Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1453

Last change on this file since 1453 was 1453, checked in by heinze, 10 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 50.7 KB

Rev	Line
[1]	1	#!/bin/ksh
[1090]	2
	3	# subjob - script for automatic generation and submission of batch-job files
	4	# for various batch queuing systems
	5
[1046]	6	#--------------------------------------------------------------------------------#
	7	# This file is part of PALM.
	8	#
	9	# PALM is free software: you can redistribute it and/or modify it under the terms
	10	# of the GNU General Public License as published by the Free Software Foundation,
	11	# either version 3 of the License, or (at your option) any later version.
	12	#
	13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
	14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
	15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
	16	#
	17	# You should have received a copy of the GNU General Public License along with
	18	# PALM. If not, see <http://www.gnu.org/licenses/>.
	19	#
[1310]	20	# Copyright 1997-2014 Leibniz Universitaet Hannover
[1046]	21	#--------------------------------------------------------------------------------#
	22	#
	23	# Current revisions:
[1090]	24	# ------------------
[1351]	25	#
[1453]	26	#
[1046]	27	# Former revisions:
	28	# -----------------
[169]	29	# $Id: subjob 1453 2014-08-22 09:42:07Z heinze $
[1200]	30	#
[1453]	31	# 1452 2014-08-22 09:41:06Z heinze
	32	# local hosts for blizzard added
	33	#
[1451]	34	# 1450 2014-08-21 07:31:51Z heinze
	35	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
	36	#
[1443]	37	# 1442 2014-07-28 07:09:10Z raasch
	38	# HLRN-III (lccrayb/lccrayh) queues adjusted
	39	#
[1379]	40	# 1378 2014-04-28 06:04:58Z raasch
	41	# -et option added for lctit
	42	#
[1351]	43	# 1350 2014-04-04 13:01:30Z maronga
	44	# location of qsub updated for lcxe6
	45	#
[1290]	46	# 1289 2014-03-04 07:12:34Z raasch
	47	# German comments translated to English
	48	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
	49	#
[1280]	50	# 1279 2014-01-28 12:10:14Z raasch
	51	# node calculation modified due to changes in mrun (tasks_per_node must not be
	52	# an integral divisor of numprocs any more)
	53	#
[1275]	54	# 1274 2014-01-09 13:14:54Z heinze
	55	# adjustments for lccrayh
	56	#
[1267]	57	# 1266 2013-12-11 12:07:34Z heinze
	58	# further adjustments for lccrayb (use msub instead of qsub)
	59	#
[1265]	60	# 1264 2013-12-09 12:46:09Z fricke
	61	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	62	#
[1263]	63	# 1262 2013-12-09 10:57:20Z fricke
	64	# further adjustments for lccrayb
	65	#
[1261]	66	# 1260 2013-12-04 12:48:04Z raasch
	67	# jaboticaba admitted
	68	#
[1256]	69	# 1255 2013-11-07 14:43:35Z raasch
	70	# further adjustments for lccrayb
	71	#
[1225]	72	# 1224 2013-09-16 07:27:23Z raasch
	73	# first adjustments for lccrayb
	74	#
[1203]	75	# 1202 2013-07-10 16:22:07Z witha
	76	# adjustments for Forwind cluster (lcflow)
	77	#
[1200]	78	# 1199 2013-07-05 14:52:22Z raasch
	79	# adjustments for CSC Helsinki (lccrayf)
	80	#
[1185]	81	# use of cluster/express queue enabled (ibmh)
	82	# vinessa added (imuk)
[1047]	83	#
[1104]	84	# 1103 2013-02-20 02:15:53Z raasch
	85	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	86	# instead typeset -L/R),
	87	# further adjustments for lckyuh
	88	#
[1100]	89	# 2013-02-10 01:47:43Z raasch
	90	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	91	# and for Forwind cluster (lcflow)
	92	#
[1096]	93	# 1094 2013-02-03 01:52:12Z raasch
	94	# new option -P for explicit setting of ssh/scp port,
	95	# decalpha parts (yonsei) removed
	96	#
[1091]	97	# 2013-02-02 07:06:13Z raasch
[1099]	98	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	99	# old changelog messages removed
	100	#
[1047]	101	# 1046 2012-11-09 14:38:45Z maronga
	102	# code put under GPL (PALM 3.9)
	103	#
[1090]	104	# 08/07/94 - Siggi - first version finished
	105	# 29/06/94 - Siggi - script development started
	106	#--------------------------------------------------------------------------------#
	107	# subjob - script for automatic generation and submission of batch-job files
	108	# for various batch queuing systems
	109	#--------------------------------------------------------------------------------#
[1]	110
	111
[1289]	112	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	113	delete_dayfile=false
[799]	114	email_notification=none
[122]	115	group_number=none
[1]	116	locat=normal
	117	no_default_queue=none
	118	no_submit=false
	119	job_catalog="~/job_queue"
	120	job_name=none
	121	local_user=$LOGNAME
	122	node_usage=shared
[475]	123	numprocs=0
[1]	124	punkte="..........................................................."
