Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1370

Last change on this file since 1370 was 1351, checked in by maronga, 11 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 50.3 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	#
	26	#
[1046]	27	# Former revisions:
	28	# -----------------
[169]	29	# $Id: subjob 1351 2014-04-04 13:10:08Z raasch $
[1200]	30	#
[1351]	31	# 1350 2014-04-04 13:01:30Z maronga
	32	# location of qsub updated for lcxe6
	33	#
[1290]	34	# 1289 2014-03-04 07:12:34Z raasch
	35	# German comments translated to English
	36	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
	37	#
[1280]	38	# 1279 2014-01-28 12:10:14Z raasch
	39	# node calculation modified due to changes in mrun (tasks_per_node must not be
	40	# an integral divisor of numprocs any more)
	41	#
[1275]	42	# 1274 2014-01-09 13:14:54Z heinze
	43	# adjustments for lccrayh
	44	#
[1267]	45	# 1266 2013-12-11 12:07:34Z heinze
	46	# further adjustments for lccrayb (use msub instead of qsub)
	47	#
[1265]	48	# 1264 2013-12-09 12:46:09Z fricke
	49	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	50	#
[1263]	51	# 1262 2013-12-09 10:57:20Z fricke
	52	# further adjustments for lccrayb
	53	#
[1261]	54	# 1260 2013-12-04 12:48:04Z raasch
	55	# jaboticaba admitted
	56	#
[1256]	57	# 1255 2013-11-07 14:43:35Z raasch
	58	# further adjustments for lccrayb
	59	#
[1225]	60	# 1224 2013-09-16 07:27:23Z raasch
	61	# first adjustments for lccrayb
	62	#
[1203]	63	# 1202 2013-07-10 16:22:07Z witha
	64	# adjustments for Forwind cluster (lcflow)
	65	#
[1200]	66	# 1199 2013-07-05 14:52:22Z raasch
	67	# adjustments for CSC Helsinki (lccrayf)
	68	#
[1185]	69	# use of cluster/express queue enabled (ibmh)
	70	# vinessa added (imuk)
[1047]	71	#
[1104]	72	# 1103 2013-02-20 02:15:53Z raasch
	73	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	74	# instead typeset -L/R),
	75	# further adjustments for lckyuh
	76	#
[1100]	77	# 2013-02-10 01:47:43Z raasch
	78	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	79	# and for Forwind cluster (lcflow)
	80	#
[1096]	81	# 1094 2013-02-03 01:52:12Z raasch
	82	# new option -P for explicit setting of ssh/scp port,
	83	# decalpha parts (yonsei) removed
	84	#
[1091]	85	# 2013-02-02 07:06:13Z raasch
[1099]	86	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	87	# old changelog messages removed
	88	#
[1047]	89	# 1046 2012-11-09 14:38:45Z maronga
	90	# code put under GPL (PALM 3.9)
	91	#
[1090]	92	# 08/07/94 - Siggi - first version finished
	93	# 29/06/94 - Siggi - script development started
	94	#--------------------------------------------------------------------------------#
	95	# subjob - script for automatic generation and submission of batch-job files
	96	# for various batch queuing systems
	97	#--------------------------------------------------------------------------------#
[1]	98
	99
[1289]	100	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	101	delete_dayfile=false
[799]	102	email_notification=none
[122]	103	group_number=none
[1]	104	locat=normal
	105	no_default_queue=none
	106	no_submit=false
	107	job_catalog="~/job_queue"
	108	job_name=none
	109	local_user=$LOGNAME
	110	node_usage=shared
[475]	111	numprocs=0
[1]	112	punkte="..........................................................."
	113	submcom=qsub
	114	queue=default
	115	remote_host=none
	116	remote_user=""
	117	verify=true
	118
	119	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	120	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	121
	122
	123
[1289]	124	# ERROR HANDLING
	125	# IN CASE OF EXIT:
[1]	126	trap 'if [[ $locat != normal ]]
	127	then
	128	case $locat in
	129	(option) printf "\n --> available optios can be displayed"
	130	printf " by typing:"
	131	printf "\n \"subjob ?\" \n";;
	132	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	133	(*) printf "\n +++ unknown error"
	134	printf "\n please inform S. Raasch!\n"
	135	esac
	136	[[ -f $job_to_send ]] && rm $job_to_send
	137	printf "\n\n+++ SUBJOB killed \n\n"
	138	fi' exit
	139
	140
[1289]	141	# IN CASE OF TERMINAL-BREAK:
[1]	142	trap '[[ -f $job_to_send ]] && rm $job_to_send
	143	printf "\n\n+++ SUBJOB killed \n\n"
	144	exit
	145	' 2
	146
	147
	148
	149
[1289]	150	# DETERMINE NAME OF LOCAL HOST
[1]	151	local_host=$(hostname)
	152
	153
	154
[1289]	155	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	156	# IS ADMITTED AT ALL)
	157	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	158	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	159	case $local_host in
[1204]	160	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
[1]	161	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[1235]	162	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
[102]	163	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	164	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1103]	165	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
[749]	166	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[1255]	167	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
[1274]	168	(hlogin\|hxc) local_addres=130.75.4.1; local_host=lccrayh;;
[12]	169	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[1174]	170	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
[372]	171	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	172	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[1197]	173	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
[1]	174	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	175	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[1099]	176	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
	177	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
