Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1446

Last change on this file since 1446 was 1443, checked in by raasch, 10 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 50.4 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	#
[1443]	26	#
[1046]	27	# Former revisions:
	28	# -----------------
[169]	29	# $Id: subjob 1443 2014-07-28 07:11:00Z maronga $
[1200]	30	#
[1443]	31	# 1442 2014-07-28 07:09:10Z raasch
	32	# HLRN-III (lccrayb/lccrayh) queues adjusted
	33	#
[1379]	34	# 1378 2014-04-28 06:04:58Z raasch
	35	# -et option added for lctit
	36	#
[1351]	37	# 1350 2014-04-04 13:01:30Z maronga
	38	# location of qsub updated for lcxe6
	39	#
[1290]	40	# 1289 2014-03-04 07:12:34Z raasch
	41	# German comments translated to English
	42	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
	43	#
[1280]	44	# 1279 2014-01-28 12:10:14Z raasch
	45	# node calculation modified due to changes in mrun (tasks_per_node must not be
	46	# an integral divisor of numprocs any more)
	47	#
[1275]	48	# 1274 2014-01-09 13:14:54Z heinze
	49	# adjustments for lccrayh
	50	#
[1267]	51	# 1266 2013-12-11 12:07:34Z heinze
	52	# further adjustments for lccrayb (use msub instead of qsub)
	53	#
[1265]	54	# 1264 2013-12-09 12:46:09Z fricke
	55	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	56	#
[1263]	57	# 1262 2013-12-09 10:57:20Z fricke
	58	# further adjustments for lccrayb
	59	#
[1261]	60	# 1260 2013-12-04 12:48:04Z raasch
	61	# jaboticaba admitted
	62	#
[1256]	63	# 1255 2013-11-07 14:43:35Z raasch
	64	# further adjustments for lccrayb
	65	#
[1225]	66	# 1224 2013-09-16 07:27:23Z raasch
	67	# first adjustments for lccrayb
	68	#
[1203]	69	# 1202 2013-07-10 16:22:07Z witha
	70	# adjustments for Forwind cluster (lcflow)
	71	#
[1200]	72	# 1199 2013-07-05 14:52:22Z raasch
	73	# adjustments for CSC Helsinki (lccrayf)
	74	#
[1185]	75	# use of cluster/express queue enabled (ibmh)
	76	# vinessa added (imuk)
[1047]	77	#
[1104]	78	# 1103 2013-02-20 02:15:53Z raasch
	79	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	80	# instead typeset -L/R),
	81	# further adjustments for lckyuh
	82	#
[1100]	83	# 2013-02-10 01:47:43Z raasch
	84	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	85	# and for Forwind cluster (lcflow)
	86	#
[1096]	87	# 1094 2013-02-03 01:52:12Z raasch
	88	# new option -P for explicit setting of ssh/scp port,
	89	# decalpha parts (yonsei) removed
	90	#
[1091]	91	# 2013-02-02 07:06:13Z raasch
[1099]	92	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	93	# old changelog messages removed
	94	#
[1047]	95	# 1046 2012-11-09 14:38:45Z maronga
	96	# code put under GPL (PALM 3.9)
	97	#
[1090]	98	# 08/07/94 - Siggi - first version finished
	99	# 29/06/94 - Siggi - script development started
	100	#--------------------------------------------------------------------------------#
	101	# subjob - script for automatic generation and submission of batch-job files
	102	# for various batch queuing systems
	103	#--------------------------------------------------------------------------------#
[1]	104
	105
[1289]	106	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	107	delete_dayfile=false
[799]	108	email_notification=none
[122]	109	group_number=none
[1]	110	locat=normal
	111	no_default_queue=none
	112	no_submit=false
	113	job_catalog="~/job_queue"
	114	job_name=none
	115	local_user=$LOGNAME
	116	node_usage=shared
[475]	117	numprocs=0
[1]	118	punkte="..........................................................."
	119	submcom=qsub
	120	queue=default
	121	remote_host=none
	122	remote_user=""
	123	verify=true
	124
	125	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	126	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	127
	128
	129
[1289]	130	# ERROR HANDLING
	131	# IN CASE OF EXIT:
[1]	132	trap 'if [[ $locat != normal ]]
	133	then
	134	case $locat in
	135	(option) printf "\n --> available optios can be displayed"
	136	printf " by typing:"
	137	printf "\n \"subjob ?\" \n";;
	138	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	139	(*) printf "\n +++ unknown error"
	140	printf "\n please inform S. Raasch!\n"
	141	esac
	142	[[ -f $job_to_send ]] && rm $job_to_send
	143	printf "\n\n+++ SUBJOB killed \n\n"
	144	fi' exit
	145
	146
[1289]	147	# IN CASE OF TERMINAL-BREAK:
[1]	148	trap '[[ -f $job_to_send ]] && rm $job_to_send
	149	printf "\n\n+++ SUBJOB killed \n\n"
	150	exit
	151	' 2
	152
	153
	154
	155
[1289]	156	# DETERMINE NAME OF LOCAL HOST
[1]	157	local_host=$(hostname)
	158
	159
	160
[1289]	161	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	162	# IS ADMITTED AT ALL)
	163	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	164	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	165	case $local_host in
[1204]	166	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
[1]	167	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[1235]	168	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
[102]	169	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	170	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1103]	171	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
[749]	172	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[1255]	173	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
[1274]	174	(hlogin\|hxc) local_addres=130.75.4.1; local_host=lccrayh;;
[12]	175	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[1174]	176	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
[372]	177	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	178	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[1197]	179	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
