Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1344

Last change on this file since 1344 was 1310, checked in by raasch, 11 years ago
update of GPL copyright
Property svn:keywords set to `Id Rev`
File size: 50.2 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	# ------------------
[1290]	25	#
[1046]	26	#
	27	# Former revisions:
	28	# -----------------
[169]	29	# $Id: subjob 1310 2014-03-14 08:01:56Z kanani $
[1200]	30	#
[1290]	31	# 1289 2014-03-04 07:12:34Z raasch
	32	# German comments translated to English
	33	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
	34	#
[1280]	35	# 1279 2014-01-28 12:10:14Z raasch
	36	# node calculation modified due to changes in mrun (tasks_per_node must not be
	37	# an integral divisor of numprocs any more)
	38	#
[1275]	39	# 1274 2014-01-09 13:14:54Z heinze
	40	# adjustments for lccrayh
	41	#
[1267]	42	# 1266 2013-12-11 12:07:34Z heinze
	43	# further adjustments for lccrayb (use msub instead of qsub)
	44	#
[1265]	45	# 1264 2013-12-09 12:46:09Z fricke
	46	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	47	#
[1263]	48	# 1262 2013-12-09 10:57:20Z fricke
	49	# further adjustments for lccrayb
	50	#
[1261]	51	# 1260 2013-12-04 12:48:04Z raasch
	52	# jaboticaba admitted
	53	#
[1256]	54	# 1255 2013-11-07 14:43:35Z raasch
	55	# further adjustments for lccrayb
	56	#
[1225]	57	# 1224 2013-09-16 07:27:23Z raasch
	58	# first adjustments for lccrayb
	59	#
[1203]	60	# 1202 2013-07-10 16:22:07Z witha
	61	# adjustments for Forwind cluster (lcflow)
	62	#
[1200]	63	# 1199 2013-07-05 14:52:22Z raasch
	64	# adjustments for CSC Helsinki (lccrayf)
	65	#
[1185]	66	# use of cluster/express queue enabled (ibmh)
	67	# vinessa added (imuk)
[1047]	68	#
[1104]	69	# 1103 2013-02-20 02:15:53Z raasch
	70	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	71	# instead typeset -L/R),
	72	# further adjustments for lckyuh
	73	#
[1100]	74	# 2013-02-10 01:47:43Z raasch
	75	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	76	# and for Forwind cluster (lcflow)
	77	#
[1096]	78	# 1094 2013-02-03 01:52:12Z raasch
	79	# new option -P for explicit setting of ssh/scp port,
	80	# decalpha parts (yonsei) removed
	81	#
[1091]	82	# 2013-02-02 07:06:13Z raasch
[1099]	83	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	84	# old changelog messages removed
	85	#
[1047]	86	# 1046 2012-11-09 14:38:45Z maronga
	87	# code put under GPL (PALM 3.9)
	88	#
[1090]	89	# 08/07/94 - Siggi - first version finished
	90	# 29/06/94 - Siggi - script development started
	91	#--------------------------------------------------------------------------------#
	92	# subjob - script for automatic generation and submission of batch-job files
	93	# for various batch queuing systems
	94	#--------------------------------------------------------------------------------#
[1]	95
	96
[1289]	97	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	98	delete_dayfile=false
[799]	99	email_notification=none
[122]	100	group_number=none
[1]	101	locat=normal
	102	no_default_queue=none
	103	no_submit=false
	104	job_catalog="~/job_queue"
	105	job_name=none
	106	local_user=$LOGNAME
	107	node_usage=shared
[475]	108	numprocs=0
[1]	109	punkte="..........................................................."
	110	submcom=qsub
	111	queue=default
	112	remote_host=none
	113	remote_user=""
	114	verify=true
	115
	116	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	117	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	118
	119
	120
[1289]	121	# ERROR HANDLING
	122	# IN CASE OF EXIT:
[1]	123	trap 'if [[ $locat != normal ]]
	124	then
	125	case $locat in
	126	(option) printf "\n --> available optios can be displayed"
	127	printf " by typing:"
	128	printf "\n \"subjob ?\" \n";;
	129	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	130	(*) printf "\n +++ unknown error"
	131	printf "\n please inform S. Raasch!\n"
	132	esac
	133	[[ -f $job_to_send ]] && rm $job_to_send
	134	printf "\n\n+++ SUBJOB killed \n\n"
	135	fi' exit
	136
	137
[1289]	138	# IN CASE OF TERMINAL-BREAK:
[1]	139	trap '[[ -f $job_to_send ]] && rm $job_to_send
	140	printf "\n\n+++ SUBJOB killed \n\n"
	141	exit
	142	' 2
	143
	144
	145
	146
[1289]	147	# DETERMINE NAME OF LOCAL HOST
[1]	148	local_host=$(hostname)
	149
	150
	151
[1289]	152	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	153	# IS ADMITTED AT ALL)
	154	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	155	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	156	case $local_host in
[1204]	157	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
[1]	158	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[1235]	159	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
[102]	160	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	161	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1103]	162	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
[749]	163	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[1255]	164	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
[1274]	165	(hlogin\|hxc) local_addres=130.75.4.1; local_host=lccrayh;;
[12]	166	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[1174]	167	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
[372]	168	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	169	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[1197]	170	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
[1]	171	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	172	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[1099]	173	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
	174	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
