Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1289

Last change on this file since 1289 was 1289, checked in by raasch, 10 years ago
script comments translated to English version update to 3.10
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	#
	20	# Copyright 1997-2012 Leibniz University Hannover
	21	#--------------------------------------------------------------------------------#
	22	#
	23	# Current revisions:
[1090]	24	# ------------------
[1289]	25	# German comments translated to English
	26	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
[1046]	27	#
	28	# Former revisions:
	29	# -----------------
[169]	30	# $Id: subjob 1289 2014-03-04 07:12:34Z raasch $
[1200]	31	#
[1280]	32	# 1279 2014-01-28 12:10:14Z raasch
	33	# node calculation modified due to changes in mrun (tasks_per_node must not be
	34	# an integral divisor of numprocs any more)
	35	#
[1275]	36	# 1274 2014-01-09 13:14:54Z heinze
	37	# adjustments for lccrayh
	38	#
[1267]	39	# 1266 2013-12-11 12:07:34Z heinze
	40	# further adjustments for lccrayb (use msub instead of qsub)
	41	#
[1265]	42	# 1264 2013-12-09 12:46:09Z fricke
	43	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	44	#
[1263]	45	# 1262 2013-12-09 10:57:20Z fricke
	46	# further adjustments for lccrayb
	47	#
[1261]	48	# 1260 2013-12-04 12:48:04Z raasch
	49	# jaboticaba admitted
	50	#
[1256]	51	# 1255 2013-11-07 14:43:35Z raasch
	52	# further adjustments for lccrayb
	53	#
[1225]	54	# 1224 2013-09-16 07:27:23Z raasch
	55	# first adjustments for lccrayb
	56	#
[1203]	57	# 1202 2013-07-10 16:22:07Z witha
	58	# adjustments for Forwind cluster (lcflow)
	59	#
[1200]	60	# 1199 2013-07-05 14:52:22Z raasch
	61	# adjustments for CSC Helsinki (lccrayf)
	62	#
[1185]	63	# use of cluster/express queue enabled (ibmh)
	64	# vinessa added (imuk)
[1047]	65	#
[1104]	66	# 1103 2013-02-20 02:15:53Z raasch
	67	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	68	# instead typeset -L/R),
	69	# further adjustments for lckyuh
	70	#
[1100]	71	# 2013-02-10 01:47:43Z raasch
	72	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	73	# and for Forwind cluster (lcflow)
	74	#
[1096]	75	# 1094 2013-02-03 01:52:12Z raasch
	76	# new option -P for explicit setting of ssh/scp port,
	77	# decalpha parts (yonsei) removed
	78	#
[1091]	79	# 2013-02-02 07:06:13Z raasch
[1099]	80	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	81	# old changelog messages removed
	82	#
[1047]	83	# 1046 2012-11-09 14:38:45Z maronga
	84	# code put under GPL (PALM 3.9)
	85	#
[1090]	86	# 08/07/94 - Siggi - first version finished
	87	# 29/06/94 - Siggi - script development started
	88	#--------------------------------------------------------------------------------#
	89	# subjob - script for automatic generation and submission of batch-job files
	90	# for various batch queuing systems
	91	#--------------------------------------------------------------------------------#
[1]	92
	93
[1289]	94	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	95	delete_dayfile=false
[799]	96	email_notification=none
[122]	97	group_number=none
[1]	98	locat=normal
	99	no_default_queue=none
	100	no_submit=false
	101	job_catalog="~/job_queue"
	102	job_name=none
	103	local_user=$LOGNAME
	104	node_usage=shared
[475]	105	numprocs=0
[1]	106	punkte="..........................................................."