	125	submcom=qsub
	126	queue=default
	127	remote_host=none
	128	remote_user=""
	129	verify=true
	130
	131	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	132	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	133
	134
	135
[1289]	136	# ERROR HANDLING
	137	# IN CASE OF EXIT:
[1]	138	trap 'if [[ $locat != normal ]]
	139	then
	140	case $locat in
	141	(option) printf "\n --> available optios can be displayed"
	142	printf " by typing:"
	143	printf "\n \"subjob ?\" \n";;
	144	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	145	(*) printf "\n +++ unknown error"
	146	printf "\n please inform S. Raasch!\n"
	147	esac
	148	[[ -f $job_to_send ]] && rm $job_to_send
	149	printf "\n\n+++ SUBJOB killed \n\n"
	150	fi' exit
	151
	152
[1289]	153	# IN CASE OF TERMINAL-BREAK:
[1]	154	trap '[[ -f $job_to_send ]] && rm $job_to_send
	155	printf "\n\n+++ SUBJOB killed \n\n"
	156	exit
	157	' 2
	158
	159
	160
	161
[1289]	162	# DETERMINE NAME OF LOCAL HOST
[1]	163	local_host=$(hostname)
	164
	165
	166
[1289]	167	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	168	# IS ADMITTED AT ALL)
	169	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	170	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	171	case $local_host in
[1204]	172	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
[1]	173	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[1235]	174	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
[102]	175	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	176	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1103]	177	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
[1452]	178	(blizzard1\|p*) local_addres=136.172.40.15; local_host=ibmh;;
	179	(blizzard2\|p*) local_addres=136.172.40.16; local_host=ibmh;;
[1255]	180	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
[1274]	181	(hlogin\|hxc) local_addres=130.75.4.1; local_host=lccrayh;;
[12]	182	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[1174]	183	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
[372]	184	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	185	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[1197]	186	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
[1]	187	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	188	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[1099]	189	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
	190	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
[367]	191	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	192	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	193	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	194	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[5]	195	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	196	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1099]	197	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
[892]	198	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[440]	199	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	200	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	201	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[1260]	202	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
[407]	203	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	204	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
[1040]	205	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
[5]	206	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1099]	207	(meller) local_addres=134.106.74.155; local_host=lcfor;;
[437]	208	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	209	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	210	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	211	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	212	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	213	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[1099]	214	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
[789]	215	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	216	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	217	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1204]	218	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
[1197]	219	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
[799]	220	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1090]	221	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
[635]	222	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1021]	223	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
[1184]	224	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
[1]	225	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	226	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	227	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	228	printf "\n please contact the PALM group at IMUK";
[1]	229	locat=parameter; exit;;
	230	esac
	231
	232
	233
[1289]	234	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	235	remote_host=$local_host
	236
	237
	238
	239
[1289]	240	# READ THE SHELLSCRIPT-OPTIONS
[1094]	241	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	242	do
	243	case $option in
	244	(c) job_catalog=$OPTARG;;
	245	(d) delete_dayfile=true;;
	246	(D) no_submit=true;;
[352]	247	(e) email_notification=$OPTARG;;
[125]	248	(g) group_number=$OPTARG;;
[1]	249	(h) remote_host=$OPTARG;;
	250	(m) memory=$OPTARG;;
	251	(n) job_name=$OPTARG;;
	252	(N) node_usage=$OPTARG;;
	253	(O) threads_per_task=$OPTARG;;
[1094]	254	(P) scp_port=$OPTARG;;
[1]	255	(q) no_default_queue=$OPTARG;;
	256	(t) cputime=$OPTARG;;
	257	(T) tasks_per_node=$OPTARG;;
	258	(u) remote_user=$OPTARG;;
	259	(v) verify=false;;
	260	(X) numprocs=$OPTARG;;
	261	(\?) printf "\n +++ Option $OPTARG unknown \n";
	262	locat=option; exit;;
	263	esac
	264	done
	265
	266
[1289]	267	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	268	(( to_shift = $OPTIND - 1 ))
	269	shift $to_shift; file_to_send=$1
[1]	270
	271
[1289]	272	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	273	if [ "$1" = "?" ]
	274	then
	275	(printf "\n *** subjob can be called as follows:\n"
	276	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	277	printf "\n Description of available options:\n"
	278	printf "\n Option Description Default-Value"
	279	printf "\n -c job-input- and output-catalog ~/job_queue"
	280	printf "\n -d no job-protocol will be created ---"
	281	printf "\n -D only the job-file will be created ---"
	282	printf "\n -h execution host, available hosts: $remote_host"
[1289]	283	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	284	printf "\n lckiaps, lctit, nech"
[1]	285	printf "\n -m memory demand per process in MByte ---"
	286	printf "\n -n jobname <jobdatei>"
	287	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	288	printf "\n -P ssh/scp port default port"
[1]	289	printf "\n -q job-queue to be used default"
	290	printf "\n -t allowed cpu-time in seconds ---"
	291	printf "\n -T tasks per node (on parallel hosts) ---"
	292	printf "\n -u username on execution host from .netrc"
	293	printf "\n -v no prompt for confirmation ---"
	294	printf "\n -X # of processors (on parallel hosts) 1"
	295	printf "\n "
	296	printf "\n The only possible positional parameter is <jobfile>:"
	297	printf "\n The complete NQS-job must be provided here."