[1]	180	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	181	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[1099]	182	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
	183	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
[367]	184	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	185	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	186	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	187	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[5]	188	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	189	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1099]	190	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
[892]	191	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[440]	192	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	193	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	194	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[1260]	195	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
[407]	196	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	197	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
[1040]	198	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
[5]	199	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1099]	200	(meller) local_addres=134.106.74.155; local_host=lcfor;;
[437]	201	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	202	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	203	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	204	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	205	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	206	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[1099]	207	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
[789]	208	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	209	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	210	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1204]	211	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
[1197]	212	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
[799]	213	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1090]	214	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
[635]	215	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1021]	216	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
[1184]	217	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
[1]	218	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	219	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	220	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	221	printf "\n please contact the PALM group at IMUK";
[1]	222	locat=parameter; exit;;
	223	esac
	224
	225
	226
[1289]	227	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	228	remote_host=$local_host
	229
	230
	231
	232
[1289]	233	# READ THE SHELLSCRIPT-OPTIONS
[1094]	234	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	235	do
	236	case $option in
	237	(c) job_catalog=$OPTARG;;
	238	(d) delete_dayfile=true;;
	239	(D) no_submit=true;;
[352]	240	(e) email_notification=$OPTARG;;
[125]	241	(g) group_number=$OPTARG;;
[1]	242	(h) remote_host=$OPTARG;;
	243	(m) memory=$OPTARG;;
	244	(n) job_name=$OPTARG;;
	245	(N) node_usage=$OPTARG;;
	246	(O) threads_per_task=$OPTARG;;
[1094]	247	(P) scp_port=$OPTARG;;
[1]	248	(q) no_default_queue=$OPTARG;;
	249	(t) cputime=$OPTARG;;
	250	(T) tasks_per_node=$OPTARG;;
	251	(u) remote_user=$OPTARG;;
	252	(v) verify=false;;
	253	(X) numprocs=$OPTARG;;
	254	(\?) printf "\n +++ Option $OPTARG unknown \n";
	255	locat=option; exit;;
	256	esac
	257	done
	258
	259
[1289]	260	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	261	(( to_shift = $OPTIND - 1 ))
	262	shift $to_shift; file_to_send=$1
[1]	263
	264
[1289]	265	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	266	if [ "$1" = "?" ]
	267	then
	268	(printf "\n *** subjob can be called as follows:\n"
	269	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	270	printf "\n Description of available options:\n"
	271	printf "\n Option Description Default-Value"
	272	printf "\n -c job-input- and output-catalog ~/job_queue"
	273	printf "\n -d no job-protocol will be created ---"
	274	printf "\n -D only the job-file will be created ---"
	275	printf "\n -h execution host, available hosts: $remote_host"
[1289]	276	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	277	printf "\n lckiaps, lctit, nech"
[1]	278	printf "\n -m memory demand per process in MByte ---"
	279	printf "\n -n jobname <jobdatei>"
	280	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	281	printf "\n -P ssh/scp port default port"
[1]	282	printf "\n -q job-queue to be used default"
	283	printf "\n -t allowed cpu-time in seconds ---"
	284	printf "\n -T tasks per node (on parallel hosts) ---"
	285	printf "\n -u username on execution host from .netrc"
	286	printf "\n -v no prompt for confirmation ---"
	287	printf "\n -X # of processors (on parallel hosts) 1"
	288	printf "\n "
	289	printf "\n The only possible positional parameter is <jobfile>:"
	290	printf "\n The complete NQS-job must be provided here."
	291	printf "\n <jobfile>=? creates this outline\n\n") \| more
	292	exit
	293	fi
	294
	295
	296
[1289]	297	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	298	if [[ "$file_to_send" = "" ]]
	299	then
	300	printf "\n +++ job-file missing"
	301	locat=parameter; exit
	302	else
	303	if [[ -f $file_to_send ]]
	304	then
	305	true
	306	else
	307	printf "\n +++ job-file: "
	308	printf "\n $file_to_send"
	309	printf "\n does not exist"
	310	locat=parameter; exit
	311	fi
	312	fi
	313
	314
	315
[1289]	316	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	317	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	318	if [[ $job_name = none ]]
	319	then
	320	job_name=$file_to_send
	321	fi