[367]	178	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	179	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	180	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	181	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[5]	182	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	183	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1099]	184	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
[892]	185	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[440]	186	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	187	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	188	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[1260]	189	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
[407]	190	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	191	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
[1040]	192	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
[5]	193	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1099]	194	(meller) local_addres=134.106.74.155; local_host=lcfor;;
[437]	195	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	196	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	197	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	198	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	199	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	200	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[1099]	201	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
[789]	202	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	203	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	204	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1204]	205	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
[1197]	206	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
[799]	207	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1090]	208	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
[635]	209	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1021]	210	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
[1184]	211	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
[1]	212	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	213	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	214	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	215	printf "\n please contact the PALM group at IMUK";
[1]	216	locat=parameter; exit;;
	217	esac
	218
	219
	220
[1289]	221	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	222	remote_host=$local_host
	223
	224
	225
	226
[1289]	227	# READ THE SHELLSCRIPT-OPTIONS
[1094]	228	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	229	do
	230	case $option in
	231	(c) job_catalog=$OPTARG;;
	232	(d) delete_dayfile=true;;
	233	(D) no_submit=true;;
[352]	234	(e) email_notification=$OPTARG;;
[125]	235	(g) group_number=$OPTARG;;
[1]	236	(h) remote_host=$OPTARG;;
	237	(m) memory=$OPTARG;;
	238	(n) job_name=$OPTARG;;
	239	(N) node_usage=$OPTARG;;
	240	(O) threads_per_task=$OPTARG;;
[1094]	241	(P) scp_port=$OPTARG;;
[1]	242	(q) no_default_queue=$OPTARG;;
	243	(t) cputime=$OPTARG;;
	244	(T) tasks_per_node=$OPTARG;;
	245	(u) remote_user=$OPTARG;;
	246	(v) verify=false;;
	247	(X) numprocs=$OPTARG;;
	248	(\?) printf "\n +++ Option $OPTARG unknown \n";
	249	locat=option; exit;;
	250	esac
	251	done
	252
	253
[1289]	254	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	255	(( to_shift = $OPTIND - 1 ))
	256	shift $to_shift; file_to_send=$1
[1]	257
	258
[1289]	259	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	260	if [ "$1" = "?" ]
	261	then
	262	(printf "\n *** subjob can be called as follows:\n"
	263	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	264	printf "\n Description of available options:\n"
	265	printf "\n Option Description Default-Value"
	266	printf "\n -c job-input- and output-catalog ~/job_queue"
	267	printf "\n -d no job-protocol will be created ---"
	268	printf "\n -D only the job-file will be created ---"
	269	printf "\n -h execution host, available hosts: $remote_host"
[1289]	270	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	271	printf "\n lckiaps, lctit, nech"
[1]	272	printf "\n -m memory demand per process in MByte ---"
	273	printf "\n -n jobname <jobdatei>"
	274	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	275	printf "\n -P ssh/scp port default port"
[1]	276	printf "\n -q job-queue to be used default"
	277	printf "\n -t allowed cpu-time in seconds ---"
	278	printf "\n -T tasks per node (on parallel hosts) ---"
	279	printf "\n -u username on execution host from .netrc"
	280	printf "\n -v no prompt for confirmation ---"
	281	printf "\n -X # of processors (on parallel hosts) 1"
	282	printf "\n "
	283	printf "\n The only possible positional parameter is <jobfile>:"
	284	printf "\n The complete NQS-job must be provided here."
	285	printf "\n <jobfile>=? creates this outline\n\n") \| more
	286	exit
	287	fi
	288
	289
	290
[1289]	291	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	292	if [[ "$file_to_send" = "" ]]
	293	then
	294	printf "\n +++ job-file missing"
	295	locat=parameter; exit
	296	else
	297	if [[ -f $file_to_send ]]
	298	then
	299	true
	300	else
	301	printf "\n +++ job-file: "
	302	printf "\n $file_to_send"
	303	printf "\n does not exist"
	304	locat=parameter; exit
	305	fi
	306	fi
	307
	308
	309
[1289]	310	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	311	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	312	if [[ $job_name = none ]]
	313	then
	314	job_name=$file_to_send
	315	fi
	316	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	317	then
	318	printf "\n +++ job-file name: "
	319	printf "\n $job_name"