	298	printf "\n <jobfile>=? creates this outline\n\n") \| more
	299	exit
	300	fi
	301
	302
	303
[1289]	304	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	305	if [[ "$file_to_send" = "" ]]
	306	then
	307	printf "\n +++ job-file missing"
	308	locat=parameter; exit
	309	else
	310	if [[ -f $file_to_send ]]
	311	then
	312	true
	313	else
	314	printf "\n +++ job-file: "
	315	printf "\n $file_to_send"
	316	printf "\n does not exist"
	317	locat=parameter; exit
	318	fi
	319	fi
	320
	321
	322
[1289]	323	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	324	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	325	if [[ $job_name = none ]]
	326	then
	327	job_name=$file_to_send
	328	fi
	329	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	330	then
	331	printf "\n +++ job-file name: "
	332	printf "\n $job_name"
	333	printf "\n must not contain \"/\"-characters"
	334	locat=parameter; exit
	335	fi
	336
	337
	338
	339
[1289]	340	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	341	# OR IF NO HOST HAS BEEN GIVEN
[1]	342	if [[ $remote_host = none ]]
	343	then
	344	printf "\n +++ host missing"
	345	locat=option; exit
	346	else
	347	case $remote_host in
[251]	348	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1184]	349	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	350	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	351	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	352	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1450]	353	(lccrayb) queue=mpp1testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1274]	354	(lccrayh) queue=mpp1testq; remote_addres=130.75.4.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1197]	355	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
[1099]	356	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
[440]	357	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[367]	358	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[1040]	359	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
[717]	360	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[1099]	361	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
[1090]	362	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
[1021]	363	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
[635]	364	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[1350]	365	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
[437]	366	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	367	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	368	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	369	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	370	locat=parameter; exit;;
[1]	371	esac
	372	fi
	373
	374
[1289]	375	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	376	if [[ $no_default_queue != none ]]
	377	then
	378	error=false
	379	ndq=$no_default_queue
	380	case $remote_host in
	381	(ibm) case $ndq in
	382	(p690_express\|p690_standard\|p690_long) error=false;;
	383	(*) error=true;;
	384	esac;;
	385	(ibmh) case $ndq in
[1184]	386	(cluster\|express) error=false;;
[1]	387	(*) error=true;;
	388	esac;;
[693]	389	(ibmkisti) case $ndq in
	390	(class.32plus\|class.1-2\|class.2-32) error=false;;
	391	(*) error=true;;
	392	esac;;
[622]	393	(ibmku) case $ndq in
	394	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	395	(*) error=true;;
	396	esac;;
[1]	397	(ibms) case $ndq in
	398	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	399	(*) error=true;;
	400	esac;;
[1224]	401	(lccrayb) case $ndq in
[1442]	402	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1224]	403	(*) error=true;;
	404	esac;;
[1274]	405	(lccrayh) case $ndq in
[1442]	406	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1274]	407	(*) error=true;;
	408	esac;;
[1197]	409	(lccrayf) case $ndq in
	410	(usup\|test*\|small\|large) error=false;;
	411	(*) error=true;;
	412	esac;;
[1099]	413	(lcflow) case $ndq in
	414	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
	415	(*) error=true;;
	416	esac;;
[1040]	417	(lckiaps) case $ndq in
	418	(express\|normal) error=false;;
	419	(*) error=true;;
	420	esac;;
[440]	421	(lckyoto) case $ndq in
	422	(eh\|ph) error=false;;
	423	(*) error=true;;
	424	esac;;
[1099]	425	(lckyuh) case $ndq in
	426	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	427	(*) error=true;;
	428	esac;;
[1090]	429	(lckyut) case $ndq in
	430	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	431	(*) error=true;;
	432	esac;;
[1]	433	(lctit) case $ndq in
[635]	434	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	435	(*) error=true;;
	436	esac;;
	437	(t3eb) case $ndq in
	438	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	439	(*) error=true;;
	440	esac;;
	441	(t3eh) case $ndq in
	442	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	443	(*) error=true;;
	444	esac;;
	445	(t3ej2\|t3ej5) case $ndq in
	446	(low\|normal\|high) error=false;;
	447	(*) error=true;;
	448	esac;;
	449	(t3es) case $ndq in
	450	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	451	(*) error=true;;
	452	esac;;
	453	esac
	454	if [[ $error = true ]]
	455	then
	456	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	457	locat=parameter; exit
	458	else
	459	queue=$no_default_queue
	460	fi
	461	fi
	462
	463
	464
[1289]	465	# CHECK THE CPU-TIME
	466	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	467	done=false
	468	while [[ $done = false ]]
	469	do
	470	if (( $cputime <= 0 ))
	471	then
	472	printf "\n +++ wrong cpu-time or cpu-time missing"
	473	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	474	printf "\n >>> "
	475	read cputime 1>/dev/null 2>&1
	476	else
	477	done=true
	478	fi
	479	done
	480	if [[ $remote_host = nech ]]
	481	then
	482	if (( tasks_per_node != 0 ))
	483	then
	484	(( cputime = cputime * tasks_per_node ))
	485	elif [[ $numprocs != 0 ]]
	486	then
	487	(( cputime = cputime * numprocs ))
	488	fi
	489	fi