	322	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	323	then
	324	printf "\n +++ job-file name: "
	325	printf "\n $job_name"
	326	printf "\n must not contain \"/\"-characters"
	327	locat=parameter; exit
	328	fi
	329
	330
	331
	332
[1289]	333	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	334	# OR IF NO HOST HAS BEEN GIVEN
[1]	335	if [[ $remote_host = none ]]
	336	then
	337	printf "\n +++ host missing"
	338	locat=option; exit
	339	else
	340	case $remote_host in
[251]	341	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1184]	342	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	343	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	344	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	345	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1262]	346	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1274]	347	(lccrayh) queue=mpp1testq; remote_addres=130.75.4.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1197]	348	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
[1099]	349	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
[440]	350	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[367]	351	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[1040]	352	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
[717]	353	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[1099]	354	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
[1090]	355	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
[1021]	356	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
[635]	357	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[1350]	358	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
[437]	359	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	360	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	361	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	362	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	363	locat=parameter; exit;;
[1]	364	esac
	365	fi
	366
	367
[1289]	368	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	369	if [[ $no_default_queue != none ]]
	370	then
	371	error=false
	372	ndq=$no_default_queue
	373	case $remote_host in
	374	(ibm) case $ndq in
	375	(p690_express\|p690_standard\|p690_long) error=false;;
	376	(*) error=true;;
	377	esac;;
	378	(ibmh) case $ndq in
[1184]	379	(cluster\|express) error=false;;
[1]	380	(*) error=true;;
	381	esac;;
[693]	382	(ibmkisti) case $ndq in
	383	(class.32plus\|class.1-2\|class.2-32) error=false;;
	384	(*) error=true;;
	385	esac;;
[622]	386	(ibmku) case $ndq in
	387	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	388	(*) error=true;;
	389	esac;;
[1]	390	(ibms) case $ndq in
	391	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	392	(*) error=true;;
	393	esac;;
[1224]	394	(lccrayb) case $ndq in
[1442]	395	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1224]	396	(*) error=true;;
	397	esac;;
[1274]	398	(lccrayh) case $ndq in
[1442]	399	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1274]	400	(*) error=true;;
	401	esac;;
[1197]	402	(lccrayf) case $ndq in
	403	(usup\|test*\|small\|large) error=false;;
	404	(*) error=true;;
	405	esac;;
[1099]	406	(lcflow) case $ndq in
	407	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
	408	(*) error=true;;
	409	esac;;
[1040]	410	(lckiaps) case $ndq in
	411	(express\|normal) error=false;;
	412	(*) error=true;;
	413	esac;;
[440]	414	(lckyoto) case $ndq in
	415	(eh\|ph) error=false;;
	416	(*) error=true;;
	417	esac;;
[1099]	418	(lckyuh) case $ndq in
	419	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	420	(*) error=true;;
	421	esac;;
[1090]	422	(lckyut) case $ndq in
	423	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	424	(*) error=true;;
	425	esac;;
[1]	426	(lctit) case $ndq in
[635]	427	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	428	(*) error=true;;
	429	esac;;
	430	(t3eb) case $ndq in
	431	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	432	(*) error=true;;
	433	esac;;
	434	(t3eh) case $ndq in
	435	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	436	(*) error=true;;
	437	esac;;
	438	(t3ej2\|t3ej5) case $ndq in
	439	(low\|normal\|high) error=false;;
	440	(*) error=true;;
	441	esac;;
	442	(t3es) case $ndq in
	443	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	444	(*) error=true;;
	445	esac;;
	446	esac
	447	if [[ $error = true ]]
	448	then
	449	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	450	locat=parameter; exit
	451	else
	452	queue=$no_default_queue
	453	fi
	454	fi
	455
	456
	457
[1289]	458	# CHECK THE CPU-TIME
	459	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	460	done=false
	461	while [[ $done = false ]]
	462	do
	463	if (( $cputime <= 0 ))
	464	then
	465	printf "\n +++ wrong cpu-time or cpu-time missing"
	466	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	467	printf "\n >>> "
	468	read cputime 1>/dev/null 2>&1
	469	else
	470	done=true
	471	fi
	472	done
	473	if [[ $remote_host = nech ]]
	474	then
	475	if (( tasks_per_node != 0 ))
	476	then
	477	(( cputime = cputime * tasks_per_node ))
	478	elif [[ $numprocs != 0 ]]
	479	then
	480	(( cputime = cputime * numprocs ))
	481	fi
	482	fi
	483	(( stunden = cputime / 3600 ))
	484	(( resttime = cputime - stunden * 3600 ))
	485	(( minuten = resttime / 60 ))
	486	(( sekunden = resttime - minuten * 60 ))
	487	timestring=${stunden}:${minuten}:${sekunden}
	488
	489
	490
[1289]	491	# CHECK THE MEMORY DEMAND
[1]	492	done=false
	493	while [[ $done = false ]]
	494	do
	495	if (( memory <= 0 ))
	496	then
	497	printf "\n +++ wrong memory demand or memory demand missing"
	498	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	499	printf "\n >>> "
	500	read memory 1>/dev/null 2>&1
	501	else