	320	printf "\n must not contain \"/\"-characters"
	321	locat=parameter; exit
	322	fi
	323
	324
	325
	326
[1289]	327	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	328	# OR IF NO HOST HAS BEEN GIVEN
[1]	329	if [[ $remote_host = none ]]
	330	then
	331	printf "\n +++ host missing"
	332	locat=option; exit
	333	else
	334	case $remote_host in
[251]	335	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1184]	336	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	337	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	338	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	339	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1262]	340	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1274]	341	(lccrayh) queue=mpp1testq; remote_addres=130.75.4.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1197]	342	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
[1099]	343	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
[440]	344	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[367]	345	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[1040]	346	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
[717]	347	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[1099]	348	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
[1090]	349	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
[1021]	350	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
[635]	351	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[1350]	352	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
[437]	353	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	354	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	355	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	356	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	357	locat=parameter; exit;;
[1]	358	esac
	359	fi
	360
	361
[1289]	362	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	363	if [[ $no_default_queue != none ]]
	364	then
	365	error=false
	366	ndq=$no_default_queue
	367	case $remote_host in
	368	(ibm) case $ndq in
	369	(p690_express\|p690_standard\|p690_long) error=false;;
	370	(*) error=true;;
	371	esac;;
	372	(ibmh) case $ndq in
[1184]	373	(cluster\|express) error=false;;
[1]	374	(*) error=true;;
	375	esac;;
[693]	376	(ibmkisti) case $ndq in
	377	(class.32plus\|class.1-2\|class.2-32) error=false;;
	378	(*) error=true;;
	379	esac;;
[622]	380	(ibmku) case $ndq in
	381	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	382	(*) error=true;;
	383	esac;;
[1]	384	(ibms) case $ndq in
	385	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	386	(*) error=true;;
	387	esac;;
[1224]	388	(lccrayb) case $ndq in
[1274]	389	(mpp1q\|testq\|specialm1q) error=false;;
[1224]	390	(*) error=true;;
	391	esac;;
[1274]	392	(lccrayh) case $ndq in
	393	(mpp1q\|mpp1testq\|specialm1q\|dataq) error=false;;
	394	(*) error=true;;
	395	esac;;
[1197]	396	(lccrayf) case $ndq in
	397	(usup\|test*\|small\|large) error=false;;
	398	(*) error=true;;
	399	esac;;
[1099]	400	(lcflow) case $ndq in
	401	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
	402	(*) error=true;;
	403	esac;;
[1040]	404	(lckiaps) case $ndq in
	405	(express\|normal) error=false;;
	406	(*) error=true;;
	407	esac;;
[440]	408	(lckyoto) case $ndq in
	409	(eh\|ph) error=false;;
	410	(*) error=true;;
	411	esac;;
[1099]	412	(lckyuh) case $ndq in
	413	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	414	(*) error=true;;
	415	esac;;
[1090]	416	(lckyut) case $ndq in
	417	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	418	(*) error=true;;
	419	esac;;
[1]	420	(lctit) case $ndq in
[635]	421	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	422	(*) error=true;;
	423	esac;;
	424	(t3eb) case $ndq in
	425	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	426	(*) error=true;;
	427	esac;;
	428	(t3eh) case $ndq in
	429	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	430	(*) error=true;;
	431	esac;;
	432	(t3ej2\|t3ej5) case $ndq in
	433	(low\|normal\|high) error=false;;
	434	(*) error=true;;
	435	esac;;
	436	(t3es) case $ndq in
	437	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	438	(*) error=true;;
	439	esac;;
	440	esac
	441	if [[ $error = true ]]
	442	then
	443	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	444	locat=parameter; exit
	445	else
	446	queue=$no_default_queue
	447	fi
	448	fi
	449
	450
	451
[1289]	452	# CHECK THE CPU-TIME
	453	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	454	done=false
	455	while [[ $done = false ]]
	456	do
	457	if (( $cputime <= 0 ))
	458	then
	459	printf "\n +++ wrong cpu-time or cpu-time missing"
	460	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	461	printf "\n >>> "
	462	read cputime 1>/dev/null 2>&1
	463	else
	464	done=true
	465	fi
	466	done
	467	if [[ $remote_host = nech ]]
	468	then
	469	if (( tasks_per_node != 0 ))
	470	then
	471	(( cputime = cputime * tasks_per_node ))
	472	elif [[ $numprocs != 0 ]]
	473	then
	474	(( cputime = cputime * numprocs ))
	475	fi
	476	fi
	477	(( stunden = cputime / 3600 ))
	478	(( resttime = cputime - stunden * 3600 ))
	479	(( minuten = resttime / 60 ))
	480	(( sekunden = resttime - minuten * 60 ))
	481	timestring=${stunden}:${minuten}:${sekunden}
	482
	483
	484
[1289]	485	# CHECK THE MEMORY DEMAND
[1]	486	done=false
	487	while [[ $done = false ]]
	488	do
	489	if (( memory <= 0 ))
	490	then
	491	printf "\n +++ wrong memory demand or memory demand missing"
	492	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	493	printf "\n >>> "
	494	read memory 1>/dev/null 2>&1
	495	else
	496	done=true
	497	fi
	498	done
	499
[1289]	500	if [[ $remote_host = nech ]]
[1]	501	then
	502	if (( tasks_per_node != 0 ))
	503	then