[367]	175	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	176	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	177	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	178	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[5]	179	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	180	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1099]	181	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
[892]	182	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[440]	183	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	184	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	185	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[1260]	186	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
[407]	187	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	188	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
[1040]	189	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
[5]	190	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1099]	191	(meller) local_addres=134.106.74.155; local_host=lcfor;;
[437]	192	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	193	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	194	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	195	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	196	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	197	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[1099]	198	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
[789]	199	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	200	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	201	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1204]	202	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
[1197]	203	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
[799]	204	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1090]	205	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
[635]	206	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1021]	207	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
[1184]	208	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
[1]	209	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	210	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	211	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	212	printf "\n please contact the PALM group at IMUK";
[1]	213	locat=parameter; exit;;
	214	esac
	215
	216
	217
[1289]	218	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	219	remote_host=$local_host
	220
	221
	222
	223
[1289]	224	# READ THE SHELLSCRIPT-OPTIONS
[1094]	225	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	226	do
	227	case $option in
	228	(c) job_catalog=$OPTARG;;
	229	(d) delete_dayfile=true;;
	230	(D) no_submit=true;;
[352]	231	(e) email_notification=$OPTARG;;
[125]	232	(g) group_number=$OPTARG;;
[1]	233	(h) remote_host=$OPTARG;;
	234	(m) memory=$OPTARG;;
	235	(n) job_name=$OPTARG;;
	236	(N) node_usage=$OPTARG;;
	237	(O) threads_per_task=$OPTARG;;
[1094]	238	(P) scp_port=$OPTARG;;
[1]	239	(q) no_default_queue=$OPTARG;;
	240	(t) cputime=$OPTARG;;
	241	(T) tasks_per_node=$OPTARG;;
	242	(u) remote_user=$OPTARG;;
	243	(v) verify=false;;
	244	(X) numprocs=$OPTARG;;
	245	(\?) printf "\n +++ Option $OPTARG unknown \n";
	246	locat=option; exit;;
	247	esac
	248	done
	249
	250
[1289]	251	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	252	(( to_shift = $OPTIND - 1 ))
	253	shift $to_shift; file_to_send=$1
[1]	254
	255
[1289]	256	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	257	if [ "$1" = "?" ]
	258	then
	259	(printf "\n *** subjob can be called as follows:\n"
	260	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	261	printf "\n Description of available options:\n"
	262	printf "\n Option Description Default-Value"
	263	printf "\n -c job-input- and output-catalog ~/job_queue"
	264	printf "\n -d no job-protocol will be created ---"
	265	printf "\n -D only the job-file will be created ---"
	266	printf "\n -h execution host, available hosts: $remote_host"
[1289]	267	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	268	printf "\n lckiaps, lctit, nech"
[1]	269	printf "\n -m memory demand per process in MByte ---"
	270	printf "\n -n jobname <jobdatei>"
	271	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	272	printf "\n -P ssh/scp port default port"
[1]	273	printf "\n -q job-queue to be used default"
	274	printf "\n -t allowed cpu-time in seconds ---"
	275	printf "\n -T tasks per node (on parallel hosts) ---"
	276	printf "\n -u username on execution host from .netrc"
	277	printf "\n -v no prompt for confirmation ---"
	278	printf "\n -X # of processors (on parallel hosts) 1"
	279	printf "\n "
	280	printf "\n The only possible positional parameter is <jobfile>:"
	281	printf "\n The complete NQS-job must be provided here."
	282	printf "\n <jobfile>=? creates this outline\n\n") \| more
	283	exit
	284	fi
	285
	286
	287
[1289]	288	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	289	if [[ "$file_to_send" = "" ]]
	290	then
	291	printf "\n +++ job-file missing"
	292	locat=parameter; exit
	293	else
	294	if [[ -f $file_to_send ]]
	295	then
	296	true
	297	else
	298	printf "\n +++ job-file: "
	299	printf "\n $file_to_send"
	300	printf "\n does not exist"
	301	locat=parameter; exit
	302	fi
	303	fi
	304
	305
	306
[1289]	307	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	308	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	309	if [[ $job_name = none ]]
	310	then
	311	job_name=$file_to_send
	312	fi
	313	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	314	then
	315	printf "\n +++ job-file name: "
	316	printf "\n $job_name"
	317	printf "\n must not contain \"/\"-characters"