	490	(( stunden = cputime / 3600 ))
	491	(( resttime = cputime - stunden * 3600 ))
	492	(( minuten = resttime / 60 ))
	493	(( sekunden = resttime - minuten * 60 ))
	494	timestring=${stunden}:${minuten}:${sekunden}
	495
	496
	497
[1289]	498	# CHECK THE MEMORY DEMAND
[1]	499	done=false
	500	while [[ $done = false ]]
	501	do
	502	if (( memory <= 0 ))
	503	then
	504	printf "\n +++ wrong memory demand or memory demand missing"
	505	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	506	printf "\n >>> "
	507	read memory 1>/dev/null 2>&1
	508	else
	509	done=true
	510	fi
	511	done
	512
[1289]	513	if [[ $remote_host = nech ]]
[1]	514	then
	515	if (( tasks_per_node != 0 ))
	516	then
	517	(( Memory = memory * tasks_per_node / 1000 ))
	518	elif [[ $numprocs != 0 ]]
	519	then
	520	(( Memory = memory * numprocs / 1000 ))
	521	else
	522	(( Memory = memory / 1000 ))
	523	fi
[635]	524	elif [[ $remote_host = lctit ]]
	525	then
	526	(( Memory = memory * tasks_per_node / 1000 ))
[1]	527	fi
	528
	529
[1289]	530	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	531	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	532	then
	533	(( memory = memory * threads_per_task ))
	534	fi
	535
	536
[1289]	537	# CALCULATE NUMBER OF REQUIRED NODES
[1]	538	if (( tasks_per_node != 0 ))
	539	then
[1279]	540	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	541	fi
	542
[1094]	543
[1289]	544	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	545	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	546
[1094]	547
[1289]	548	# CALCULATE NUMBER OF MPI TASKS
[696]	549	(( mpi_tasks = numprocs / threads_per_task ))
[1]	550
[696]	551
[1289]	552	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	553	if [[ "$scp_port" != "" ]]
	554	then
	555	PORTOPT="-P $scp_port"
	556	SSH_PORTOPT="-p $scp_port"
	557	fi
	558
	559
[1289]	560	# HEADER-OUTPUT
[1]	561	if [[ $verify = true ]]
	562	then
	563	printf "\n\n"
	564	printf "#--------------------------------------------------------------# \n"
	565	spalte1=SUBJOB;spalte2=$(date)
[1103]	566	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	567	printf "\| \| \n"
	568	printf "\| values of parameters/options: \| \n"
[1103]	569	spalte1=$(echo local_host$punkte \| cut -c-20)
	570	spalte2=$punkte$local_host
	571	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	572	spalte1=$(echo remote_host$punkte \| cut -c-20)
	573	spalte2=$punkte$remote_host
	574	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	575	spalte1=$(echo queue$punkte \| cut -c-20)
	576	spalte2=$punkte$queue
	577	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	578	spalte1=$(echo memory$punkte \| cut -c-20)
	579	spalte2="$punkte$memory mb"
	580	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	581	spalte1=$(echo cputime$punkte \| cut -c-20)
	582	spalte2="$punkte$cputime sec"
	583	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	584	spalte1=$(echo job_name$punkte \| cut -c-20)
	585	spalte2="$punkte$job_name"
	586	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	587	printf "#--------------------------------------------------------------# \n\n"
	588
	589
[1289]	590	# QUERY CHECK
[1]	591	antwort="dummy"
	592	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	593	do
	594	read antwort?" >>> continue (y/n) ? "
	595	done
	596	if [[ $antwort = n \|\| $antwort = N ]]
	597	then
	598	locat=verify; exit
	599	fi
	600	printf "\n"
	601	fi
	602
	603
	604
[1289]	605	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	606	identifier=$RANDOM
	607	job_on_remhost=${job_name}_${identifier}_$local_host
	608	job_to_send=job_to_send_$identifier
[1]	609	if [[ $delete_dayfile = false ]]
	610	then
[1289]	611	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	612	local_dayfile=${remote_host}_${job_name}
	613	else
	614	remote_dayfile=/dev/null
	615	fi
	616
	617
	618
[1289]	619	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	620	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	621	then
	622
[1289]	623	# GENERAL LOADLEVELER SETTINGS
[622]	624	execute_in_shell="#!/bin/ksh"
	625	use_shell="# @ shell = /bin/ksh"
	626	consumable_memory="ConsumableMemory($memory mb)"
	627	class="# @ class = $queue"
	628	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	629	network_to_use="# @ network.mpi = sn_all,shared,us"
	630	data_limit="# @ data_limit = 1.76gb"
	631	image_size="# @ image_size = 50"
[693]	632	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	633
[693]	634	if [[ $email_notification = none ]]
	635	then
	636	notify_user=""
	637	else
	638	notify_user="# @ notify_user = $email_notification"
	639	if [[ $delete_dayfile = true ]]
	640	then
	641	notification='# @ notification = never'
	642	fi
	643	fi
[622]	644
[312]	645	if [[ $remote_host = ibmh ]]
[1]	646	then
[312]	647	data_limit=""
	648	network_to_use=""
[1184]	649	class="# @ class = $queue"
[312]	650	environment=""
[814]	651	rset="# @ rset = RSET_MCM_AFFINITY"
	652	task_affinity="# @ task_affinity = core(1)"
[693]	653	elif [[ $remote_host = ibmkisti ]]
	654	then
	655	network_to_use="# @ network.MPI = sn_all,shared,US"
	656	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	657	if [[ $threads_per_task = 1 ]]
	658	then
	659	rset="# @ rset = RSET_MCM_AFFINITY"
	660	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	661	fi
[693]	662	environment=""
	663	use_shell=""
	664	data_limit=""
	665	image_size=""
[622]	666	elif [[ $remote_host = ibmku ]]
	667	then
	668	execute_in_shell="#!/usr/bin/ksh"
	669	use_shell="# @ shell = /usr/bin/ksh"
	670	consumable_memory=""
	671	environment=""
	672	network_to_use="# @ network.mpi = sn_all,shared,us"
	673	data_limit=""
	674	image_size=""
	675	elif [[ $remote_host = ibms ]]
	676	then
	677	network_to_use="# @ network.mpi = csss,shared,us"
[1]	678	fi
	679
	680	cat > $job_to_send << %%END%%
[622]	681	$execute_in_shell
	682	$use_shell
[1]	683
	684	# @ job_type = parallel