	504	(( Memory = memory * tasks_per_node / 1000 ))
	505	elif [[ $numprocs != 0 ]]
	506	then
	507	(( Memory = memory * numprocs / 1000 ))
	508	else
	509	(( Memory = memory / 1000 ))
	510	fi
[635]	511	elif [[ $remote_host = lctit ]]
	512	then
	513	(( Memory = memory * tasks_per_node / 1000 ))
[1]	514	fi
	515
	516
[1289]	517	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	518	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	519	then
	520	(( memory = memory * threads_per_task ))
	521	fi
	522
	523
[1289]	524	# CALCULATE NUMBER OF REQUIRED NODES
[1]	525	if (( tasks_per_node != 0 ))
	526	then
[1279]	527	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	528	fi
	529
[1094]	530
[1289]	531	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	532	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	533
[1094]	534
[1289]	535	# CALCULATE NUMBER OF MPI TASKS
[696]	536	(( mpi_tasks = numprocs / threads_per_task ))
[1]	537
[696]	538
[1289]	539	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	540	if [[ "$scp_port" != "" ]]
	541	then
	542	PORTOPT="-P $scp_port"
	543	SSH_PORTOPT="-p $scp_port"
	544	fi
	545
	546
[1289]	547	# HEADER-OUTPUT
[1]	548	if [[ $verify = true ]]
	549	then
	550	printf "\n\n"
	551	printf "#--------------------------------------------------------------# \n"
	552	spalte1=SUBJOB;spalte2=$(date)
[1103]	553	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	554	printf "\| \| \n"
	555	printf "\| values of parameters/options: \| \n"
[1103]	556	spalte1=$(echo local_host$punkte \| cut -c-20)
	557	spalte2=$punkte$local_host
	558	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	559	spalte1=$(echo remote_host$punkte \| cut -c-20)
	560	spalte2=$punkte$remote_host
	561	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	562	spalte1=$(echo queue$punkte \| cut -c-20)
	563	spalte2=$punkte$queue
	564	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	565	spalte1=$(echo memory$punkte \| cut -c-20)
	566	spalte2="$punkte$memory mb"
	567	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	568	spalte1=$(echo cputime$punkte \| cut -c-20)
	569	spalte2="$punkte$cputime sec"
	570	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	571	spalte1=$(echo job_name$punkte \| cut -c-20)
	572	spalte2="$punkte$job_name"
	573	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	574	printf "#--------------------------------------------------------------# \n\n"
	575
	576
[1289]	577	# QUERY CHECK
[1]	578	antwort="dummy"
	579	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	580	do
	581	read antwort?" >>> continue (y/n) ? "
	582	done
	583	if [[ $antwort = n \|\| $antwort = N ]]
	584	then
	585	locat=verify; exit
	586	fi
	587	printf "\n"
	588	fi
	589
	590
	591
[1289]	592	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	593	identifier=$RANDOM
	594	job_on_remhost=${job_name}_${identifier}_$local_host
	595	job_to_send=job_to_send_$identifier
[1]	596	if [[ $delete_dayfile = false ]]
	597	then
[1289]	598	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	599	local_dayfile=${remote_host}_${job_name}
	600	else
	601	remote_dayfile=/dev/null
	602	fi
	603
	604
	605
[1289]	606	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	607	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	608	then
	609
[1289]	610	# GENERAL LOADLEVELER SETTINGS
[622]	611	execute_in_shell="#!/bin/ksh"
	612	use_shell="# @ shell = /bin/ksh"
	613	consumable_memory="ConsumableMemory($memory mb)"
	614	class="# @ class = $queue"
	615	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	616	network_to_use="# @ network.mpi = sn_all,shared,us"
	617	data_limit="# @ data_limit = 1.76gb"
	618	image_size="# @ image_size = 50"
[693]	619	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	620
[693]	621	if [[ $email_notification = none ]]
	622	then
	623	notify_user=""
	624	else
	625	notify_user="# @ notify_user = $email_notification"
	626	if [[ $delete_dayfile = true ]]
	627	then
	628	notification='# @ notification = never'
	629	fi
	630	fi
[622]	631
[312]	632	if [[ $remote_host = ibmh ]]
[1]	633	then
[312]	634	data_limit=""
	635	network_to_use=""
[1184]	636	class="# @ class = $queue"
[312]	637	environment=""
[814]	638	rset="# @ rset = RSET_MCM_AFFINITY"
	639	task_affinity="# @ task_affinity = core(1)"
[693]	640	elif [[ $remote_host = ibmkisti ]]
	641	then
	642	network_to_use="# @ network.MPI = sn_all,shared,US"
	643	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	644	if [[ $threads_per_task = 1 ]]
	645	then
	646	rset="# @ rset = RSET_MCM_AFFINITY"
	647	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	648	fi
[693]	649	environment=""
	650	use_shell=""
	651	data_limit=""
	652	image_size=""
[622]	653	elif [[ $remote_host = ibmku ]]
	654	then
	655	execute_in_shell="#!/usr/bin/ksh"
	656	use_shell="# @ shell = /usr/bin/ksh"
	657	consumable_memory=""
	658	environment=""
	659	network_to_use="# @ network.mpi = sn_all,shared,us"
	660	data_limit=""
	661	image_size=""
	662	elif [[ $remote_host = ibms ]]
	663	then
	664	network_to_use="# @ network.mpi = csss,shared,us"
[1]	665	fi
	666
	667	cat > $job_to_send << %%END%%
[622]	668	$execute_in_shell
	669	$use_shell
[1]	670
	671	# @ job_type = parallel
[693]	672	# @ job_name = $job_name
[1]	673	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	674	# @ output = $remote_dayfile
	675	# @ error = $remote_dayfile
[693]	676	$wall_clock_limit
[622]	677	$image_size
[312]	678	$class
	679	$environment
[1]	680	$network_to_use
	681	$data_limit
[693]	682	$rset
	683	$mcm_affinity_options
[814]	684	$task_affinity
[1]	685	$notification
[693]	686	$notify_user
[1]	687
	688	%%END%%
	689
	690	if (( nodes > 0 ))
	691	then