	318	locat=parameter; exit
	319	fi
	320
	321
	322
	323
[1289]	324	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	325	# OR IF NO HOST HAS BEEN GIVEN
[1]	326	if [[ $remote_host = none ]]
	327	then
	328	printf "\n +++ host missing"
	329	locat=option; exit
	330	else
	331	case $remote_host in
[251]	332	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1184]	333	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	334	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	335	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	336	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1262]	337	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1274]	338	(lccrayh) queue=mpp1testq; remote_addres=130.75.4.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1197]	339	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
[1099]	340	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
[440]	341	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[367]	342	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[1040]	343	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
[717]	344	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[1099]	345	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
[1090]	346	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
[1021]	347	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
[635]	348	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[892]	349	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
[437]	350	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	351	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	352	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	353	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	354	locat=parameter; exit;;
[1]	355	esac
	356	fi
	357
	358
[1289]	359	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	360	if [[ $no_default_queue != none ]]
	361	then
	362	error=false
	363	ndq=$no_default_queue
	364	case $remote_host in
	365	(ibm) case $ndq in
	366	(p690_express\|p690_standard\|p690_long) error=false;;
	367	(*) error=true;;
	368	esac;;
	369	(ibmh) case $ndq in
[1184]	370	(cluster\|express) error=false;;
[1]	371	(*) error=true;;
	372	esac;;
[693]	373	(ibmkisti) case $ndq in
	374	(class.32plus\|class.1-2\|class.2-32) error=false;;
	375	(*) error=true;;
	376	esac;;
[622]	377	(ibmku) case $ndq in
	378	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	379	(*) error=true;;
	380	esac;;
[1]	381	(ibms) case $ndq in
	382	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	383	(*) error=true;;
	384	esac;;
[1224]	385	(lccrayb) case $ndq in
[1274]	386	(mpp1q\|testq\|specialm1q) error=false;;
[1224]	387	(*) error=true;;
	388	esac;;
[1274]	389	(lccrayh) case $ndq in
	390	(mpp1q\|mpp1testq\|specialm1q\|dataq) error=false;;
	391	(*) error=true;;
	392	esac;;
[1197]	393	(lccrayf) case $ndq in
	394	(usup\|test*\|small\|large) error=false;;
	395	(*) error=true;;
	396	esac;;
[1099]	397	(lcflow) case $ndq in
	398	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
	399	(*) error=true;;
	400	esac;;
[1040]	401	(lckiaps) case $ndq in
	402	(express\|normal) error=false;;
	403	(*) error=true;;
	404	esac;;
[440]	405	(lckyoto) case $ndq in
	406	(eh\|ph) error=false;;
	407	(*) error=true;;
	408	esac;;
[1099]	409	(lckyuh) case $ndq in
	410	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	411	(*) error=true;;
	412	esac;;
[1090]	413	(lckyut) case $ndq in
	414	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	415	(*) error=true;;
	416	esac;;
[1]	417	(lctit) case $ndq in
[635]	418	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	419	(*) error=true;;
	420	esac;;
	421	(t3eb) case $ndq in
	422	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	423	(*) error=true;;
	424	esac;;
	425	(t3eh) case $ndq in
	426	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	427	(*) error=true;;
	428	esac;;
	429	(t3ej2\|t3ej5) case $ndq in
	430	(low\|normal\|high) error=false;;
	431	(*) error=true;;
	432	esac;;
	433	(t3es) case $ndq in
	434	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	435	(*) error=true;;
	436	esac;;
	437	esac
	438	if [[ $error = true ]]
	439	then
	440	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	441	locat=parameter; exit
	442	else
	443	queue=$no_default_queue
	444	fi
	445	fi
	446
	447
	448
[1289]	449	# CHECK THE CPU-TIME
	450	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	451	done=false
	452	while [[ $done = false ]]
	453	do
	454	if (( $cputime <= 0 ))
	455	then
	456	printf "\n +++ wrong cpu-time or cpu-time missing"
	457	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	458	printf "\n >>> "
	459	read cputime 1>/dev/null 2>&1
	460	else
	461	done=true
	462	fi
	463	done
	464	if [[ $remote_host = nech ]]
	465	then
	466	if (( tasks_per_node != 0 ))
	467	then
	468	(( cputime = cputime * tasks_per_node ))
	469	elif [[ $numprocs != 0 ]]
	470	then
	471	(( cputime = cputime * numprocs ))
	472	fi
	473	fi
	474	(( stunden = cputime / 3600 ))
	475	(( resttime = cputime - stunden * 3600 ))
	476	(( minuten = resttime / 60 ))
	477	(( sekunden = resttime - minuten * 60 ))
	478	timestring=${stunden}:${minuten}:${sekunden}
	479
	480
	481
[1289]	482	# CHECK THE MEMORY DEMAND
[1]	483	done=false
	484	while [[ $done = false ]]
	485	do
	486	if (( memory <= 0 ))
	487	then
	488	printf "\n +++ wrong memory demand or memory demand missing"
	489	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	490	printf "\n >>> "
	491	read memory 1>/dev/null 2>&1
	492	else
	493	done=true
	494	fi
	495	done
	496
[1289]	497	if [[ $remote_host = nech ]]
[1]	498	then
	499	if (( tasks_per_node != 0 ))
	500	then