	107	submcom=qsub
	108	queue=default
	109	remote_host=none
	110	remote_user=""
	111	verify=true
	112
	113	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	114	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	115
	116
	117
[1289]	118	# ERROR HANDLING
	119	# IN CASE OF EXIT:
[1]	120	trap 'if [[ $locat != normal ]]
	121	then
	122	case $locat in
	123	(option) printf "\n --> available optios can be displayed"
	124	printf " by typing:"
	125	printf "\n \"subjob ?\" \n";;
	126	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	127	(*) printf "\n +++ unknown error"
	128	printf "\n please inform S. Raasch!\n"
	129	esac
	130	[[ -f $job_to_send ]] && rm $job_to_send
	131	printf "\n\n+++ SUBJOB killed \n\n"
	132	fi' exit
	133
	134
[1289]	135	# IN CASE OF TERMINAL-BREAK:
[1]	136	trap '[[ -f $job_to_send ]] && rm $job_to_send
	137	printf "\n\n+++ SUBJOB killed \n\n"
	138	exit
	139	' 2
	140
	141
	142
	143
[1289]	144	# DETERMINE NAME OF LOCAL HOST
[1]	145	local_host=$(hostname)
	146
	147
	148
[1289]	149	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	150	# IS ADMITTED AT ALL)
	151	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	152	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	153	case $local_host in
[1204]	154	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
[1]	155	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[1235]	156	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
[102]	157	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	158	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1103]	159	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
[749]	160	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[1255]	161	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
[1274]	162	(hlogin\|hxc) local_addres=130.75.4.1; local_host=lccrayh;;
[12]	163	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[1174]	164	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
[372]	165	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	166	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[1197]	167	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
[1]	168	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	169	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[1099]	170	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
	171	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
[367]	172	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	173	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	174	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	175	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[5]	176	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	177	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1099]	178	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
[892]	179	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[440]	180	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	181	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	182	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[1260]	183	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
[407]	184	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	185	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
[1040]	186	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
[5]	187	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1099]	188	(meller) local_addres=134.106.74.155; local_host=lcfor;;
[437]	189	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	190	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	191	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	192	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	193	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	194	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[1099]	195	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
[789]	196	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	197	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	198	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1204]	199	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
[1197]	200	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
[799]	201	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1090]	202	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
[635]	203	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1021]	204	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
[1184]	205	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
[1]	206	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	207	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	208	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	209	printf "\n please contact the PALM group at IMUK";
[1]	210	locat=parameter; exit;;
	211	esac
	212
	213
	214
[1289]	215	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	216	remote_host=$local_host
	217
	218
	219
	220
[1289]	221	# READ THE SHELLSCRIPT-OPTIONS
[1094]	222	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	223	do
	224	case $option in
	225	(c) job_catalog=$OPTARG;;
	226	(d) delete_dayfile=true;;
	227	(D) no_submit=true;;
[352]	228	(e) email_notification=$OPTARG;;
[125]	229	(g) group_number=$OPTARG;;
[1]	230	(h) remote_host=$OPTARG;;
	231	(m) memory=$OPTARG;;
	232	(n) job_name=$OPTARG;;
	233	(N) node_usage=$OPTARG;;
	234	(O) threads_per_task=$OPTARG;;
[1094]	235	(P) scp_port=$OPTARG;;
[1]	236	(q) no_default_queue=$OPTARG;;
	237	(t) cputime=$OPTARG;;
	238	(T) tasks_per_node=$OPTARG;;
	239	(u) remote_user=$OPTARG;;
	240	(v) verify=false;;
	241	(X) numprocs=$OPTARG;;
	242	(\?) printf "\n +++ Option $OPTARG unknown \n";
	243	locat=option; exit;;
	244	esac
	245	done
	246
	247
[1289]	248	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	249	(( to_shift = $OPTIND - 1 ))
	250	shift $to_shift; file_to_send=$1
[1]	251
	252
[1289]	253	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	254	if [ "$1" = "?" ]
	255	then
	256	(printf "\n *** subjob can be called as follows:\n"
	257	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	258	printf "\n Description of available options:\n"
	259	printf "\n Option Description Default-Value"
	260	printf "\n -c job-input- and output-catalog ~/job_queue"
	261	printf "\n -d no job-protocol will be created ---"
	262	printf "\n -D only the job-file will be created ---"
	263	printf "\n -h execution host, available hosts: $remote_host"
[1289]	264	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	265	printf "\n lckiaps, lctit, nech"
[1]	266	printf "\n -m memory demand per process in MByte ---"
	267	printf "\n -n jobname <jobdatei>"
	268	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	269	printf "\n -P ssh/scp port default port"
[1]	270	printf "\n -q job-queue to be used default"
	271	printf "\n -t allowed cpu-time in seconds ---"
	272	printf "\n -T tasks per node (on parallel hosts) ---"
	273	printf "\n -u username on execution host from .netrc"
	274	printf "\n -v no prompt for confirmation ---"
	275	printf "\n -X # of processors (on parallel hosts) 1"
	276	printf "\n "
	277	printf "\n The only possible positional parameter is <jobfile>:"
	278	printf "\n The complete NQS-job must be provided here."