	692
[693]	693	if [[ $remote_host != ibmkisti ]]
	694	then
	695
	696	cat >> $job_to_send << %%END%%
[1]	697	# @ node = $nodes
[622]	698	# @ tasks_per_node = $processes_per_node
[1]	699	# @ node_usage = $node_usage
	700	# @ queue
	701
	702	%%END%%
	703
[693]	704	else
	705
	706	cat >> $job_to_send << %%END%%
[696]	707	# @ total_tasks = $mpi_tasks
[693]	708	# @ blocking = unlimited
	709	# @ queue
	710
	711	%%END%%
	712
	713	fi
	714
[1]	715	else
	716
[1289]	717	cat >> $job_to_send << %%END%%
[1]	718	# @ blocking = unlimited
	719	# @ total_tasks = $numprocs
	720	# @ node_usage = $node_usage
	721	# @ queue
	722
	723	%%END%%
	724
	725	fi
	726
[1289]	727	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	728	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	729	then
	730	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	731	fi
	732
[1]	733	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	734	then
	735
	736	cat > $job_to_send << %%END%%
	737	#!/bin/ksh
	738
	739	# @ job_type = serial
	740	# @ node_usage = $node_usage
	741	# @ job_name = palm
	742	# @ wall_clock_limit = ${timestring},$timestring
	743	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	744	# @ output = $remote_dayfile
	745	# @ error = $remote_dayfile
[312]	746	$class
[1]	747	$notification
	748
	749	# @ queue
	750
	751	%%END%%
	752
[1274]	753	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	754	then
	755
	756	if [[ $numprocs != 0 ]]
	757	then
	758	cat > $job_to_send << %%END%%
[1255]	759	#!/bin/bash
[1224]	760	#PBS -N $job_name
	761	#PBS -l walltime=$timestring
[1264]	762	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	763	#PBS -o $remote_dayfile
	764	#PBS -j oe
	765	#PBS -q $queue
	766
	767	$module_calls
	768
	769	%%END%%
	770
	771	else
	772
	773	continue
	774
	775	fi
	776
[1197]	777	elif [[ $remote_host = lccrayf ]]
	778	then
	779
	780	if [[ $numprocs != 0 ]]
	781	then
	782	cat > $job_to_send << %%END%%
	783	#!/bin/bash -l
	784	#SBATCH -J $job_name
	785	#SBATCH -t $timestring
	786	#SBATCH -N $nodes
	787	#SBATCH --ntasks-per-node=$processes_per_node
	788	#SBATCH -p $queue
	789	#SBATCH -o $remote_dayfile
	790	#SBATCH -e $remote_dayfile
	791
	792	$init_cmds
	793	$module_calls
	794
	795	%%END%%
	796
	797	else
	798	cat > $job_to_send << %%END%%
	799	#!/bin/bash -l
	800	#SBATCH -J $job_name
	801	#SBATCH -t $timestring
	802	#SBATCH -l ncpus=1
	803	#SBATCH -l pmem=${memory}mb
	804	#SBATCH -m abe
	805	#SBATCH -o $remote_dayfile
	806	#SBATCH -e $remote_dayfile
	807
	808	$init_cmds
	809	$module_calls
	810
	811	%%END%%
	812
	813	fi
	814
[1099]	815	elif [[ $remote_host = lcflow ]]
	816	then
[1202]	817	if [ $memory -gt 1800 ]; then
[1099]	818	use_himem=""
	819	else
	820	use_himem="#"
	821	fi
	822	if [[ $numprocs != 0 ]]
	823	then
[1202]	824	pe_set="#$ -pe impi41 $numprocs"
[1099]	825	else
[1202]	826	pe_set="#$ -pe impi41 1"
[1099]	827	fi
	828	if [[ $queue = default ]]
	829	then
	830	queue_set=""
	831	else
	832	queue_set="#$ -q $queue"
	833	fi
	834	[[ "$disc_space" = "" ]] && disc_space=50
	835
	836	cat > $job_to_send << %%END%%
	837	#!/bin/bash
	838	#$ -S /bin/bash
	839	#$ -N $job_name
	840	#$ -cwd
	841	#$ -l h_rt=$timestring
	842	#$ -l h_vmem=${memory}M
	843	#$ -o $remote_dayfile
	844	#$ -j y
	845	$pe_set
	846	#$ -R y
	847	${use_himem}#$ -l highmem=true
	848	#$ -l h_fsize=${disc_space}G
	849	$queue_set
	850
	851	%%END%%
	852
[1021]	853	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	854	then
	855
	856	if [[ $numprocs != 0 ]]
	857	then
	858	cat > $job_to_send << %%END%%
	859	#!/bin/ksh
	860	#PBS -N $job_name
	861	#PBS -l walltime=$timestring
	862	#PBS -l ncpus=$numprocs
	863	#PBS -l pmem=${memory}mb
	864	#PBS -o $remote_dayfile
[1021]	865	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	866	#PBS -j oe
	867
	868	mpd &
	869
	870	%%END%%
	871
	872	else
	873	cat > $job_to_send << %%END%%
	874	#!/bin/ksh
	875	#PBS -N $job_name
	876	#PBS -l walltime=$timestring
	877	#PBS -l ncpus=1
	878	#PBS -l pmem=${memory}mb
	879	#PBS -o $remote_dayfile
	880	#PBS -j oe
	881
	882	%%END%%
	883
	884	fi
	885
[1040]	886	elif [[ $remote_host = lckiaps ]]
	887	then
	888
	889	if [[ $numprocs != 0 ]]
	890	then
	891	cat > $job_to_send << %%END%%
	892	#!/bin/ksh
	893	#PBS -N $job_name
	894	#PBS -l walltime=$timestring
	895	#PBS -l select=1:ncpus=$numprocs
	896	#PBS -l pmem=${memory}mb
	897	#PBS -q $queue
	898	#PBS -o $remote_dayfile
	899	#PBS -j oe
	900	#PBS -V
	901
	902	%%END%%
	903
	904	else
	905	cat > $job_to_send << %%END%%
	906	#!/bin/ksh
	907	#PBS -N $job_name
	908	#PBS -l walltime=$timestring
	909	#PBS -l ncpus=1
	910	#PBS -l pmem=${memory}mb
	911	#PBS -o $remote_dayfile
	912	#PBS -j oe
	913
	914	%%END%%
	915
	916	fi
	917
[693]	918	elif [[ $remote_host = lcyon ]]
	919	then
	920
	921	if [[ $numprocs != 0 ]]
	922	then
	923	cat > $job_to_send << %%END%%
	924	#!/bin/ksh
	925	#PBS -N $job_name
	926	#PBS -l walltime=$timestring
	927	#PBS -l ncpus=$numprocs
	928	#PBS -l pmem=${memory}mb
	929	#PBS -o $remote_dayfile
	930	#PBS -j oe
	931
	932	%%END%%
	933
	934	else
	935	cat > $job_to_send << %%END%%
	936	#!/bin/ksh
	937	#PBS -N $job_name
	938	#PBS -l walltime=$timestring
	939	#PBS -l ncpus=1
	940	#PBS -l pmem=${memory}mb
	941	#PBS -o $remote_dayfile
	942	#PBS -j oe
	943
	944	%%END%%
	945
	946	fi
	947
[892]	948	elif [[ $remote_host = lcxe6 ]]
[164]	949	then
	950
	951	if [[ $numprocs != 0 ]]
	952	then
	953	cat > $job_to_send << %%END%%
[799]	954	#!/bin/ksh
[164]	955	#PBS -S /bin/ksh
	956	#PBS -N $job_name
[552]	957	#PBS -A $project_account
[206]	958	#PBS -j oe
[164]	959	#PBS -l walltime=$timestring
	960	#PBS -l mppwidth=${numprocs}
[622]	961	#PBS -l mppnppn=${processes_per_node}