[693]	685	# @ job_name = $job_name
[1]	686	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	687	# @ output = $remote_dayfile
	688	# @ error = $remote_dayfile
[693]	689	$wall_clock_limit
[622]	690	$image_size
[312]	691	$class
	692	$environment
[1]	693	$network_to_use
	694	$data_limit
[693]	695	$rset
	696	$mcm_affinity_options
[814]	697	$task_affinity
[1]	698	$notification
[693]	699	$notify_user
[1]	700
	701	%%END%%
	702
	703	if (( nodes > 0 ))
	704	then
	705
[693]	706	if [[ $remote_host != ibmkisti ]]
	707	then
	708
	709	cat >> $job_to_send << %%END%%
[1]	710	# @ node = $nodes
[622]	711	# @ tasks_per_node = $processes_per_node
[1]	712	# @ node_usage = $node_usage
	713	# @ queue
	714
	715	%%END%%
	716
[693]	717	else
	718
	719	cat >> $job_to_send << %%END%%
[696]	720	# @ total_tasks = $mpi_tasks
[693]	721	# @ blocking = unlimited
	722	# @ queue
	723
	724	%%END%%
	725
	726	fi
	727
[1]	728	else
	729
[1289]	730	cat >> $job_to_send << %%END%%
[1]	731	# @ blocking = unlimited
	732	# @ total_tasks = $numprocs
	733	# @ node_usage = $node_usage
	734	# @ queue
	735
	736	%%END%%
	737
	738	fi
	739
[1289]	740	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	741	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	742	then
	743	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	744	fi
	745
[1]	746	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	747	then
	748
	749	cat > $job_to_send << %%END%%
	750	#!/bin/ksh
	751
	752	# @ job_type = serial
	753	# @ node_usage = $node_usage
	754	# @ job_name = palm
	755	# @ wall_clock_limit = ${timestring},$timestring
	756	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	757	# @ output = $remote_dayfile
	758	# @ error = $remote_dayfile
[312]	759	$class
[1]	760	$notification
	761
	762	# @ queue
	763
	764	%%END%%
	765
[1274]	766	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	767	then
	768
	769	if [[ $numprocs != 0 ]]
	770	then
	771	cat > $job_to_send << %%END%%
[1255]	772	#!/bin/bash
[1224]	773	#PBS -N $job_name
	774	#PBS -l walltime=$timestring
[1264]	775	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	776	#PBS -o $remote_dayfile
	777	#PBS -j oe
	778	#PBS -q $queue
	779
	780	$module_calls
	781
	782	%%END%%
	783
	784	else
	785
	786	continue
	787
	788	fi
	789
[1197]	790	elif [[ $remote_host = lccrayf ]]
	791	then
	792
	793	if [[ $numprocs != 0 ]]
	794	then
	795	cat > $job_to_send << %%END%%
	796	#!/bin/bash -l
	797	#SBATCH -J $job_name
	798	#SBATCH -t $timestring
	799	#SBATCH -N $nodes
	800	#SBATCH --ntasks-per-node=$processes_per_node
	801	#SBATCH -p $queue
	802	#SBATCH -o $remote_dayfile
	803	#SBATCH -e $remote_dayfile
	804
	805	$init_cmds
	806	$module_calls
	807
	808	%%END%%
	809
	810	else
	811	cat > $job_to_send << %%END%%
	812	#!/bin/bash -l
	813	#SBATCH -J $job_name
	814	#SBATCH -t $timestring
	815	#SBATCH -l ncpus=1
	816	#SBATCH -l pmem=${memory}mb
	817	#SBATCH -m abe
	818	#SBATCH -o $remote_dayfile
	819	#SBATCH -e $remote_dayfile
	820
	821	$init_cmds
	822	$module_calls
	823
	824	%%END%%
	825
	826	fi
	827
[1099]	828	elif [[ $remote_host = lcflow ]]
	829	then
[1202]	830	if [ $memory -gt 1800 ]; then
[1099]	831	use_himem=""
	832	else
	833	use_himem="#"
	834	fi
	835	if [[ $numprocs != 0 ]]
	836	then
[1202]	837	pe_set="#$ -pe impi41 $numprocs"
[1099]	838	else
[1202]	839	pe_set="#$ -pe impi41 1"
[1099]	840	fi
	841	if [[ $queue = default ]]
	842	then
	843	queue_set=""
	844	else
	845	queue_set="#$ -q $queue"
	846	fi
	847	[[ "$disc_space" = "" ]] && disc_space=50
	848
	849	cat > $job_to_send << %%END%%
	850	#!/bin/bash
	851	#$ -S /bin/bash
	852	#$ -N $job_name
	853	#$ -cwd
	854	#$ -l h_rt=$timestring
	855	#$ -l h_vmem=${memory}M
	856	#$ -o $remote_dayfile
	857	#$ -j y
	858	$pe_set
	859	#$ -R y
	860	${use_himem}#$ -l highmem=true
	861	#$ -l h_fsize=${disc_space}G
	862	$queue_set
	863
	864	%%END%%
	865
[1021]	866	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	867	then
	868
	869	if [[ $numprocs != 0 ]]
	870	then
	871	cat > $job_to_send << %%END%%
	872	#!/bin/ksh
	873	#PBS -N $job_name
	874	#PBS -l walltime=$timestring
	875	#PBS -l ncpus=$numprocs
	876	#PBS -l pmem=${memory}mb
	877	#PBS -o $remote_dayfile
[1021]	878	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	879	#PBS -j oe
	880
	881	mpd &
	882
	883	%%END%%
	884
	885	else
	886	cat > $job_to_send << %%END%%
	887	#!/bin/ksh
	888	#PBS -N $job_name
	889	#PBS -l walltime=$timestring
	890	#PBS -l ncpus=1
	891	#PBS -l pmem=${memory}mb
	892	#PBS -o $remote_dayfile
	893	#PBS -j oe
	894
	895	%%END%%
	896
	897	fi
	898
[1040]	899	elif [[ $remote_host = lckiaps ]]
	900	then
	901
	902	if [[ $numprocs != 0 ]]
	903	then
	904	cat > $job_to_send << %%END%%
	905	#!/bin/ksh
	906	#PBS -N $job_name
	907	#PBS -l walltime=$timestring
	908	#PBS -l select=1:ncpus=$numprocs
	909	#PBS -l pmem=${memory}mb
	910	#PBS -q $queue
	911	#PBS -o $remote_dayfile
	912	#PBS -j oe
	913	#PBS -V
	914
	915	%%END%%
	916
	917	else
	918	cat > $job_to_send << %%END%%
	919	#!/bin/ksh
	920	#PBS -N $job_name
	921	#PBS -l walltime=$timestring
	922	#PBS -l ncpus=1
	923	#PBS -l pmem=${memory}mb
	924	#PBS -o $remote_dayfile
	925	#PBS -j oe
	926
	927	%%END%%
	928
	929	fi
	930
[693]	931	elif [[ $remote_host = lcyon ]]
	932	then
	933
	934	if [[ $numprocs != 0 ]]
	935	then
	936	cat > $job_to_send << %%END%%
	937	#!/bin/ksh
	938	#PBS -N $job_name
	939	#PBS -l walltime=$timestring
	940	#PBS -l ncpus=$numprocs
	941	#PBS -l pmem=${memory}mb
	942	#PBS -o $remote_dayfile
	943	#PBS -j oe
	944
	945	%%END%%
	946
	947	else
	948	cat > $job_to_send << %%END%%
	949	#!/bin/ksh
	950	#PBS -N $job_name
	951	#PBS -l walltime=$timestring
	952	#PBS -l ncpus=1
	953	#PBS -l pmem=${memory}mb
	954	#PBS -o $remote_dayfile
	955	#PBS -j oe
	956
	957	%%END%%
	958
	959	fi
	960
[892]	961	elif [[ $remote_host = lcxe6 ]]
[164]	962	then
	963
	964	if [[ $numprocs != 0 ]]
	965	then
	966	cat > $job_to_send << %%END%%