	502	done=true
	503	fi
	504	done
	505
[1289]	506	if [[ $remote_host = nech ]]
[1]	507	then
	508	if (( tasks_per_node != 0 ))
	509	then
	510	(( Memory = memory * tasks_per_node / 1000 ))
	511	elif [[ $numprocs != 0 ]]
	512	then
	513	(( Memory = memory * numprocs / 1000 ))
	514	else
	515	(( Memory = memory / 1000 ))
	516	fi
[635]	517	elif [[ $remote_host = lctit ]]
	518	then
	519	(( Memory = memory * tasks_per_node / 1000 ))
[1]	520	fi
	521
	522
[1289]	523	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	524	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	525	then
	526	(( memory = memory * threads_per_task ))
	527	fi
	528
	529
[1289]	530	# CALCULATE NUMBER OF REQUIRED NODES
[1]	531	if (( tasks_per_node != 0 ))
	532	then
[1279]	533	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	534	fi
	535
[1094]	536
[1289]	537	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	538	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	539
[1094]	540
[1289]	541	# CALCULATE NUMBER OF MPI TASKS
[696]	542	(( mpi_tasks = numprocs / threads_per_task ))
[1]	543
[696]	544
[1289]	545	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	546	if [[ "$scp_port" != "" ]]
	547	then
	548	PORTOPT="-P $scp_port"
	549	SSH_PORTOPT="-p $scp_port"
	550	fi
	551
	552
[1289]	553	# HEADER-OUTPUT
[1]	554	if [[ $verify = true ]]
	555	then
	556	printf "\n\n"
	557	printf "#--------------------------------------------------------------# \n"
	558	spalte1=SUBJOB;spalte2=$(date)
[1103]	559	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	560	printf "\| \| \n"
	561	printf "\| values of parameters/options: \| \n"
[1103]	562	spalte1=$(echo local_host$punkte \| cut -c-20)
	563	spalte2=$punkte$local_host
	564	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	565	spalte1=$(echo remote_host$punkte \| cut -c-20)
	566	spalte2=$punkte$remote_host
	567	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	568	spalte1=$(echo queue$punkte \| cut -c-20)
	569	spalte2=$punkte$queue
	570	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	571	spalte1=$(echo memory$punkte \| cut -c-20)
	572	spalte2="$punkte$memory mb"
	573	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	574	spalte1=$(echo cputime$punkte \| cut -c-20)
	575	spalte2="$punkte$cputime sec"
	576	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	577	spalte1=$(echo job_name$punkte \| cut -c-20)
	578	spalte2="$punkte$job_name"
	579	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	580	printf "#--------------------------------------------------------------# \n\n"
	581
	582
[1289]	583	# QUERY CHECK
[1]	584	antwort="dummy"
	585	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	586	do
	587	read antwort?" >>> continue (y/n) ? "
	588	done
	589	if [[ $antwort = n \|\| $antwort = N ]]
	590	then
	591	locat=verify; exit
	592	fi
	593	printf "\n"
	594	fi
	595
	596
	597
[1289]	598	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	599	identifier=$RANDOM
	600	job_on_remhost=${job_name}_${identifier}_$local_host
	601	job_to_send=job_to_send_$identifier
[1]	602	if [[ $delete_dayfile = false ]]
	603	then
[1289]	604	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	605	local_dayfile=${remote_host}_${job_name}
	606	else
	607	remote_dayfile=/dev/null
	608	fi
	609
	610
	611
[1289]	612	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	613	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	614	then
	615
[1289]	616	# GENERAL LOADLEVELER SETTINGS
[622]	617	execute_in_shell="#!/bin/ksh"
	618	use_shell="# @ shell = /bin/ksh"
	619	consumable_memory="ConsumableMemory($memory mb)"
	620	class="# @ class = $queue"
	621	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	622	network_to_use="# @ network.mpi = sn_all,shared,us"
	623	data_limit="# @ data_limit = 1.76gb"
	624	image_size="# @ image_size = 50"
[693]	625	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	626
[693]	627	if [[ $email_notification = none ]]
	628	then
	629	notify_user=""
	630	else
	631	notify_user="# @ notify_user = $email_notification"
	632	if [[ $delete_dayfile = true ]]
	633	then
	634	notification='# @ notification = never'
	635	fi
	636	fi
[622]	637
[312]	638	if [[ $remote_host = ibmh ]]
[1]	639	then
[312]	640	data_limit=""
	641	network_to_use=""
[1184]	642	class="# @ class = $queue"
[312]	643	environment=""
[814]	644	rset="# @ rset = RSET_MCM_AFFINITY"
	645	task_affinity="# @ task_affinity = core(1)"
[693]	646	elif [[ $remote_host = ibmkisti ]]
	647	then
	648	network_to_use="# @ network.MPI = sn_all,shared,US"
	649	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	650	if [[ $threads_per_task = 1 ]]
	651	then
	652	rset="# @ rset = RSET_MCM_AFFINITY"
	653	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	654	fi
[693]	655	environment=""
	656	use_shell=""
	657	data_limit=""
	658	image_size=""
[622]	659	elif [[ $remote_host = ibmku ]]
	660	then
	661	execute_in_shell="#!/usr/bin/ksh"
	662	use_shell="# @ shell = /usr/bin/ksh"
	663	consumable_memory=""
	664	environment=""
	665	network_to_use="# @ network.mpi = sn_all,shared,us"
	666	data_limit=""
	667	image_size=""
	668	elif [[ $remote_host = ibms ]]
	669	then
	670	network_to_use="# @ network.mpi = csss,shared,us"
[1]	671	fi
	672
	673	cat > $job_to_send << %%END%%
[622]	674	$execute_in_shell
	675	$use_shell
[1]	676
	677	# @ job_type = parallel
[693]	678	# @ job_name = $job_name
[1]	679	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	680	# @ output = $remote_dayfile
	681	# @ error = $remote_dayfile
[693]	682	$wall_clock_limit
[622]	683	$image_size
[312]	684	$class
	685	$environment
[1]	686	$network_to_use
	687	$data_limit
[693]	688	$rset
	689	$mcm_affinity_options
[814]	690	$task_affinity