[164]	962	#PBS -m abe
	963	#PBS -o $remote_dayfile
[492]	964	$email_directive
[164]	965
[892]	966	$init_cmds
[493]	967	$module_calls
[343]	968
[164]	969	%%END%%
	970
[108]	971	else
	972	cat > $job_to_send << %%END%%
	973	#!/bin/ksh
[168]	974	#PBS -S /bin/ksh
[108]	975	#PBS -N $job_name
[552]	976	#PBS -A $project_account
[206]	977	#PBS -j oe
[108]	978	#PBS -l walltime=$timestring
	979	#PBS -l ncpus=1
	980	#PBS -l pmem=${memory}mb
	981	#PBS -m abe
[492]	982	$email_directive
[108]	983	#PBS -o $remote_dayfile
	984
[892]	985	$init_cmds
[493]	986	$module_calls
[343]	987
[108]	988	%%END%%
	989
	990	fi
	991
[440]	992	elif [[ $remote_host = lckyoto ]]
[437]	993	then
	994
[440]	995	cat > $job_to_send << %%END%%
[799]	996	#!/bin/ksh
[440]	997	# @\$-o $remote_dayfile
	998	# @\$-eo -oi
	999	# @\$-lP 16
[799]	1000	# @\$-lp 1
[440]	1001	# @\$-lm 28gb -llm unlimited -ls unlimited
	1002	# @\$-q $queue
	1003	# @\$-Pvn abs_pack
	1004	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1005	#. /thin/local/etc/setprofile/intel-11.0.sh
	1006	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1007	. ~/.myprofile
	1008	#. /home2/t/t51254/palm/current_version/myprofile
	1009	#. /thin/apps/pgi/mpi.sh
	1010	#
	1011	env
	1012	#
	1013	set -x
	1014
	1015	%%END%%
	1016
	1017	elif [[ $remote_host = lcxt5m ]]
	1018	then
	1019
[437]	1020	if [[ $numprocs != 0 ]]
	1021	then
	1022	cat > $job_to_send << %%END%%
[799]	1023	#!/bin/ksh
[437]	1024	#PBS -S /bin/ksh
	1025	#PBS -N $job_name
	1026	#PBS -j oe
	1027	#PBS -l walltime=$timestring
	1028	#PBS -l mppwidth=${numprocs}
[622]	1029	#PBS -l mppnppn=${processes_per_node}
[437]	1030	#PBS -m abe
	1031	#PBS -o $remote_dayfile
	1032
[892]	1033	$init_cmds
[493]	1034	$module_calls
[437]	1035
	1036	%%END%%
	1037
	1038	else
	1039	cat > $job_to_send << %%END%%
	1040	#!/bin/ksh
	1041	#PBS -S /bin/ksh
	1042	#PBS -N $job_name
	1043	#PBS -j oe
	1044	#PBS -l walltime=$timestring
	1045	#PBS -l ncpus=1
	1046	#PBS -l pmem=${memory}mb
	1047	#PBS -m abe
	1048	#PBS -o $remote_dayfile
	1049
[892]	1050	$init_cmds
[493]	1051	$module_calls
[437]	1052
	1053	%%END%%
	1054
	1055	fi
	1056
[1099]	1057	elif [[ $remote_host = lckyuh ]]
	1058	then
	1059	cat > $job_to_send << %%END%%
	1060	#!/bin/bash
	1061	#PJM -L "rscgrp=$queue"
	1062	#PJM -L "node=$nodes"
	1063	#PJM --mpi "proc=$numprocs"
	1064	#PJM -L "elapse=$timestring"
	1065	#PJM -o $remote_dayfile
	1066	#PJM -j
	1067	#PJM -X
	1068	#PJM --no-stging
	1069
	1070	export LANG=en_US.UTF-8
	1071	%%END%%
	1072
[1090]	1073	elif [[ $remote_host = lckyut ]]
	1074	then
	1075	cat > $job_to_send << %%END%%
	1076	#!/bin/bash
[1099]	1077	#PJM -L "rscgrp=$queue"
[1090]	1078	#PJM -L "vnode=$numprocs"
	1079	#PJM -L "vnode-core=1"
	1080	#PJM -L "elapse=$timestring"
	1081	#PJM --mpi proc=$numprocs
	1082	#PJM -o $remote_dayfile
	1083	#PJM -j
[1099]	1084	#PJM -X
	1085	#PJM --no-stging
[1090]	1086
	1087	export LANG=en_US.UTF-8
	1088	%%END%%
	1089
[1]	1090	elif [[ $remote_host = nech ]]
	1091	then
	1092
	1093	if (( nodes > 1 ))
	1094	then
	1095	cat > $job_to_send << %%END%%
	1096	#!/bin/ksh
[622]	1097	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1098	#PBS -l ${qsubmem}=${Memory}gb
	1099	#PBS -b $nodes
	1100	#PBS -o $remote_dayfile
	1101	#PBS -N palm
	1102	#PBS -j o
	1103	#PBS -T mpisx
	1104
	1105	%%END%%
	1106
	1107	elif [[ $numprocs != 0 ]]
	1108	then
	1109	cat > $job_to_send << %%END%%
	1110	#!/bin/ksh
[622]	1111	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1112	#PBS -l ${qsubmem}=${Memory}gb
	1113	#PBS -o $remote_dayfile
	1114	#PBS -N palm
	1115	#PBS -j o
	1116
	1117	%%END%%
	1118
	1119	else
	1120	cat > $job_to_send << %%END%%
	1121	#!/bin/ksh
	1122	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1123	#PBS -o $remote_dayfile
	1124	#PBS -j o
	1125
	1126	%%END%%
	1127
	1128	fi
	1129
	1130	elif [[ $remote_host = lctit ]]
	1131	then
	1132	cat > $job_to_send << %%END%%
[635]	1133	#!/bin/ksh
[892]	1134	$init_cmds
[678]	1135	$module_calls
	1136
[1]	1137	%%END%%
	1138
[1289]	1139	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1140	if [[ $tasks_per_node != $processes_per_node ]]
	1141	then
	1142	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -o $remote_dayfile -j oe -q $queue "
	1143	else
	1144	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -q $queue "
	1145	fi
[1]	1146
	1147	else
	1148
	1149	cat > $job_to_send << %%END%%
	1150	# @\$-q ${queue}
	1151	# @\$-l${qsubtime} $timestring
	1152	# @\$-l${qsubmem} ${memory}mb
	1153	# @\$-o $remote_dayfile
	1154	# @\$-eo
	1155
	1156	%%END%%
	1157
	1158	fi
	1159
	1160
[1289]	1161	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1162	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1163	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1164	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1165	then
	1166	echo "set +vx" >> $job_to_send
	1167	echo "trap '" >> $job_to_send
	1168	echo "set +vx" >> $job_to_send
[1289]	1169	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]	1170	then
[622]	1171	if [[ $remote_host = ibmh ]]
[1]	1172	then
	1173	return_queue=c1
[693]	1174	elif [[ $remote_host = ibmkisti ]]
	1175	then
	1176	return_queue=class.1-2
[622]	1177	elif [[ $remote_host = ibmku ]]
	1178	then
	1179	return_queue=sdbg2
[1]	1180	elif [[ $remote_host = ibms ]]
	1181	then
	1182	return_queue=p_normal
[1274]	1183	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1184	then
	1185	return_queue=dataq
[1040]	1186	elif [[ $remote_host = lckiaps ]]
	1187	then
	1188	return_queue=express
[1099]	1189	elif [[ $remote_host = lckyuh ]]
	1190	then
	1191	return_queue=cx-single
[1090]	1192	elif [[ $remote_host = lckyut ]]