[799]	967	#!/bin/ksh
[164]	968	#PBS -S /bin/ksh
	969	#PBS -N $job_name
[552]	970	#PBS -A $project_account
[206]	971	#PBS -j oe
[164]	972	#PBS -l walltime=$timestring
	973	#PBS -l mppwidth=${numprocs}
[622]	974	#PBS -l mppnppn=${processes_per_node}
[164]	975	#PBS -m abe
	976	#PBS -o $remote_dayfile
[492]	977	$email_directive
[164]	978
[892]	979	$init_cmds
[493]	980	$module_calls
[343]	981
[164]	982	%%END%%
	983
[108]	984	else
	985	cat > $job_to_send << %%END%%
	986	#!/bin/ksh
[168]	987	#PBS -S /bin/ksh
[108]	988	#PBS -N $job_name
[552]	989	#PBS -A $project_account
[206]	990	#PBS -j oe
[108]	991	#PBS -l walltime=$timestring
	992	#PBS -l ncpus=1
	993	#PBS -l pmem=${memory}mb
	994	#PBS -m abe
[492]	995	$email_directive
[108]	996	#PBS -o $remote_dayfile
	997
[892]	998	$init_cmds
[493]	999	$module_calls
[343]	1000
[108]	1001	%%END%%
	1002
	1003	fi
	1004
[440]	1005	elif [[ $remote_host = lckyoto ]]
[437]	1006	then
	1007
[440]	1008	cat > $job_to_send << %%END%%
[799]	1009	#!/bin/ksh
[440]	1010	# @\$-o $remote_dayfile
	1011	# @\$-eo -oi
	1012	# @\$-lP 16
[799]	1013	# @\$-lp 1
[440]	1014	# @\$-lm 28gb -llm unlimited -ls unlimited
	1015	# @\$-q $queue
	1016	# @\$-Pvn abs_pack
	1017	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1018	#. /thin/local/etc/setprofile/intel-11.0.sh
	1019	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1020	. ~/.myprofile
	1021	#. /home2/t/t51254/palm/current_version/myprofile
	1022	#. /thin/apps/pgi/mpi.sh
	1023	#
	1024	env
	1025	#
	1026	set -x
	1027
	1028	%%END%%
	1029
	1030	elif [[ $remote_host = lcxt5m ]]
	1031	then
	1032
[437]	1033	if [[ $numprocs != 0 ]]
	1034	then
	1035	cat > $job_to_send << %%END%%
[799]	1036	#!/bin/ksh
[437]	1037	#PBS -S /bin/ksh
	1038	#PBS -N $job_name
	1039	#PBS -j oe
	1040	#PBS -l walltime=$timestring
	1041	#PBS -l mppwidth=${numprocs}
[622]	1042	#PBS -l mppnppn=${processes_per_node}
[437]	1043	#PBS -m abe
	1044	#PBS -o $remote_dayfile
	1045
[892]	1046	$init_cmds
[493]	1047	$module_calls
[437]	1048
	1049	%%END%%
	1050
	1051	else
	1052	cat > $job_to_send << %%END%%
	1053	#!/bin/ksh
	1054	#PBS -S /bin/ksh
	1055	#PBS -N $job_name
	1056	#PBS -j oe
	1057	#PBS -l walltime=$timestring
	1058	#PBS -l ncpus=1
	1059	#PBS -l pmem=${memory}mb
	1060	#PBS -m abe
	1061	#PBS -o $remote_dayfile
	1062
[892]	1063	$init_cmds
[493]	1064	$module_calls
[437]	1065
	1066	%%END%%
	1067
	1068	fi
	1069
[1099]	1070	elif [[ $remote_host = lckyuh ]]
	1071	then
	1072	cat > $job_to_send << %%END%%
	1073	#!/bin/bash
	1074	#PJM -L "rscgrp=$queue"
	1075	#PJM -L "node=$nodes"
	1076	#PJM --mpi "proc=$numprocs"
	1077	#PJM -L "elapse=$timestring"
	1078	#PJM -o $remote_dayfile
	1079	#PJM -j
	1080	#PJM -X
	1081	#PJM --no-stging
	1082
	1083	export LANG=en_US.UTF-8
	1084	%%END%%
	1085
[1090]	1086	elif [[ $remote_host = lckyut ]]
	1087	then
	1088	cat > $job_to_send << %%END%%
	1089	#!/bin/bash
[1099]	1090	#PJM -L "rscgrp=$queue"
[1090]	1091	#PJM -L "vnode=$numprocs"
	1092	#PJM -L "vnode-core=1"
	1093	#PJM -L "elapse=$timestring"
	1094	#PJM --mpi proc=$numprocs
	1095	#PJM -o $remote_dayfile
	1096	#PJM -j
[1099]	1097	#PJM -X
	1098	#PJM --no-stging
[1090]	1099
	1100	export LANG=en_US.UTF-8
	1101	%%END%%
	1102
[1]	1103	elif [[ $remote_host = nech ]]
	1104	then
	1105
	1106	if (( nodes > 1 ))
	1107	then
	1108	cat > $job_to_send << %%END%%
	1109	#!/bin/ksh
[622]	1110	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1111	#PBS -l ${qsubmem}=${Memory}gb
	1112	#PBS -b $nodes
	1113	#PBS -o $remote_dayfile
	1114	#PBS -N palm
	1115	#PBS -j o
	1116	#PBS -T mpisx
	1117
	1118	%%END%%
	1119
	1120	elif [[ $numprocs != 0 ]]
	1121	then
	1122	cat > $job_to_send << %%END%%
	1123	#!/bin/ksh
[622]	1124	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1125	#PBS -l ${qsubmem}=${Memory}gb
	1126	#PBS -o $remote_dayfile
	1127	#PBS -N palm
	1128	#PBS -j o
	1129
	1130	%%END%%
	1131
	1132	else
	1133	cat > $job_to_send << %%END%%
	1134	#!/bin/ksh
	1135	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1136	#PBS -o $remote_dayfile
	1137	#PBS -j o
	1138
	1139	%%END%%
	1140
	1141	fi
	1142
	1143	elif [[ $remote_host = lctit ]]
	1144	then
	1145	cat > $job_to_send << %%END%%
[635]	1146	#!/bin/ksh
[892]	1147	$init_cmds
[678]	1148	$module_calls
	1149
[1]	1150	%%END%%
	1151
[1289]	1152	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1153	if [[ $tasks_per_node != $processes_per_node ]]
	1154	then
[1378]	1155	submcom="$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 -et 1 -q $queue "
[678]	1156	else
[1378]	1157	submcom="$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 -et 1 -q $queue "
[678]	1158	fi
[1]	1159
	1160	else
	1161
	1162	cat > $job_to_send << %%END%%
	1163	# @\$-q ${queue}
	1164	# @\$-l${qsubtime} $timestring
	1165	# @\$-l${qsubmem} ${memory}mb
	1166	# @\$-o $remote_dayfile
	1167	# @\$-eo
	1168
	1169	%%END%%
	1170
	1171	fi
	1172
	1173
[1289]	1174	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1175	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1176	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1177	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1178	then
	1179	echo "set +vx" >> $job_to_send
	1180	echo "trap '" >> $job_to_send
	1181	echo "set +vx" >> $job_to_send
[1289]	1182	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = lccrayb \|\| $remote_host = lccrayh \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lcflow \|\| $remote_host = lckiaps \|\| $remote_host = lckyu* ]]
[1]	1183	then
[622]	1184	if [[ $remote_host = ibmh ]]
[1]	1185	then
	1186	return_queue=c1
[693]	1187	elif [[ $remote_host = ibmkisti ]]
	1188	then
	1189	return_queue=class.1-2
[622]	1190	elif [[ $remote_host = ibmku ]]
	1191	then
	1192	return_queue=sdbg2
[1]	1193	elif [[ $remote_host = ibms ]]
	1194	then
	1195	return_queue=p_normal
[1274]	1196	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1197	then
