283 | echo "#PBS -l feature=data" >> $job_to_send |
|---|
| 1284 | echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send |
|---|
| 1285 | echo "#PBS -j oe" >> $job_to_send |
|---|
| 1286 | echo " " >> $job_to_send |
|---|
| 1287 | echo ". /usr/share/modules/init/bash" >> $job_to_send |
|---|
| 1288 | echo "set -x" >> $job_to_send |
|---|
| 1289 | echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send |
|---|
| 1290 | echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send |
|---|
| 1291 | echo "%%END%%" >> $job_to_send |
|---|
| 1292 | |
|---|
| 1293 | else |
|---|
| 1294 | |
|---|
| 1295 | echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send |
|---|
| 1296 | echo "# @\\\$-q $return_queue" >> $job_to_send |
|---|
| 1297 | echo "# @\\\$-l${qsubtime} 10" >> $job_to_send |
|---|
| 1298 | echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send |
|---|
| 1299 | if [[ $remote_host = t3ej2 || $remote_host = t3ej5 || $remote_host = t3es ]] |
|---|
| 1300 | then |
|---|
| 1301 | echo "# @\$-l mpp_p=0" >> $job_to_send |
|---|
| 1302 | fi |
|---|
| 1303 | echo '# @\$-lF 10mb' >> $job_to_send |
|---|
| 1304 | # echo '# @\$-o /dev/null' >> $job_to_send |
|---|
| 1305 | echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send |
|---|
| 1306 | echo '# @\\\$-eo' >> $job_to_send |
|---|
| 1307 | echo " " >> $job_to_send |
|---|
| 1308 | if [[ $remote_host = t3ej2 || $remote_host = t3ej5 ]] |
|---|
| 1309 | then |
|---|
| 1310 | echo "set +vx" >> $job_to_send |
|---|
| 1311 | echo ". .profile" >> $job_to_send |
|---|
| 1312 | fi |
|---|
| 1313 | echo "set -x" >> $job_to_send |
|---|
| 1314 | echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send |
|---|
| 1315 | echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send |
|---|
| 1316 | echo "%%END%%" >> $job_to_send |
|---|
| 1317 | fi |
|---|
| 1318 | |
|---|
| 1319 | if [[ $(echo $remote_host | cut -c1-3) = ibm ]] |
|---|
| 1320 | then |
|---|
| 1321 | echo "llsubmit scpjob.$kennung" >> $job_to_send |
|---|
| 1322 | elif [[ $(echo $remote_host | cut -c1-5) = lcsgi ]] |
|---|
| 1323 | then |
|---|
| 1324 | echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send |
|---|
| 1325 | echo "chmod u+x scpjob.$kennung" >> $job_to_send |
|---|
| 1326 | echo "msub scpjob.$kennung" >> $job_to_send |
|---|
| 1327 | elif [[ $remote_host = t3eb || $remote_host = t3eh || $remote_host = t3ej2 || $remote_host = t3ej5 ]] |
|---|
| 1328 | then |
|---|
| 1329 | echo "qsub -J n scpjob.$kennung" >> $job_to_send |
|---|
| 1330 | elif [[ $remote_host = t3es ]] |
|---|
| 1331 | then |
|---|
| 1332 | echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send |
|---|
| 1333 | else |
|---|
| 1334 | echo "qsub scpjob.$kennung" >> $job_to_send |
|---|
| 1335 | fi |
|---|
| 1336 | if [[ $remote_host != ibmku ]] |
|---|
| 1337 | then |
|---|
| 1338 | echo "rm scpjob.$kennung" >> $job_to_send |
|---|
| 1339 | fi |
|---|
| 1340 | if [[ $remote_host = nech ]] |
|---|
| 1341 | then |
|---|
| 1342 | echo "cd -" >> $job_to_send |
|---|
| 1343 | fi |
|---|
| 1344 | else |
|---|
| 1345 | # echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send |
|---|
| 1346 | echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send |
|---|
| 1347 | fi |
|---|
| 1348 | echo "set -x" >> $job_to_send |
|---|
| 1349 | echo " ' exit" >> $job_to_send |
|---|
| 1350 | echo "set -x" >> $job_to_send |
|---|
| 1351 | fi |
|---|
| 1352 | |
|---|
| 1353 | |
|---|
| 1354 | |
|---|
| 1355 | # EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN |
|---|
| 1356 | cat $file_to_send >> $job_to_send |
|---|
| 1357 | |
|---|
| 1358 | if [[ $remote_host = ibm ]] |
|---|
| 1359 | then |
|---|
| 1360 | echo " " >> $job_to_send |
|---|
| 1361 | echo "exit" >> $job_to_send |
|---|
| 1362 | fi |
|---|
| 1363 | |
|---|
| 1364 | # remove job file |
|---|
| 1365 | if [[ $remote_host = lctit || $remote_host = ibmku ]] |
|---|
| 1366 | then |
|---|
| 1367 | echo " " >> $job_to_send |
|---|
| 1368 | echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send |
|---|
| 1369 | fi |
|---|
| 1370 | |
|---|
| 1371 | |
|---|
| 1372 | |
|---|
| 1373 | # USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN |
|---|
| 1374 | if [[ -z $remote_user ]] |
|---|
| 1375 | then |
|---|
| 1376 | if [[ $remote_host = t3eb || $remote_host = t3eh || $remote_host = t3ej2 || $remote_host = t3ej5 || $remote_host = t3es || $remote_host = vpp ]] |
|---|
| 1377 | then |
|---|