[1]	691	$notification
[693]	692	$notify_user
[1]	693
	694	%%END%%
	695
	696	if (( nodes > 0 ))
	697	then
	698
[693]	699	if [[ $remote_host != ibmkisti ]]
	700	then
	701
	702	cat >> $job_to_send << %%END%%
[1]	703	# @ node = $nodes
[622]	704	# @ tasks_per_node = $processes_per_node
[1]	705	# @ node_usage = $node_usage
	706	# @ queue
	707
	708	%%END%%
	709
[693]	710	else
	711
	712	cat >> $job_to_send << %%END%%
[696]	713	# @ total_tasks = $mpi_tasks
[693]	714	# @ blocking = unlimited
	715	# @ queue
	716
	717	%%END%%
	718
	719	fi
	720
[1]	721	else
	722
[1289]	723	cat >> $job_to_send << %%END%%
[1]	724	# @ blocking = unlimited
	725	# @ total_tasks = $numprocs
	726	# @ node_usage = $node_usage
	727	# @ queue
	728
	729	%%END%%
	730
	731	fi
	732
[1289]	733	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	734	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	735	then
	736	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	737	fi
	738
[1]	739	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	740	then
	741
	742	cat > $job_to_send << %%END%%
	743	#!/bin/ksh
	744
	745	# @ job_type = serial
	746	# @ node_usage = $node_usage
	747	# @ job_name = palm
	748	# @ wall_clock_limit = ${timestring},$timestring
	749	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	750	# @ output = $remote_dayfile
	751	# @ error = $remote_dayfile
[312]	752	$class
[1]	753	$notification
	754
	755	# @ queue
	756
	757	%%END%%
	758
[1274]	759	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	760	then
	761
	762	if [[ $numprocs != 0 ]]
	763	then
	764	cat > $job_to_send << %%END%%
[1255]	765	#!/bin/bash
[1224]	766	#PBS -N $job_name
	767	#PBS -l walltime=$timestring
[1264]	768	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	769	#PBS -o $remote_dayfile
	770	#PBS -j oe
	771	#PBS -q $queue
	772
	773	$module_calls
	774
	775	%%END%%
	776
	777	else
	778
	779	continue
	780
	781	fi
	782
[1197]	783	elif [[ $remote_host = lccrayf ]]
	784	then
	785
	786	if [[ $numprocs != 0 ]]
	787	then
	788	cat > $job_to_send << %%END%%
	789	#!/bin/bash -l
	790	#SBATCH -J $job_name
	791	#SBATCH -t $timestring
	792	#SBATCH -N $nodes
	793	#SBATCH --ntasks-per-node=$processes_per_node
	794	#SBATCH -p $queue
	795	#SBATCH -o $remote_dayfile
	796	#SBATCH -e $remote_dayfile
	797
	798	$init_cmds
	799	$module_calls
	800
	801	%%END%%
	802
	803	else
	804	cat > $job_to_send << %%END%%
	805	#!/bin/bash -l
	806	#SBATCH -J $job_name
	807	#SBATCH -t $timestring
	808	#SBATCH -l ncpus=1
	809	#SBATCH -l pmem=${memory}mb
	810	#SBATCH -m abe
	811	#SBATCH -o $remote_dayfile
	812	#SBATCH -e $remote_dayfile
	813
	814	$init_cmds
	815	$module_calls
	816
	817	%%END%%
	818
	819	fi
	820
[1099]	821	elif [[ $remote_host = lcflow ]]
	822	then
[1202]	823	if [ $memory -gt 1800 ]; then
[1099]	824	use_himem=""
	825	else
	826	use_himem="#"
	827	fi
	828	if [[ $numprocs != 0 ]]
	829	then
[1202]	830	pe_set="#$ -pe impi41 $numprocs"
[1099]	831	else
[1202]	832	pe_set="#$ -pe impi41 1"
[1099]	833	fi
	834	if [[ $queue = default ]]
	835	then
	836	queue_set=""
	837	else
	838	queue_set="#$ -q $queue"
	839	fi
	840	[[ "$disc_space" = "" ]] && disc_space=50
	841
	842	cat > $job_to_send << %%END%%
	843	#!/bin/bash
	844	#$ -S /bin/bash
	845	#$ -N $job_name
	846	#$ -cwd
	847	#$ -l h_rt=$timestring
	848	#$ -l h_vmem=${memory}M
	849	#$ -o $remote_dayfile
	850	#$ -j y
	851	$pe_set
	852	#$ -R y
	853	${use_himem}#$ -l highmem=true
	854	#$ -l h_fsize=${disc_space}G
	855	$queue_set
	856
	857	%%END%%
	858
[1021]	859	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	860	then
	861
	862	if [[ $numprocs != 0 ]]
	863	then
	864	cat > $job_to_send << %%END%%
	865	#!/bin/ksh
	866	#PBS -N $job_name
	867	#PBS -l walltime=$timestring
	868	#PBS -l ncpus=$numprocs
	869	#PBS -l pmem=${memory}mb
	870	#PBS -o $remote_dayfile
[1021]	871	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	872	#PBS -j oe
	873
	874	mpd &
	875
	876	%%END%%
	877
	878	else
	879	cat > $job_to_send << %%END%%
	880	#!/bin/ksh
	881	#PBS -N $job_name
	882	#PBS -l walltime=$timestring
	883	#PBS -l ncpus=1
	884	#PBS -l pmem=${memory}mb
	885	#PBS -o $remote_dayfile
	886	#PBS -j oe
	887
	888	%%END%%
	889
	890	fi
	891
[1040]	892	elif [[ $remote_host = lckiaps ]]
	893	then
	894
	895	if [[ $numprocs != 0 ]]
	896	then
	897	cat > $job_to_send << %%END%%
	898	#!/bin/ksh
	899	#PBS -N $job_name
	900	#PBS -l walltime=$timestring
	901	#PBS -l select=1:ncpus=$numprocs
	902	#PBS -l pmem=${memory}mb
	903	#PBS -q $queue
	904	#PBS -o $remote_dayfile
	905	#PBS -j oe
	906	#PBS -V
	907
	908	%%END%%
	909
	910	else
	911	cat > $job_to_send << %%END%%
	912	#!/bin/ksh
	913	#PBS -N $job_name
	914	#PBS -l walltime=$timestring
	915	#PBS -l ncpus=1
	916	#PBS -l pmem=${memory}mb
	917	#PBS -o $remote_dayfile
	918	#PBS -j oe
	919
	920	%%END%%
	921
	922	fi
	923
[693]	924	elif [[ $remote_host = lcyon ]]
	925	then
	926
	927	if [[ $numprocs != 0 ]]
	928	then
	929	cat > $job_to_send << %%END%%
	930	#!/bin/ksh
	931	#PBS -N $job_name
	932	#PBS -l walltime=$timestring
	933	#PBS -l ncpus=$numprocs
	934	#PBS -l pmem=${memory}mb
	935	#PBS -o $remote_dayfile
	936	#PBS -j oe
	937
	938	%%END%%
	939
	940	else
	941	cat > $job_to_send << %%END%%
	942	#!/bin/ksh
	943	#PBS -N $job_name
	944	#PBS -l walltime=$timestring
	945	#PBS -l ncpus=1
	946	#PBS -l pmem=${memory}mb
	947	#PBS -o $remote_dayfile
	948	#PBS -j oe
	949
	950	%%END%%
	951
	952	fi
	953
[892]	954	elif [[ $remote_host = lcxe6 ]]
[164]	955	then
	956
	957	if [[ $numprocs != 0 ]]
	958	then
	959	cat > $job_to_send << %%END%%
[799]	960	#!/bin/ksh
[164]	961	#PBS -S /bin/ksh
	962	#PBS -N $job_name
[552]	963	#PBS -A $project_account
[206]	964	#PBS -j oe
[164]	965	#PBS -l walltime=$timestring