	1198	return_queue=dataq
[1040]	1199	elif [[ $remote_host = lckiaps ]]
	1200	then
	1201	return_queue=express
[1099]	1202	elif [[ $remote_host = lckyuh ]]
	1203	then
	1204	return_queue=cx-single
[1090]	1205	elif [[ $remote_host = lckyut ]]
	1206	then
	1207	return_queue=cx-single
[1]	1208	else
	1209	return_queue=unknown
	1210	fi
	1211
	1212	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1213	then
	1214
[622]	1215	if [[ $remote_host = ibmku ]]
	1216	then
[1289]	1217	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1218	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1219	else
[1289]	1220	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1221	fi
[1289]	1222	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1223	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1224	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1225	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1226	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1227	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1228	if [[ $host != "ibmh" ]]
	1229	then
[1289]	1230	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1231	fi
[1289]	1232	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1233	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1234
[1289]	1235	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1236	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1237
[1289]	1238	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
	1239	echo "echo \"batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$identifier" >> $job_to_send
[622]	1240	if [[ $remote_host = ibmku ]]
	1241	then
[1289]	1242	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1243	fi
[1289]	1244	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1245
	1246	elif [[ $remote_host = nech ]]
	1247	then
	1248	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1249	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1250	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1251	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1252	echo "#PBS -j o" >> $job_to_send
	1253	echo " " >> $job_to_send
	1254	echo "set -x" >> $job_to_send
	1255	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1094]	1256	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1257	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1258	echo "%%END%%" >> $job_to_send
	1259
[1099]	1260	elif [[ $remote_host = lckyuh ]]
	1261	then
[1289]	1262	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1263	echo "#!/bin/bash" >> $job_to_send
	1264	echo "#PJM -L \"node=1\"" >> $job_to_send
	1265	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1266	echo "#PJM --no-stging" >> $job_to_send
	1267	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1268	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1269	echo "#PJM -j" >> $job_to_send
	1270	echo " " >> $job_to_send
	1271	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1272	echo "set -x" >> $job_to_send
	1273	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1274	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1275	echo "%%END%%" >> $job_to_send
	1276
[1090]	1277	elif [[ $remote_host = lckyut ]]
	1278	then
[1289]	1279	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1280	echo "#!/bin/bash" >> $job_to_send
	1281	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1282	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1283	echo "#PJM --no-stging" >> $job_to_send
	1284	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1285	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1286	echo "#PJM -j" >> $job_to_send
	1287	echo " " >> $job_to_send
	1288	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1289	echo "set -x" >> $job_to_send
[1094]	1290	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1291	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1292	echo "%%END%%" >> $job_to_send
	1293
[1274]	1294	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1295	then
[1289]	1296	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1297	echo "#!/bin/bash" >> $job_to_send
	1298	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1299	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1300	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1301	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1302	echo "#PBS -j oe" >> $job_to_send
	1303	echo " " >> $job_to_send
	1304	echo "set -x" >> $job_to_send
	1305	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1306	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1307	echo "%%END%%" >> $job_to_send
	1308
[1099]	1309	elif [[ $remote_host = lcflow ]]
	1310	then
[1289]	1311	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1312	echo "#!/bin/bash" >> $job_to_send
	1313	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1314	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1315	echo "SGEPREFIX -cwd" >> $job_to_send
	1316	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1317	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
	1318	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
[1099]	1319	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1320	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1321	echo " " >> $job_to_send
	1322	echo "set -x" >> $job_to_send
	1323	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1324	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1325	echo "" >> $job_to_send
	1326	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1327	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1328	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1329	echo "%%END%%" >> $job_to_send
[1289]	1330	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1331	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1099]	1332
[1]	1333	else
	1334
[1289]	1335	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1336	echo "# @\\\$-q $return_queue" >> $job_to_send