| 1378 | grep $remote_addres ~/.netrc | read dum dum dum remote_user dum dum |
|---|
| 1379 | fi |
|---|
| 1380 | fi |
|---|
| 1381 | |
|---|
| 1382 | |
|---|
| 1383 | |
|---|
| 1384 | # JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN |
|---|
| 1385 | if [[ $no_submit = false ]] |
|---|
| 1386 | then |
|---|
| 1387 | if [[ $remote_host != $local_host ]] |
|---|
| 1388 | then |
|---|
| 1389 | [[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..." |
|---|
| 1390 | if [[ $remote_host = ibms || $remote_host = ibmy ]] # ssh on ibms cannot handle "~/" |
|---|
| 1391 | then |
|---|
| 1392 | job_catalog_save=$job_catalog |
|---|
| 1393 | job_catalog=job_queue |
|---|
| 1394 | elif [[ $remote_host = nech ]] |
|---|
| 1395 | then |
|---|
| 1396 | job_catalog_save=$job_catalog |
|---|
| 1397 | job_catalog=/hpf/b/${remote_user}/job_queue |
|---|
| 1398 | fi |
|---|
| 1399 | if [[ $local_host = decalpha ]] |
|---|
| 1400 | then |
|---|
| 1401 | # VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE |
|---|
| 1402 | # PASSWORT AB |
|---|
| 1403 | /bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost |
|---|
| 1404 | elif [[ $remote_host = nech ]] |
|---|
| 1405 | then |
|---|
| 1406 | # DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ |
|---|
| 1407 | # TRANSFERIERT WERDEN |
|---|
| 1408 | scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost |
|---|
| 1409 | else |
|---|
| 1410 | scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost |
|---|
| 1411 | fi |
|---|
| 1412 | if [[ $? = 1 ]] |
|---|
| 1413 | then |
|---|
| 1414 | locat=scp; exit |
|---|
| 1415 | fi |
|---|
| 1416 | if [[ $remote_host = ibms ]] |
|---|
| 1417 | then |
|---|
| 1418 | job_catalog=$job_catalog_save |
|---|
| 1419 | fi |
|---|
| 1420 | [[ $verify = true ]] && printf "\n >>> finished\n" |
|---|
| 1421 | else |
|---|
| 1422 | eval job_catalog=$job_catalog |
|---|
| 1423 | cp $job_to_send ${job_catalog}/$job_on_remhost |
|---|
| 1424 | fi |
|---|
| 1425 | |
|---|
| 1426 | |
|---|
| 1427 | |
|---|
| 1428 | # NQS- BZW. LOADLEVELER-JOB STARTEN |
|---|
| 1429 | if [[ $remote_host != $local_host ]] |
|---|
| 1430 | then |
|---|
| 1431 | [[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n" |
|---|
| 1432 | |
|---|
| 1433 | if [[ $(echo $remote_host | cut -c1-5) = lcsgi && $prio = true ]] |
|---|
| 1434 | then |
|---|
| 1435 | printf "\n >>> submit with HLRN qos-feature hiprio...\n" |
|---|
| 1436 | ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost" |
|---|
| 1437 | elif [[ $remote_host = ibmku ]] |
|---|
| 1438 | then |
|---|
| 1439 | ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost" |
|---|
| 1440 | else |
|---|
| 1441 | ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost" |
|---|
| 1442 | fi |
|---|
| 1443 | |
|---|
| 1444 | [[ $verify = true ]] && printf " >>> o.k.\n" |
|---|
| 1445 | else |
|---|
| 1446 | cd $job_catalog |
|---|
| 1447 | if [[ $(echo $local_host | cut -c1-5) = lcsgi || $(echo $local_host | cut -c1-3) = ibm ]] |
|---|
| 1448 | then |
|---|
| 1449 | eval $submcom $job_on_remhost |
|---|
| 1450 | elif [[ $local_host = lcfimm || $local_host = lctit || $localhost = lcxe6 || $localhost = lck || $localhost = lckordi || $localhost = lcyon || $localhost = lcsb ]] |
|---|
| 1451 | then |
|---|
| 1452 | chmod u+x $job_on_remhost |
|---|
| 1453 | echo "$submcom $job_on_remhost" |
|---|
| 1454 | eval $submcom $job_on_remhost |
|---|
| 1455 | elif [[ $local_host = nech ]] |
|---|
| 1456 | then |
|---|
| 1457 | if [[ $queue = default ]] |
|---|
| 1458 | then |
|---|
| 1459 | eval $submcom $job_on_remhost |
|---|
| 1460 | else |
|---|
| 1461 | eval $submcom -q $queue $job_on_remhost |
|---|
| 1462 | fi |
|---|
| 1463 | else |
|---|
| 1464 | qsub $job_on_remhost |
|---|
| 1465 | fi |
|---|
| 1466 | |
|---|
| 1467 | # Jobfile must not be deleted on lctit/ibmku!! This will be done |
|---|
| 1468 | # only at the end of the job. |
|---|
| 1469 | if [[ $local_host != lctit && $local_host != ibmku ]] |
|---|
| 1470 | then |
|---|
| 1471 | rm $job_on_remhost |
|---|
| 1472 | fi |
|---|
| 1473 | cd - > /dev/null |
|---|
| 1474 | fi |
|---|
| 1475 | fi |
|---|
| 1476 | |
|---|
| 1477 | |
|---|
| 1478 | |
|---|
| 1479 | # ABSCHLUSSARBEITEN |
|---|
| 1480 | |
|---|
| 1481 | if [[ $no_submit = false ]] |
|---|
| 1482 | then |
|---|
| 1483 | rm $job_to_send |
|---|
| 1484 | fi |
|---|
| 1485 | [[ $verify = true ]] && printf "\n\n *** SUBJOB finished \n\n" |
|---|