	501	(( Memory = memory * tasks_per_node / 1000 ))
	502	elif [[ $numprocs != 0 ]]
	503	then
	504	(( Memory = memory * numprocs / 1000 ))
	505	else
	506	(( Memory = memory / 1000 ))
	507	fi
[635]	508	elif [[ $remote_host = lctit ]]
	509	then
	510	(( Memory = memory * tasks_per_node / 1000 ))
[1]	511	fi
	512
	513
[1289]	514	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	515	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	516	then
	517	(( memory = memory * threads_per_task ))
	518	fi
	519
	520
[1289]	521	# CALCULATE NUMBER OF REQUIRED NODES
[1]	522	if (( tasks_per_node != 0 ))
	523	then
[1279]	524	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	525	fi
	526
[1094]	527
[1289]	528	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	529	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	530
[1094]	531
[1289]	532	# CALCULATE NUMBER OF MPI TASKS
[696]	533	(( mpi_tasks = numprocs / threads_per_task ))
[1]	534
[696]	535
[1289]	536	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	537	if [[ "$scp_port" != "" ]]
	538	then
	539	PORTOPT="-P $scp_port"
	540	SSH_PORTOPT="-p $scp_port"
	541	fi
	542
	543
[1289]	544	# HEADER-OUTPUT
[1]	545	if [[ $verify = true ]]
	546	then
	547	printf "\n\n"
	548	printf "#--------------------------------------------------------------# \n"
	549	spalte1=SUBJOB;spalte2=$(date)
[1103]	550	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	551	printf "\| \| \n"
	552	printf "\| values of parameters/options: \| \n"
[1103]	553	spalte1=$(echo local_host$punkte \| cut -c-20)
	554	spalte2=$punkte$local_host
	555	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	556	spalte1=$(echo remote_host$punkte \| cut -c-20)
	557	spalte2=$punkte$remote_host
	558	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	559	spalte1=$(echo queue$punkte \| cut -c-20)
	560	spalte2=$punkte$queue
	561	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	562	spalte1=$(echo memory$punkte \| cut -c-20)
	563	spalte2="$punkte$memory mb"
	564	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	565	spalte1=$(echo cputime$punkte \| cut -c-20)
	566	spalte2="$punkte$cputime sec"
	567	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	568	spalte1=$(echo job_name$punkte \| cut -c-20)
	569	spalte2="$punkte$job_name"
	570	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	571	printf "#--------------------------------------------------------------# \n\n"
	572
	573
[1289]	574	# QUERY CHECK
[1]	575	antwort="dummy"
	576	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	577	do
	578	read antwort?" >>> continue (y/n) ? "
	579	done
	580	if [[ $antwort = n \|\| $antwort = N ]]
	581	then
	582	locat=verify; exit
	583	fi
	584	printf "\n"
	585	fi
	586
	587
	588
[1289]	589	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	590	identifier=$RANDOM
	591	job_on_remhost=${job_name}_${identifier}_$local_host
	592	job_to_send=job_to_send_$identifier
[1]	593	if [[ $delete_dayfile = false ]]
	594	then
[1289]	595	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	596	local_dayfile=${remote_host}_${job_name}
	597	else
	598	remote_dayfile=/dev/null
	599	fi
	600
	601
	602
[1289]	603	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	604	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	605	then
	606
[1289]	607	# GENERAL LOADLEVELER SETTINGS
[622]	608	execute_in_shell="#!/bin/ksh"
	609	use_shell="# @ shell = /bin/ksh"
	610	consumable_memory="ConsumableMemory($memory mb)"
	611	class="# @ class = $queue"
	612	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	613	network_to_use="# @ network.mpi = sn_all,shared,us"
	614	data_limit="# @ data_limit = 1.76gb"
	615	image_size="# @ image_size = 50"
[693]	616	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	617
[693]	618	if [[ $email_notification = none ]]
	619	then
	620	notify_user=""
	621	else
	622	notify_user="# @ notify_user = $email_notification"
	623	if [[ $delete_dayfile = true ]]
	624	then
	625	notification='# @ notification = never'
	626	fi
	627	fi
[622]	628
[312]	629	if [[ $remote_host = ibmh ]]
[1]	630	then
[312]	631	data_limit=""
	632	network_to_use=""
[1184]	633	class="# @ class = $queue"
[312]	634	environment=""
[814]	635	rset="# @ rset = RSET_MCM_AFFINITY"
	636	task_affinity="# @ task_affinity = core(1)"
[693]	637	elif [[ $remote_host = ibmkisti ]]
	638	then
	639	network_to_use="# @ network.MPI = sn_all,shared,US"
	640	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	641	if [[ $threads_per_task = 1 ]]
	642	then
	643	rset="# @ rset = RSET_MCM_AFFINITY"
	644	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	645	fi
[693]	646	environment=""
	647	use_shell=""
	648	data_limit=""
	649	image_size=""
[622]	650	elif [[ $remote_host = ibmku ]]
	651	then
	652	execute_in_shell="#!/usr/bin/ksh"
	653	use_shell="# @ shell = /usr/bin/ksh"
	654	consumable_memory=""
	655	environment=""
	656	network_to_use="# @ network.mpi = sn_all,shared,us"
	657	data_limit=""
	658	image_size=""
	659	elif [[ $remote_host = ibms ]]
	660	then
	661	network_to_use="# @ network.mpi = csss,shared,us"
[1]	662	fi
	663
	664	cat > $job_to_send << %%END%%
[622]	665	$execute_in_shell
	666	$use_shell
[1]	667
	668	# @ job_type = parallel
[693]	669	# @ job_name = $job_name
[1]	670	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	671	# @ output = $remote_dayfile
	672	# @ error = $remote_dayfile
[693]	673	$wall_clock_limit
[622]	674	$image_size
[312]	675	$class
	676	$environment
[1]	677	$network_to_use
	678	$data_limit
[693]	679	$rset
	680	$mcm_affinity_options
[814]	681	$task_affinity
[1]	682	$notification
[693]	683	$notify_user
[1]	684
	685	%%END%%
	686
	687	if (( nodes > 0 ))
	688	then
	689