	1337	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1338	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1339	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1340	then
	1341	echo "# @\$-l mpp_p=0" >> $job_to_send
	1342	fi
	1343	echo '# @\$-lF 10mb' >> $job_to_send
	1344	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1345	echo '# @\\\$-eo' >> $job_to_send
	1346	echo " " >> $job_to_send
	1347	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1348	then
	1349	echo "set +vx" >> $job_to_send
	1350	echo ". .profile" >> $job_to_send
	1351	fi
	1352	echo "set -x" >> $job_to_send
[1094]	1353	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
[1]	1354	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1355	echo "%%END%%" >> $job_to_send
	1356	fi
	1357
	1358	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1359	then
[1289]	1360	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1361	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1362	then
[1289]	1363	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1364	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1365	then
[1289]	1366	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1367	elif [[ $remote_host = t3es ]]
	1368	then
[1289]	1369	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1370	elif [[ $remote_host = lckiaps ]]
	1371	then
[1289]	1372	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1373	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
	1374	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1375	elif [[ $remote_host = lckyu* ]]
[1090]	1376	then
[1289]	1377	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
	1378	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1379	elif [[ $remote_host = lcflow ]]
	1380	then
[1289]	1381	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1382	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1383	else
[1289]	1384	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1385	fi
[1043]	1386	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1387	then
[1289]	1388	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1389	fi
[1]	1390	if [[ $remote_host = nech ]]
	1391	then
	1392	echo "cd -" >> $job_to_send
	1393	fi
	1394	else
	1395	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1396	# ??? funktioniert das ÃŒberhaupt noch ???
[1]	1397	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1398	fi
	1399	echo "set -x" >> $job_to_send
	1400	echo " ' exit" >> $job_to_send
	1401	echo "set -x" >> $job_to_send
	1402	fi
	1403
	1404
	1405
[1289]	1406	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1407	cat $file_to_send >> $job_to_send
[69]	1408
[1]	1409	if [[ $remote_host = ibm ]]
	1410	then
	1411	echo " " >> $job_to_send
	1412	echo "exit" >> $job_to_send
	1413	fi
[635]	1414
[1289]	1415	# REMOVE JOB-FILE
[1099]	1416	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1417	then
	1418	echo " " >> $job_to_send
	1419	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1420	fi
[1]	1421
	1422
[1289]	1423	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1424	if [[ $no_submit = false ]]
	1425	then
	1426	if [[ $remote_host != $local_host ]]
	1427	then
	1428	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1429	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1430	then
[82]	1431	job_catalog_save=$job_catalog
	1432	job_catalog=job_queue
	1433	elif [[ $remote_host = nech ]]
	1434	then
	1435	job_catalog_save=$job_catalog
	1436	job_catalog=/hpf/b/${remote_user}/job_queue
	1437	fi
[1096]	1438	if [[ $remote_host = nech ]]
[82]	1439	then
[1289]	1440	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1441	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1442	else
[1094]	1443	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1444	fi
[82]	1445	if [[ $? = 1 ]]
	1446	then
	1447	locat=scp; exit
	1448	fi
	1449	if [[ $remote_host = ibms ]]
	1450	then
	1451	job_catalog=$job_catalog_save
	1452	fi
[1]	1453	[[ $verify = true ]] && printf "\n >>> finished\n"
	1454	else
	1455	eval job_catalog=$job_catalog
	1456	cp $job_to_send ${job_catalog}/$job_on_remhost
	1457	fi
	1458
	1459
	1460
[1289]	1461	# START NQS- / LOADLEVELER-JOB
[1]	1462	if [[ $remote_host != $local_host ]]
	1463	then
	1464	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1465
[1289]	1466	if [[ $remote_host = ibmku ]]
[1]	1467	then
[1094]	1468	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1469	elif [[ $remote_host = lcflow ]]
	1470	then
[1202]	1471	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1472	else
[1094]	1473	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1474	fi
[1]	1475
	1476	[[ $verify = true ]] && printf " >>> o.k.\n"
	1477	else
	1478	cd $job_catalog
[1289]	1479	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1480	then
	1481	eval $submcom $job_on_remhost
[1289]	1482	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1483	then
[635]	1484	chmod u+x $job_on_remhost
[108]	1485	eval $submcom $job_on_remhost
[1]	1486	elif [[ $local_host = nech ]]
	1487	then
	1488	if [[ $queue = default ]]
	1489	then
[799]	1490	eval $submcom $job_on_remhost
[1]	1491	else
[799]	1492	eval $submcom -q $queue $job_on_remhost
[1]	1493	fi
	1494	else
	1495	qsub $job_on_remhost
	1496	fi
[622]	1497
[1289]	1498	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1499	# AT THE END OF THE JOB
[1099]	1500	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1501	then
	1502	rm $job_on_remhost
	1503	fi
[1]	1504	cd - > /dev/null
	1505	fi
	1506	fi
	1507
	1508
	1509
[1289]	1510	# FINAL ACTIONS
[1]	1511	if [[ $no_submit = false ]]
	1512	then
[1099]	1513	rm -f $job_to_send
[1]	1514	fi
[1266]	1515	[[ $verify = true ]] && printf "\n\n *** SUBJOB finished \n\n"

Note: See TracBrowser for help on using the repository browser.

Download in other formats:

| Impressum | ©Leibniz Universität Hannover |