	1193	then
	1194	return_queue=cx-single
[1]	1195	else
	1196	return_queue=unknown
	1197	fi
	1198
	1199	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1200	then
	1201
[622]	1202	if [[ $remote_host = ibmku ]]
	1203	then
[1289]	1204	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1205	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1206	else
[1289]	1207	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1208	fi
[1289]	1209	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1210	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1211	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1212	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1213	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1214	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1215	if [[ $host != "ibmh" ]]
	1216	then
[1289]	1217	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1218	fi
[1289]	1219	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1220	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1221
[1289]	1222	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1223	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1224
[1289]	1225	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
	1226	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]	1227	if [[ $remote_host = ibmku ]]
	1228	then
[1289]	1229	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1230	fi
[1289]	1231	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1232
	1233	elif [[ $remote_host = nech ]]
	1234	then
	1235	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1236	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1237	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1238	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1239	echo "#PBS -j o" >> $job_to_send
	1240	echo " " >> $job_to_send
	1241	echo "set -x" >> $job_to_send
	1242	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1094]	1243	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1244	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1245	echo "%%END%%" >> $job_to_send
	1246
[1099]	1247	elif [[ $remote_host = lckyuh ]]
	1248	then
[1289]	1249	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1250	echo "#!/bin/bash" >> $job_to_send
	1251	echo "#PJM -L \"node=1\"" >> $job_to_send
	1252	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1253	echo "#PJM --no-stging" >> $job_to_send
	1254	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1255	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1256	echo "#PJM -j" >> $job_to_send
	1257	echo " " >> $job_to_send
	1258	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1259	echo "set -x" >> $job_to_send
	1260	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1261	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1262	echo "%%END%%" >> $job_to_send
	1263
[1090]	1264	elif [[ $remote_host = lckyut ]]
	1265	then
[1289]	1266	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1267	echo "#!/bin/bash" >> $job_to_send
	1268	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1269	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1270	echo "#PJM --no-stging" >> $job_to_send
	1271	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1272	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1273	echo "#PJM -j" >> $job_to_send
	1274	echo " " >> $job_to_send
	1275	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1276	echo "set -x" >> $job_to_send
[1094]	1277	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1278	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1279	echo "%%END%%" >> $job_to_send
	1280
[1274]	1281	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1282	then
[1289]	1283	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1284	echo "#!/bin/bash" >> $job_to_send
	1285	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1286	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1287	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1288	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1289	echo "#PBS -j oe" >> $job_to_send
	1290	echo " " >> $job_to_send
	1291	echo "set -x" >> $job_to_send
	1292	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1293	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1294	echo "%%END%%" >> $job_to_send
	1295
[1099]	1296	elif [[ $remote_host = lcflow ]]
	1297	then
[1289]	1298	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1299	echo "#!/bin/bash" >> $job_to_send
	1300	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1301	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1302	echo "SGEPREFIX -cwd" >> $job_to_send
	1303	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1304	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
	1305	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
[1099]	1306	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1307	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1308	echo " " >> $job_to_send
	1309	echo "set -x" >> $job_to_send
	1310	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1311	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1312	echo "" >> $job_to_send
	1313	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1314	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1315	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1316	echo "%%END%%" >> $job_to_send
[1289]	1317	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1318	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1099]	1319
[1]	1320	else
	1321
[1289]	1322	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1323	echo "# @\\\$-q $return_queue" >> $job_to_send