[693]	690	if [[ $remote_host != ibmkisti ]]
	691	then
	692
	693	cat >> $job_to_send << %%END%%
[1]	694	# @ node = $nodes
[622]	695	# @ tasks_per_node = $processes_per_node
[1]	696	# @ node_usage = $node_usage
	697	# @ queue
	698
	699	%%END%%
	700
[693]	701	else
	702
	703	cat >> $job_to_send << %%END%%
[696]	704	# @ total_tasks = $mpi_tasks
[693]	705	# @ blocking = unlimited
	706	# @ queue
	707
	708	%%END%%
	709
	710	fi
	711
[1]	712	else
	713
[1289]	714	cat >> $job_to_send << %%END%%
[1]	715	# @ blocking = unlimited
	716	# @ total_tasks = $numprocs
	717	# @ node_usage = $node_usage
	718	# @ queue
	719
	720	%%END%%
	721
	722	fi
	723
[1289]	724	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	725	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	726	then
	727	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	728	fi
	729
[1]	730	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	731	then
	732
	733	cat > $job_to_send << %%END%%
	734	#!/bin/ksh
	735
	736	# @ job_type = serial
	737	# @ node_usage = $node_usage
	738	# @ job_name = palm
	739	# @ wall_clock_limit = ${timestring},$timestring
	740	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	741	# @ output = $remote_dayfile
	742	# @ error = $remote_dayfile
[312]	743	$class
[1]	744	$notification
	745
	746	# @ queue
	747
	748	%%END%%
	749
[1274]	750	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	751	then
	752
	753	if [[ $numprocs != 0 ]]
	754	then
	755	cat > $job_to_send << %%END%%
[1255]	756	#!/bin/bash
[1224]	757	#PBS -N $job_name
	758	#PBS -l walltime=$timestring
[1264]	759	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	760	#PBS -o $remote_dayfile
	761	#PBS -j oe
	762	#PBS -q $queue
	763
	764	$module_calls
	765
	766	%%END%%
	767
	768	else
	769
	770	continue
	771
	772	fi
	773
[1197]	774	elif [[ $remote_host = lccrayf ]]
	775	then
	776
	777	if [[ $numprocs != 0 ]]
	778	then
	779	cat > $job_to_send << %%END%%
	780	#!/bin/bash -l
	781	#SBATCH -J $job_name
	782	#SBATCH -t $timestring
	783	#SBATCH -N $nodes
	784	#SBATCH --ntasks-per-node=$processes_per_node
	785	#SBATCH -p $queue
	786	#SBATCH -o $remote_dayfile
	787	#SBATCH -e $remote_dayfile
	788
	789	$init_cmds
	790	$module_calls
	791
	792	%%END%%
	793
	794	else
	795	cat > $job_to_send << %%END%%
	796	#!/bin/bash -l
	797	#SBATCH -J $job_name
	798	#SBATCH -t $timestring
	799	#SBATCH -l ncpus=1
	800	#SBATCH -l pmem=${memory}mb
	801	#SBATCH -m abe
	802	#SBATCH -o $remote_dayfile
	803	#SBATCH -e $remote_dayfile
	804
	805	$init_cmds
	806	$module_calls
	807
	808	%%END%%
	809
	810	fi
	811
[1099]	812	elif [[ $remote_host = lcflow ]]
	813	then
[1202]	814	if [ $memory -gt 1800 ]; then
[1099]	815	use_himem=""
	816	else
	817	use_himem="#"
	818	fi
	819	if [[ $numprocs != 0 ]]
	820	then
[1202]	821	pe_set="#$ -pe impi41 $numprocs"
[1099]	822	else
[1202]	823	pe_set="#$ -pe impi41 1"
[1099]	824	fi
	825	if [[ $queue = default ]]
	826	then
	827	queue_set=""
	828	else
	829	queue_set="#$ -q $queue"
	830	fi
	831	[[ "$disc_space" = "" ]] && disc_space=50
	832
	833	cat > $job_to_send << %%END%%
	834	#!/bin/bash
	835	#$ -S /bin/bash
	836	#$ -N $job_name
	837	#$ -cwd
	838	#$ -l h_rt=$timestring
	839	#$ -l h_vmem=${memory}M
	840	#$ -o $remote_dayfile
	841	#$ -j y
	842	$pe_set
	843	#$ -R y
	844	${use_himem}#$ -l highmem=true
	845	#$ -l h_fsize=${disc_space}G
	846	$queue_set
	847
	848	%%END%%
	849
[1021]	850	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	851	then
	852
	853	if [[ $numprocs != 0 ]]
	854	then
	855	cat > $job_to_send << %%END%%
	856	#!/bin/ksh
	857	#PBS -N $job_name
	858	#PBS -l walltime=$timestring
	859	#PBS -l ncpus=$numprocs
	860	#PBS -l pmem=${memory}mb
	861	#PBS -o $remote_dayfile
[1021]	862	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	863	#PBS -j oe
	864
	865	mpd &
	866
	867	%%END%%
	868
	869	else
	870	cat > $job_to_send << %%END%%
	871	#!/bin/ksh
	872	#PBS -N $job_name
	873	#PBS -l walltime=$timestring
	874	#PBS -l ncpus=1
	875	#PBS -l pmem=${memory}mb
	876	#PBS -o $remote_dayfile
	877	#PBS -j oe
	878
	879	%%END%%
	880
	881	fi
	882
[1040]	883	elif [[ $remote_host = lckiaps ]]
	884	then
	885
	886	if [[ $numprocs != 0 ]]
	887	then
	888	cat > $job_to_send << %%END%%
	889	#!/bin/ksh
	890	#PBS -N $job_name
	891	#PBS -l walltime=$timestring
	892	#PBS -l select=1:ncpus=$numprocs
	893	#PBS -l pmem=${memory}mb
	894	#PBS -q $queue
	895	#PBS -o $remote_dayfile
	896	#PBS -j oe
	897	#PBS -V
	898
	899	%%END%%
	900
	901	else
	902	cat > $job_to_send << %%END%%
	903	#!/bin/ksh
	904	#PBS -N $job_name
	905	#PBS -l walltime=$timestring
	906	#PBS -l ncpus=1
	907	#PBS -l pmem=${memory}mb
	908	#PBS -o $remote_dayfile
	909	#PBS -j oe
	910
	911	%%END%%
	912
	913	fi
	914
[693]	915	elif [[ $remote_host = lcyon ]]
	916	then
	917
	918	if [[ $numprocs != 0 ]]
	919	then
	920	cat > $job_to_send << %%END%%
	921	#!/bin/ksh
	922	#PBS -N $job_name
	923	#PBS -l walltime=$timestring
	924	#PBS -l ncpus=$numprocs
	925	#PBS -l pmem=${memory}mb
	926	#PBS -o $remote_dayfile
	927	#PBS -j oe
	928
	929	%%END%%
	930
	931	else
	932	cat > $job_to_send << %%END%%
	933	#!/bin/ksh
	934	#PBS -N $job_name
	935	#PBS -l walltime=$timestring
	936	#PBS -l ncpus=1
	937	#PBS -l pmem=${memory}mb
	938	#PBS -o $remote_dayfile
	939	#PBS -j oe
	940
	941	%%END%%
	942
	943	fi
	944
[892]	945	elif [[ $remote_host = lcxe6 ]]
[164]	946	then
	947
	948	if [[ $numprocs != 0 ]]
	949	then
	950	cat > $job_to_send << %%END%%
[799]	951	#!/bin/ksh
[164]	952	#PBS -S /bin/ksh
	953	#PBS -N $job_name
[552]	954	#PBS -A $project_account
[206]	955	#PBS -j oe
[164]	956	#PBS -l walltime=$timestring
	957	#PBS -l mppwidth=${numprocs}
[622]	958	#PBS -l mppnppn=${processes_per_node}