	966	#PBS -l mppwidth=${numprocs}
[622]	967	#PBS -l mppnppn=${processes_per_node}
[164]	968	#PBS -m abe
	969	#PBS -o $remote_dayfile
[492]	970	$email_directive
[164]	971
[892]	972	$init_cmds
[493]	973	$module_calls
[343]	974
[164]	975	%%END%%
	976
[108]	977	else
	978	cat > $job_to_send << %%END%%
	979	#!/bin/ksh
[168]	980	#PBS -S /bin/ksh
[108]	981	#PBS -N $job_name
[552]	982	#PBS -A $project_account
[206]	983	#PBS -j oe
[108]	984	#PBS -l walltime=$timestring
	985	#PBS -l ncpus=1
	986	#PBS -l pmem=${memory}mb
	987	#PBS -m abe
[492]	988	$email_directive
[108]	989	#PBS -o $remote_dayfile
	990
[892]	991	$init_cmds
[493]	992	$module_calls
[343]	993
[108]	994	%%END%%
	995
	996	fi
	997
[440]	998	elif [[ $remote_host = lckyoto ]]
[437]	999	then
	1000
[440]	1001	cat > $job_to_send << %%END%%
[799]	1002	#!/bin/ksh
[440]	1003	# @\$-o $remote_dayfile
	1004	# @\$-eo -oi
	1005	# @\$-lP 16
[799]	1006	# @\$-lp 1
[440]	1007	# @\$-lm 28gb -llm unlimited -ls unlimited
	1008	# @\$-q $queue
	1009	# @\$-Pvn abs_pack
	1010	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1011	#. /thin/local/etc/setprofile/intel-11.0.sh
	1012	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1013	. ~/.myprofile
	1014	#. /home2/t/t51254/palm/current_version/myprofile
	1015	#. /thin/apps/pgi/mpi.sh
	1016	#
	1017	env
	1018	#
	1019	set -x
	1020
	1021	%%END%%
	1022
	1023	elif [[ $remote_host = lcxt5m ]]
	1024	then
	1025
[437]	1026	if [[ $numprocs != 0 ]]
	1027	then
	1028	cat > $job_to_send << %%END%%
[799]	1029	#!/bin/ksh
[437]	1030	#PBS -S /bin/ksh
	1031	#PBS -N $job_name
	1032	#PBS -j oe
	1033	#PBS -l walltime=$timestring
	1034	#PBS -l mppwidth=${numprocs}
[622]	1035	#PBS -l mppnppn=${processes_per_node}
[437]	1036	#PBS -m abe
	1037	#PBS -o $remote_dayfile
	1038
[892]	1039	$init_cmds
[493]	1040	$module_calls
[437]	1041
	1042	%%END%%
	1043
	1044	else
	1045	cat > $job_to_send << %%END%%
	1046	#!/bin/ksh
	1047	#PBS -S /bin/ksh
	1048	#PBS -N $job_name
	1049	#PBS -j oe
	1050	#PBS -l walltime=$timestring
	1051	#PBS -l ncpus=1
	1052	#PBS -l pmem=${memory}mb
	1053	#PBS -m abe
	1054	#PBS -o $remote_dayfile
	1055
[892]	1056	$init_cmds
[493]	1057	$module_calls
[437]	1058
	1059	%%END%%
	1060
	1061	fi
	1062
[1099]	1063	elif [[ $remote_host = lckyuh ]]
	1064	then
	1065	cat > $job_to_send << %%END%%
	1066	#!/bin/bash
	1067	#PJM -L "rscgrp=$queue"
	1068	#PJM -L "node=$nodes"
	1069	#PJM --mpi "proc=$numprocs"
	1070	#PJM -L "elapse=$timestring"
	1071	#PJM -o $remote_dayfile
	1072	#PJM -j
	1073	#PJM -X
	1074	#PJM --no-stging
	1075
	1076	export LANG=en_US.UTF-8
	1077	%%END%%
	1078
[1090]	1079	elif [[ $remote_host = lckyut ]]
	1080	then
	1081	cat > $job_to_send << %%END%%
	1082	#!/bin/bash
[1099]	1083	#PJM -L "rscgrp=$queue"
[1090]	1084	#PJM -L "vnode=$numprocs"
	1085	#PJM -L "vnode-core=1"
	1086	#PJM -L "elapse=$timestring"
	1087	#PJM --mpi proc=$numprocs
	1088	#PJM -o $remote_dayfile
	1089	#PJM -j
[1099]	1090	#PJM -X
	1091	#PJM --no-stging
[1090]	1092
	1093	export LANG=en_US.UTF-8
	1094	%%END%%
	1095
[1]	1096	elif [[ $remote_host = nech ]]
	1097	then
	1098
	1099	if (( nodes > 1 ))
	1100	then
	1101	cat > $job_to_send << %%END%%
	1102	#!/bin/ksh
[622]	1103	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1104	#PBS -l ${qsubmem}=${Memory}gb
	1105	#PBS -b $nodes
	1106	#PBS -o $remote_dayfile
	1107	#PBS -N palm
	1108	#PBS -j o
	1109	#PBS -T mpisx
	1110
	1111	%%END%%
	1112
	1113	elif [[ $numprocs != 0 ]]
	1114	then
	1115	cat > $job_to_send << %%END%%
	1116	#!/bin/ksh
[622]	1117	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1118	#PBS -l ${qsubmem}=${Memory}gb
	1119	#PBS -o $remote_dayfile
	1120	#PBS -N palm
	1121	#PBS -j o
	1122
	1123	%%END%%
	1124
	1125	else
	1126	cat > $job_to_send << %%END%%
	1127	#!/bin/ksh
	1128	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1129	#PBS -o $remote_dayfile
	1130	#PBS -j o
	1131
	1132	%%END%%
	1133
	1134	fi
	1135
	1136	elif [[ $remote_host = lctit ]]
	1137	then
	1138	cat > $job_to_send << %%END%%
[635]	1139	#!/bin/ksh
[892]	1140	$init_cmds
[678]	1141	$module_calls
	1142
[1]	1143	%%END%%
	1144
[1289]	1145	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1146	if [[ $tasks_per_node != $processes_per_node ]]
	1147	then
[1378]	1148	submcom="$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 $queue "
[678]	1149	else
[1378]	1150	submcom="$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]	1151	fi
[1]	1152
	1153	else
	1154
	1155	cat > $job_to_send << %%END%%
	1156	# @\$-q ${queue}
	1157	# @\$-l${qsubtime} $timestring
	1158	# @\$-l${qsubmem} ${memory}mb
	1159	# @\$-o $remote_dayfile
	1160	# @\$-eo
	1161
	1162	%%END%%
	1163
	1164	fi
	1165
	1166
[1289]	1167	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1168	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1169	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1170	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1171	then
	1172	echo "set +vx" >> $job_to_send
	1173	echo "trap '" >> $job_to_send
	1174	echo "set +vx" >> $job_to_send
[1289]	1175	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]	1176	then
[622]	1177	if [[ $remote_host = ibmh ]]
[1]	1178	then
	1179	return_queue=c1
[693]	1180	elif [[ $remote_host = ibmkisti ]]
	1181	then
	1182	return_queue=class.1-2
[622]	1183	elif [[ $remote_host = ibmku ]]
	1184	then
	1185	return_queue=sdbg2
[1]	1186	elif [[ $remote_host = ibms ]]
	1187	then
	1188	return_queue=p_normal
[1274]	1189	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1190	then
	1191	return_queue=dataq
[1040]	1192	elif [[ $remote_host = lckiaps ]]
	1193	then
	1194	return_queue=express
[1099]	1195	elif [[ $remote_host = lckyuh ]]
	1196	then
	1197	return_queue=cx-single