	1324	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1325	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1326	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1327	then
	1328	echo "# @\$-l mpp_p=0" >> $job_to_send
	1329	fi
	1330	echo '# @\$-lF 10mb' >> $job_to_send
	1331	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1332	echo '# @\\\$-eo' >> $job_to_send
	1333	echo " " >> $job_to_send
	1334	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1335	then
	1336	echo "set +vx" >> $job_to_send
	1337	echo ". .profile" >> $job_to_send
	1338	fi
	1339	echo "set -x" >> $job_to_send
[1094]	1340	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]	1341	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1342	echo "%%END%%" >> $job_to_send
	1343	fi
	1344
	1345	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1346	then
[1289]	1347	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1348	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1349	then
[1289]	1350	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1351	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1352	then
[1289]	1353	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1354	elif [[ $remote_host = t3es ]]
	1355	then
[1289]	1356	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1357	elif [[ $remote_host = lckiaps ]]
	1358	then
[1289]	1359	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1360	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
	1361	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1362	elif [[ $remote_host = lckyu* ]]
[1090]	1363	then
[1289]	1364	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
	1365	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1366	elif [[ $remote_host = lcflow ]]
	1367	then
[1289]	1368	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1369	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1370	else
[1289]	1371	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1372	fi
[1043]	1373	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1374	then
[1289]	1375	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1376	fi
[1]	1377	if [[ $remote_host = nech ]]
	1378	then
	1379	echo "cd -" >> $job_to_send
	1380	fi
	1381	else
	1382	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1383	# ??? funktioniert das ÃŒberhaupt noch ???
[1]	1384	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1385	fi
	1386	echo "set -x" >> $job_to_send
	1387	echo " ' exit" >> $job_to_send
	1388	echo "set -x" >> $job_to_send
	1389	fi
	1390
	1391
	1392
[1289]	1393	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1394	cat $file_to_send >> $job_to_send
[69]	1395
[1]	1396	if [[ $remote_host = ibm ]]
	1397	then
	1398	echo " " >> $job_to_send
	1399	echo "exit" >> $job_to_send
	1400	fi
[635]	1401
[1289]	1402	# REMOVE JOB-FILE
[1099]	1403	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1404	then
	1405	echo " " >> $job_to_send
	1406	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1407	fi
[1]	1408
	1409
[1289]	1410	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1411	if [[ $no_submit = false ]]
	1412	then
	1413	if [[ $remote_host != $local_host ]]
	1414	then
	1415	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1416	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1417	then
[82]	1418	job_catalog_save=$job_catalog
	1419	job_catalog=job_queue
	1420	elif [[ $remote_host = nech ]]
	1421	then
	1422	job_catalog_save=$job_catalog
	1423	job_catalog=/hpf/b/${remote_user}/job_queue
	1424	fi
[1096]	1425	if [[ $remote_host = nech ]]
[82]	1426	then
[1289]	1427	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1428	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1429	else
[1094]	1430	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1431	fi
[82]	1432	if [[ $? = 1 ]]
	1433	then
	1434	locat=scp; exit
	1435	fi
	1436	if [[ $remote_host = ibms ]]
	1437	then
	1438	job_catalog=$job_catalog_save
	1439	fi
[1]	1440	[[ $verify = true ]] && printf "\n >>> finished\n"
	1441	else
	1442	eval job_catalog=$job_catalog
	1443	cp $job_to_send ${job_catalog}/$job_on_remhost
	1444	fi
	1445
	1446
	1447
[1289]	1448	# START NQS- / LOADLEVELER-JOB
[1]	1449	if [[ $remote_host != $local_host ]]
	1450	then
	1451	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1452
[1289]	1453	if [[ $remote_host = ibmku ]]
[1]	1454	then
[1094]	1455	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1456	elif [[ $remote_host = lcflow ]]
	1457	then
[1202]	1458	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1459	else
[1094]	1460	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1461	fi
[1]	1462
	1463	[[ $verify = true ]] && printf " >>> o.k.\n"
	1464	else
	1465	cd $job_catalog
[1289]	1466	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1467	then
	1468	eval $submcom $job_on_remhost
[1289]	1469	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1470	then
[635]	1471	chmod u+x $job_on_remhost
[108]	1472	eval $submcom $job_on_remhost
[1]	1473	elif [[ $local_host = nech ]]
	1474	then
	1475	if [[ $queue = default ]]
	1476	then
[799]	1477	eval $submcom $job_on_remhost
[1]	1478	else
[799]	1479	eval $submcom -q $queue $job_on_remhost
[1]	1480	fi
	1481	else
	1482	qsub $job_on_remhost
	1483	fi
[622]	1484
[1289]	1485	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1486	# AT THE END OF THE JOB
[1099]	1487	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1488	then
	1489	rm $job_on_remhost
	1490	fi
[1]	1491	cd - > /dev/null
	1492	fi
	1493	fi
	1494
	1495
	1496
[1289]	1497	# FINAL ACTIONS
[1]	1498	if [[ $no_submit = false ]]
	1499	then
[1099]	1500	rm -f $job_to_send
[1]	1501	fi
[1266]	1502	[[ $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 |