[164]	959	#PBS -m abe
	960	#PBS -o $remote_dayfile
[492]	961	$email_directive
[164]	962
[892]	963	$init_cmds
[493]	964	$module_calls
[343]	965
[164]	966	%%END%%
	967
[108]	968	else
	969	cat > $job_to_send << %%END%%
	970	#!/bin/ksh
[168]	971	#PBS -S /bin/ksh
[108]	972	#PBS -N $job_name
[552]	973	#PBS -A $project_account
[206]	974	#PBS -j oe
[108]	975	#PBS -l walltime=$timestring
	976	#PBS -l ncpus=1
	977	#PBS -l pmem=${memory}mb
	978	#PBS -m abe
[492]	979	$email_directive
[108]	980	#PBS -o $remote_dayfile
	981
[892]	982	$init_cmds
[493]	983	$module_calls
[343]	984
[108]	985	%%END%%
	986
	987	fi
	988
[440]	989	elif [[ $remote_host = lckyoto ]]
[437]	990	then
	991
[440]	992	cat > $job_to_send << %%END%%
[799]	993	#!/bin/ksh
[440]	994	# @\$-o $remote_dayfile
	995	# @\$-eo -oi
	996	# @\$-lP 16
[799]	997	# @\$-lp 1
[440]	998	# @\$-lm 28gb -llm unlimited -ls unlimited
	999	# @\$-q $queue
	1000	# @\$-Pvn abs_pack
	1001	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1002	#. /thin/local/etc/setprofile/intel-11.0.sh
	1003	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1004	. ~/.myprofile
	1005	#. /home2/t/t51254/palm/current_version/myprofile
	1006	#. /thin/apps/pgi/mpi.sh
	1007	#
	1008	env
	1009	#
	1010	set -x
	1011
	1012	%%END%%
	1013
	1014	elif [[ $remote_host = lcxt5m ]]
	1015	then
	1016
[437]	1017	if [[ $numprocs != 0 ]]
	1018	then
	1019	cat > $job_to_send << %%END%%
[799]	1020	#!/bin/ksh
[437]	1021	#PBS -S /bin/ksh
	1022	#PBS -N $job_name
	1023	#PBS -j oe
	1024	#PBS -l walltime=$timestring
	1025	#PBS -l mppwidth=${numprocs}
[622]	1026	#PBS -l mppnppn=${processes_per_node}
[437]	1027	#PBS -m abe
	1028	#PBS -o $remote_dayfile
	1029
[892]	1030	$init_cmds
[493]	1031	$module_calls
[437]	1032
	1033	%%END%%
	1034
	1035	else
	1036	cat > $job_to_send << %%END%%
	1037	#!/bin/ksh
	1038	#PBS -S /bin/ksh
	1039	#PBS -N $job_name
	1040	#PBS -j oe
	1041	#PBS -l walltime=$timestring
	1042	#PBS -l ncpus=1
	1043	#PBS -l pmem=${memory}mb
	1044	#PBS -m abe
	1045	#PBS -o $remote_dayfile
	1046
[892]	1047	$init_cmds
[493]	1048	$module_calls
[437]	1049
	1050	%%END%%
	1051
	1052	fi
	1053
[1099]	1054	elif [[ $remote_host = lckyuh ]]
	1055	then
	1056	cat > $job_to_send << %%END%%
	1057	#!/bin/bash
	1058	#PJM -L "rscgrp=$queue"
	1059	#PJM -L "node=$nodes"
	1060	#PJM --mpi "proc=$numprocs"
	1061	#PJM -L "elapse=$timestring"
	1062	#PJM -o $remote_dayfile
	1063	#PJM -j
	1064	#PJM -X
	1065	#PJM --no-stging
	1066
	1067	export LANG=en_US.UTF-8
	1068	%%END%%
	1069
[1090]	1070	elif [[ $remote_host = lckyut ]]
	1071	then
	1072	cat > $job_to_send << %%END%%
	1073	#!/bin/bash
[1099]	1074	#PJM -L "rscgrp=$queue"
[1090]	1075	#PJM -L "vnode=$numprocs"
	1076	#PJM -L "vnode-core=1"
	1077	#PJM -L "elapse=$timestring"
	1078	#PJM --mpi proc=$numprocs
	1079	#PJM -o $remote_dayfile
	1080	#PJM -j
[1099]	1081	#PJM -X
	1082	#PJM --no-stging
[1090]	1083
	1084	export LANG=en_US.UTF-8
	1085	%%END%%
	1086
[1]	1087	elif [[ $remote_host = nech ]]
	1088	then
	1089
	1090	if (( nodes > 1 ))
	1091	then
	1092	cat > $job_to_send << %%END%%
	1093	#!/bin/ksh
[622]	1094	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1095	#PBS -l ${qsubmem}=${Memory}gb
	1096	#PBS -b $nodes
	1097	#PBS -o $remote_dayfile
	1098	#PBS -N palm
	1099	#PBS -j o
	1100	#PBS -T mpisx
	1101
	1102	%%END%%
	1103
	1104	elif [[ $numprocs != 0 ]]
	1105	then
	1106	cat > $job_to_send << %%END%%
	1107	#!/bin/ksh
[622]	1108	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1109	#PBS -l ${qsubmem}=${Memory}gb
	1110	#PBS -o $remote_dayfile
	1111	#PBS -N palm
	1112	#PBS -j o
	1113
	1114	%%END%%
	1115
	1116	else
	1117	cat > $job_to_send << %%END%%
	1118	#!/bin/ksh
	1119	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1120	#PBS -o $remote_dayfile
	1121	#PBS -j o
	1122
	1123	%%END%%
	1124
	1125	fi
	1126
	1127	elif [[ $remote_host = lctit ]]
	1128	then
	1129	cat > $job_to_send << %%END%%
[635]	1130	#!/bin/ksh
[892]	1131	$init_cmds
[678]	1132	$module_calls
	1133
[1]	1134	%%END%%
	1135
[1289]	1136	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1137	if [[ $tasks_per_node != $processes_per_node ]]
	1138	then
	1139	submcom="$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 $queue "
	1140	else
	1141	submcom="$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 "
	1142	fi
[1]	1143
	1144	else
	1145
	1146	cat > $job_to_send << %%END%%
	1147	# @\$-q ${queue}
	1148	# @\$-l${qsubtime} $timestring
	1149	# @\$-l${qsubmem} ${memory}mb
	1150	# @\$-o $remote_dayfile
	1151	# @\$-eo
	1152
	1153	%%END%%
	1154
	1155	fi
	1156
	1157
[1289]	1158	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1159	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1160	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1161	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1162	then
	1163	echo "set +vx" >> $job_to_send
	1164	echo "trap '" >> $job_to_send
	1165	echo "set +vx" >> $job_to_send
[1289]	1166	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]	1167	then
[622]	1168	if [[ $remote_host = ibmh ]]
[1]	1169	then
	1170	return_queue=c1
[693]	1171	elif [[ $remote_host = ibmkisti ]]
	1172	then
	1173	return_queue=class.1-2
[622]	1174	elif [[ $remote_host = ibmku ]]
	1175	then
	1176	return_queue=sdbg2
[1]	1177	elif [[ $remote_host = ibms ]]
	1178	then
	1179	return_queue=p_normal
[1274]	1180	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1181	then
	1182	return_queue=dataq
[1040]	1183	elif [[ $remote_host = lckiaps ]]
	1184	then
	1185	return_queue=express
[1099]	1186	elif [[ $remote_host = lckyuh ]]
	1187	then
	1188	return_queue=cx-single
[1090]	1189	elif [[ $remote_host = lckyut ]]
	1190	then