[1090]	1198	elif [[ $remote_host = lckyut ]]
	1199	then
	1200	return_queue=cx-single
[1]	1201	else
	1202	return_queue=unknown
	1203	fi
	1204
	1205	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1206	then
	1207
[622]	1208	if [[ $remote_host = ibmku ]]
	1209	then
[1289]	1210	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1211	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1212	else
[1289]	1213	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1214	fi
[1289]	1215	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1216	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1217	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1218	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1219	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1220	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1221	if [[ $host != "ibmh" ]]
	1222	then
[1289]	1223	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1224	fi
[1289]	1225	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1226	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1227
[1289]	1228	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1229	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1230
[1289]	1231	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
	1232	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]	1233	if [[ $remote_host = ibmku ]]
	1234	then
[1289]	1235	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1236	fi
[1289]	1237	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1238
	1239	elif [[ $remote_host = nech ]]
	1240	then
	1241	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1242	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1243	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1244	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1245	echo "#PBS -j o" >> $job_to_send
	1246	echo " " >> $job_to_send
	1247	echo "set -x" >> $job_to_send
	1248	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1094]	1249	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1250	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1251	echo "%%END%%" >> $job_to_send
	1252
[1099]	1253	elif [[ $remote_host = lckyuh ]]
	1254	then
[1289]	1255	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1256	echo "#!/bin/bash" >> $job_to_send
	1257	echo "#PJM -L \"node=1\"" >> $job_to_send
	1258	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1259	echo "#PJM --no-stging" >> $job_to_send
	1260	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1261	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1262	echo "#PJM -j" >> $job_to_send
	1263	echo " " >> $job_to_send
	1264	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1265	echo "set -x" >> $job_to_send
	1266	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1267	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1268	echo "%%END%%" >> $job_to_send
	1269
[1090]	1270	elif [[ $remote_host = lckyut ]]
	1271	then
[1289]	1272	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1273	echo "#!/bin/bash" >> $job_to_send
	1274	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1275	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1276	echo "#PJM --no-stging" >> $job_to_send
	1277	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1278	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1279	echo "#PJM -j" >> $job_to_send
	1280	echo " " >> $job_to_send
	1281	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1282	echo "set -x" >> $job_to_send
[1094]	1283	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1284	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1285	echo "%%END%%" >> $job_to_send
	1286
[1274]	1287	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1288	then
[1289]	1289	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1290	echo "#!/bin/bash" >> $job_to_send
	1291	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1292	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1293	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1294	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1295	echo "#PBS -j oe" >> $job_to_send
	1296	echo " " >> $job_to_send
	1297	echo "set -x" >> $job_to_send
	1298	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1299	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1300	echo "%%END%%" >> $job_to_send
	1301
[1099]	1302	elif [[ $remote_host = lcflow ]]
	1303	then
[1289]	1304	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1305	echo "#!/bin/bash" >> $job_to_send
	1306	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1307	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1308	echo "SGEPREFIX -cwd" >> $job_to_send
	1309	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1310	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
	1311	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
[1099]	1312	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1313	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1314	echo " " >> $job_to_send
	1315	echo "set -x" >> $job_to_send
	1316	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1317	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1318	echo "" >> $job_to_send
	1319	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1320	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1321	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1322	echo "%%END%%" >> $job_to_send
[1289]	1323	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1324	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1099]	1325
[1]	1326	else
	1327
[1289]	1328	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1329	echo "# @\\\$-q $return_queue" >> $job_to_send