	1191	return_queue=cx-single
[1]	1192	else
	1193	return_queue=unknown
	1194	fi
	1195
	1196	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1197	then
	1198
[622]	1199	if [[ $remote_host = ibmku ]]
	1200	then
[1289]	1201	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1202	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1203	else
[1289]	1204	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1205	fi
[1289]	1206	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1207	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1208	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1209	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1210	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1211	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1212	if [[ $host != "ibmh" ]]
	1213	then
[1289]	1214	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1215	fi
[1289]	1216	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1217	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1218
[1289]	1219	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1220	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1221
[1289]	1222	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
	1223	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]	1224	if [[ $remote_host = ibmku ]]
	1225	then
[1289]	1226	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1227	fi
[1289]	1228	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1229
	1230	elif [[ $remote_host = nech ]]
	1231	then
	1232	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1233	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1234	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1235	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1236	echo "#PBS -j o" >> $job_to_send
	1237	echo " " >> $job_to_send
	1238	echo "set -x" >> $job_to_send
	1239	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1094]	1240	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1241	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1242	echo "%%END%%" >> $job_to_send
	1243
[1099]	1244	elif [[ $remote_host = lckyuh ]]
	1245	then
[1289]	1246	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1247	echo "#!/bin/bash" >> $job_to_send
	1248	echo "#PJM -L \"node=1\"" >> $job_to_send
	1249	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1250	echo "#PJM --no-stging" >> $job_to_send
	1251	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1252	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1253	echo "#PJM -j" >> $job_to_send
	1254	echo " " >> $job_to_send
	1255	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1256	echo "set -x" >> $job_to_send
	1257	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1258	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1259	echo "%%END%%" >> $job_to_send
	1260
[1090]	1261	elif [[ $remote_host = lckyut ]]
	1262	then
[1289]	1263	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1264	echo "#!/bin/bash" >> $job_to_send
	1265	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1266	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1267	echo "#PJM --no-stging" >> $job_to_send
	1268	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1269	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1270	echo "#PJM -j" >> $job_to_send
	1271	echo " " >> $job_to_send
	1272	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1273	echo "set -x" >> $job_to_send
[1094]	1274	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1275	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1276	echo "%%END%%" >> $job_to_send
	1277
[1274]	1278	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1279	then
[1289]	1280	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1281	echo "#!/bin/bash" >> $job_to_send
	1282	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1283	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1284	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1285	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1286	echo "#PBS -j oe" >> $job_to_send
	1287	echo " " >> $job_to_send
	1288	echo "set -x" >> $job_to_send
	1289	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1290	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1291	echo "%%END%%" >> $job_to_send
	1292
[1099]	1293	elif [[ $remote_host = lcflow ]]
	1294	then
[1289]	1295	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1296	echo "#!/bin/bash" >> $job_to_send
	1297	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1298	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1299	echo "SGEPREFIX -cwd" >> $job_to_send
	1300	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1301	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
	1302	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
[1099]	1303	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1304	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1305	echo " " >> $job_to_send
	1306	echo "set -x" >> $job_to_send
	1307	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1308	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1309	echo "" >> $job_to_send
	1310	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1311	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1312	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1313	echo "%%END%%" >> $job_to_send
[1289]	1314	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1315	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1099]	1316
[1]	1317	else
	1318
[1289]	1319	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1320	echo "# @\\\$-q $return_queue" >> $job_to_send