	1330	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1331	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1332	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1333	then
	1334	echo "# @\$-l mpp_p=0" >> $job_to_send
	1335	fi
	1336	echo '# @\$-lF 10mb' >> $job_to_send
	1337	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1338	echo '# @\\\$-eo' >> $job_to_send
	1339	echo " " >> $job_to_send
	1340	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1341	then
	1342	echo "set +vx" >> $job_to_send
	1343	echo ". .profile" >> $job_to_send
	1344	fi
	1345	echo "set -x" >> $job_to_send
[1094]	1346	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]	1347	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1348	echo "%%END%%" >> $job_to_send
	1349	fi
	1350
	1351	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1352	then
[1289]	1353	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1354	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1355	then
[1289]	1356	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1357	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1358	then
[1289]	1359	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1360	elif [[ $remote_host = t3es ]]
	1361	then
[1289]	1362	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1363	elif [[ $remote_host = lckiaps ]]
	1364	then
[1289]	1365	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1366	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
	1367	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1368	elif [[ $remote_host = lckyu* ]]
[1090]	1369	then
[1289]	1370	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
	1371	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1372	elif [[ $remote_host = lcflow ]]
	1373	then
[1289]	1374	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1375	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1376	else
[1289]	1377	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1378	fi
[1043]	1379	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1380	then
[1289]	1381	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1382	fi
[1]	1383	if [[ $remote_host = nech ]]
	1384	then
	1385	echo "cd -" >> $job_to_send
	1386	fi
	1387	else
	1388	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1389	# ??? funktioniert das ÃŒberhaupt noch ???
[1]	1390	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1391	fi
	1392	echo "set -x" >> $job_to_send
	1393	echo " ' exit" >> $job_to_send
	1394	echo "set -x" >> $job_to_send
	1395	fi
	1396
	1397
	1398
[1289]	1399	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1400	cat $file_to_send >> $job_to_send
[69]	1401
[1]	1402	if [[ $remote_host = ibm ]]
	1403	then
	1404	echo " " >> $job_to_send
	1405	echo "exit" >> $job_to_send
	1406	fi
[635]	1407
[1289]	1408	# REMOVE JOB-FILE
[1099]	1409	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1410	then
	1411	echo " " >> $job_to_send
	1412	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1413	fi
[1]	1414
	1415
[1289]	1416	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1417	if [[ $no_submit = false ]]
	1418	then
	1419	if [[ $remote_host != $local_host ]]
	1420	then
	1421	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1422	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1423	then
[82]	1424	job_catalog_save=$job_catalog
	1425	job_catalog=job_queue
	1426	elif [[ $remote_host = nech ]]
	1427	then
	1428	job_catalog_save=$job_catalog
	1429	job_catalog=/hpf/b/${remote_user}/job_queue
	1430	fi
[1096]	1431	if [[ $remote_host = nech ]]
[82]	1432	then
[1289]	1433	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1434	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1435	else
[1094]	1436	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1437	fi
[82]	1438	if [[ $? = 1 ]]
	1439	then
	1440	locat=scp; exit
	1441	fi
	1442	if [[ $remote_host = ibms ]]
	1443	then
	1444	job_catalog=$job_catalog_save
	1445	fi
[1]	1446	[[ $verify = true ]] && printf "\n >>> finished\n"
	1447	else
	1448	eval job_catalog=$job_catalog
	1449	cp $job_to_send ${job_catalog}/$job_on_remhost
	1450	fi
	1451
	1452
	1453
[1289]	1454	# START NQS- / LOADLEVELER-JOB
[1]	1455	if [[ $remote_host != $local_host ]]
	1456	then
	1457	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1458
[1289]	1459	if [[ $remote_host = ibmku ]]
[1]	1460	then
[1094]	1461	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1462	elif [[ $remote_host = lcflow ]]
	1463	then
[1202]	1464	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1465	else
[1094]	1466	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1467	fi
[1]	1468
	1469	[[ $verify = true ]] && printf " >>> o.k.\n"
	1470	else
	1471	cd $job_catalog
[1289]	1472	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1473	then
	1474	eval $submcom $job_on_remhost
[1289]	1475	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1476	then
[635]	1477	chmod u+x $job_on_remhost
[108]	1478	eval $submcom $job_on_remhost
[1]	1479	elif [[ $local_host = nech ]]
	1480	then
	1481	if [[ $queue = default ]]
	1482	then
[799]	1483	eval $submcom $job_on_remhost
[1]	1484	else
[799]	1485	eval $submcom -q $queue $job_on_remhost
[1]	1486	fi
	1487	else
	1488	qsub $job_on_remhost
	1489	fi
[622]	1490
[1289]	1491	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1492	# AT THE END OF THE JOB
[1099]	1493	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1494	then
	1495	rm $job_on_remhost
	1496	fi
[1]	1497	cd - > /dev/null
	1498	fi
	1499	fi
	1500
	1501
	1502
[1289]	1503	# FINAL ACTIONS
[1]	1504	if [[ $no_submit = false ]]
	1505	then
[1099]	1506	rm -f $job_to_send
[1]	1507	fi
[1266]	1508	[[ $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 |