	1321	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1322	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1323	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1324	then
	1325	echo "# @\$-l mpp_p=0" >> $job_to_send
	1326	fi
	1327	echo '# @\$-lF 10mb' >> $job_to_send
	1328	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1329	echo '# @\\\$-eo' >> $job_to_send
	1330	echo " " >> $job_to_send
	1331	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1332	then
	1333	echo "set +vx" >> $job_to_send
	1334	echo ". .profile" >> $job_to_send
	1335	fi
	1336	echo "set -x" >> $job_to_send
[1094]	1337	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]	1338	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1339	echo "%%END%%" >> $job_to_send
	1340	fi
	1341
	1342	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1343	then
[1289]	1344	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1345	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1346	then
[1289]	1347	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1348	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1349	then
[1289]	1350	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1351	elif [[ $remote_host = t3es ]]
	1352	then
[1289]	1353	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1354	elif [[ $remote_host = lckiaps ]]
	1355	then
[1289]	1356	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1357	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
	1358	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1359	elif [[ $remote_host = lckyu* ]]
[1090]	1360	then
[1289]	1361	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
	1362	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1363	elif [[ $remote_host = lcflow ]]
	1364	then
[1289]	1365	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1366	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1367	else
[1289]	1368	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1369	fi
[1043]	1370	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1371	then
[1289]	1372	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1373	fi
[1]	1374	if [[ $remote_host = nech ]]
	1375	then
	1376	echo "cd -" >> $job_to_send
	1377	fi
	1378	else
	1379	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1380	# ??? funktioniert das ÃŒberhaupt noch ???
[1]	1381	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1382	fi
	1383	echo "set -x" >> $job_to_send
	1384	echo " ' exit" >> $job_to_send
	1385	echo "set -x" >> $job_to_send
	1386	fi
	1387
	1388
	1389
[1289]	1390	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1391	cat $file_to_send >> $job_to_send
[69]	1392
[1]	1393	if [[ $remote_host = ibm ]]
	1394	then
	1395	echo " " >> $job_to_send
	1396	echo "exit" >> $job_to_send
	1397	fi
[635]	1398
[1289]	1399	# REMOVE JOB-FILE
[1099]	1400	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1401	then
	1402	echo " " >> $job_to_send
	1403	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1404	fi
[1]	1405
	1406
[1289]	1407	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1408	if [[ $no_submit = false ]]
	1409	then
	1410	if [[ $remote_host != $local_host ]]
	1411	then
	1412	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1413	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1414	then
[82]	1415	job_catalog_save=$job_catalog
	1416	job_catalog=job_queue
	1417	elif [[ $remote_host = nech ]]
	1418	then
	1419	job_catalog_save=$job_catalog
	1420	job_catalog=/hpf/b/${remote_user}/job_queue
	1421	fi
[1096]	1422	if [[ $remote_host = nech ]]
[82]	1423	then
[1289]	1424	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1425	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1426	else
[1094]	1427	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1428	fi
[82]	1429	if [[ $? = 1 ]]
	1430	then
	1431	locat=scp; exit
	1432	fi
	1433	if [[ $remote_host = ibms ]]
	1434	then
	1435	job_catalog=$job_catalog_save
	1436	fi
[1]	1437	[[ $verify = true ]] && printf "\n >>> finished\n"
	1438	else
	1439	eval job_catalog=$job_catalog
	1440	cp $job_to_send ${job_catalog}/$job_on_remhost
	1441	fi
	1442
	1443
	1444
[1289]	1445	# START NQS- / LOADLEVELER-JOB
[1]	1446	if [[ $remote_host != $local_host ]]
	1447	then
	1448	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1449
[1289]	1450	if [[ $remote_host = ibmku ]]
[1]	1451	then
[1094]	1452	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1453	elif [[ $remote_host = lcflow ]]
	1454	then
[1202]	1455	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1456	else
[1094]	1457	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1458	fi
[1]	1459
	1460	[[ $verify = true ]] && printf " >>> o.k.\n"
	1461	else
	1462	cd $job_catalog
[1289]	1463	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1464	then
	1465	eval $submcom $job_on_remhost
[1289]	1466	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1467	then
[635]	1468	chmod u+x $job_on_remhost
[108]	1469	eval $submcom $job_on_remhost
[1]	1470	elif [[ $local_host = nech ]]
	1471	then
	1472	if [[ $queue = default ]]
	1473	then
[799]	1474	eval $submcom $job_on_remhost
[1]	1475	else
[799]	1476	eval $submcom -q $queue $job_on_remhost
[1]	1477	fi
	1478	else
	1479	qsub $job_on_remhost
	1480	fi
[622]	1481
[1289]	1482	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1483	# AT THE END OF THE JOB
[1099]	1484	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1485	then
	1486	rm $job_on_remhost
	1487	fi
[1]	1488	cd - > /dev/null
	1489	fi
	1490	fi
	1491
	1492
	1493
[1289]	1494	# FINAL ACTIONS
[1]	1495	if [[ $no_submit = false ]]
	1496	then
[1099]	1497	rm -f $job_to_send
[1]	1498	fi
[1266]	1499	[[ $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 |