	279	printf "\n <jobfile>=? creates this outline\n\n") \| more
	280	exit
	281	fi
	282
	283
	284
[1289]	285	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	286	if [[ "$file_to_send" = "" ]]
	287	then
	288	printf "\n +++ job-file missing"
	289	locat=parameter; exit
	290	else
	291	if [[ -f $file_to_send ]]
	292	then
	293	true
	294	else
	295	printf "\n +++ job-file: "
	296	printf "\n $file_to_send"
	297	printf "\n does not exist"
	298	locat=parameter; exit
	299	fi
	300	fi
	301
	302
	303
[1289]	304	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	305	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	306	if [[ $job_name = none ]]
	307	then
	308	job_name=$file_to_send
	309	fi
	310	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	311	then
	312	printf "\n +++ job-file name: "
	313	printf "\n $job_name"
	314	printf "\n must not contain \"/\"-characters"
	315	locat=parameter; exit
	316	fi
	317
	318
	319
	320
[1289]	321	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	322	# OR IF NO HOST HAS BEEN GIVEN
[1]	323	if [[ $remote_host = none ]]
	324	then
	325	printf "\n +++ host missing"
	326	locat=option; exit
	327	else
	328	case $remote_host in
[251]	329	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1184]	330	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	331	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	332	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	333	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1262]	334	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1274]	335	(lccrayh) queue=mpp1testq; remote_addres=130.75.4.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1197]	336	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
[1099]	337	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
[440]	338	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[367]	339	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[1040]	340	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
[717]	341	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[1099]	342	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
[1090]	343	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
[1021]	344	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
[635]	345	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[892]	346	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
[437]	347	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	348	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	349	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	350	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	351	locat=parameter; exit;;
[1]	352	esac
	353	fi
	354
	355
[1289]	356	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	357	if [[ $no_default_queue != none ]]
	358	then
	359	error=false
	360	ndq=$no_default_queue
	361	case $remote_host in
	362	(ibm) case $ndq in
	363	(p690_express\|p690_standard\|p690_long) error=false;;
	364	(*) error=true;;
	365	esac;;
	366	(ibmh) case $ndq in
[1184]	367	(cluster\|express) error=false;;
[1]	368	(*) error=true;;
	369	esac;;
[693]	370	(ibmkisti) case $ndq in
	371	(class.32plus\|class.1-2\|class.2-32) error=false;;
	372	(*) error=true;;
	373	esac;;
[622]	374	(ibmku) case $ndq in
	375	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	376	(*) error=true;;
	377	esac;;
[1]	378	(ibms) case $ndq in
	379	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	380	(*) error=true;;
	381	esac;;
[1224]	382	(lccrayb) case $ndq in
[1274]	383	(mpp1q\|testq\|specialm1q) error=false;;
[1224]	384	(*) error=true;;
	385	esac;;
[1274]	386	(lccrayh) case $ndq in
	387	(mpp1q\|mpp1testq\|specialm1q\|dataq) error=false;;
	388	(*) error=true;;
	389	esac;;
[1197]	390	(lccrayf) case $ndq in
	391	(usup\|test*\|small\|large) error=false;;
	392	(*) error=true;;
	393	esac;;
[1099]	394	(lcflow) case $ndq in
	395	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
	396	(*) error=true;;
	397	esac;;
[1040]	398	(lckiaps) case $ndq in
	399	(express\|normal) error=false;;
	400	(*) error=true;;
	401	esac;;
[440]	402	(lckyoto) case $ndq in
	403	(eh\|ph) error=false;;
	404	(*) error=true;;
	405	esac;;
[1099]	406	(lckyuh) case $ndq in
	407	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	408	(*) error=true;;
	409	esac;;
[1090]	410	(lckyut) case $ndq in
	411	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	412	(*) error=true;;
	413	esac;;
[1]	414	(lctit) case $ndq in
[635]	415	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	416	(*) error=true;;
	417	esac;;
	418	(t3eb) case $ndq in
	419	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	420	(*) error=true;;
	421	esac;;
	422	(t3eh) case $ndq in
	423	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	424	(*) error=true;;
	425	esac;;
	426	(t3ej2\|t3ej5) case $ndq in
	427	(low\|normal\|high) error=false;;
	428	(*) error=true;;
	429	esac;;
	430	(t3es) case $ndq in
	431	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	432	(*) error=true;;
	433	esac;;
	434	esac
	435	if [[ $error = true ]]
	436	then
	437	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	438	locat=parameter; exit
	439	else
	440	queue=$no_default_queue
	441	fi
	442	fi
	443
	444
	445
[1289]	446	# CHECK THE CPU-TIME
	447	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	448	done=false
	449	while [[ $done = false ]]
	450	do
	451	if (( $cputime <= 0 ))
	452	then
	453	printf "\n +++ wrong cpu-time or cpu-time missing"
	454	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	455	printf "\n >>> "
	456	read cputime 1>/dev/null 2>&1
	457	else
	458	done=true
	459	fi
	460	done
	461	if [[ $remote_host = nech ]]
	462	then
	463	if (( tasks_per_node != 0 ))
	464	then
	465	(( cputime = cputime * tasks_per_node ))
	466	elif [[ $numprocs != 0 ]]
	467	then
	468	(( cputime = cputime * numprocs ))
	469	fi
	470	fi
	471	(( stunden = cputime / 3600 ))
	472	(( resttime = cputime - stunden * 3600 ))
	473	(( minuten = resttime / 60 ))
	474	(( sekunden = resttime - minuten * 60 ))
	475	timestring=${stunden}:${minuten}:${sekunden}
	476
	477
	478
[1289]	479	# CHECK THE MEMORY DEMAND
[1]	480	done=false
	481	while [[ $done = false ]]
	482	do
	483	if (( memory <= 0 ))
	484	then
	485	printf "\n +++ wrong memory demand or memory demand missing"
	486	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	487	printf "\n >>> "
	488	read memory 1>/dev/null 2>&1
	489	else
	490	done=true
	491	fi
	492	done
	493
[1289]	494	if [[ $remote_host = nech ]]
[1]	495	then
	496	if (( tasks_per_node != 0 ))
	497	then
	498	(( Memory = memory * tasks_per_node / 1000 ))
	499	elif [[ $numprocs != 0 ]]
	500	then
	501	(( Memory = memory * numprocs / 1000 ))
	502	else
	503	(( Memory = memory / 1000 ))
	504	fi
[635]	505	elif [[ $remote_host = lctit ]]
	506	then
	507	(( Memory = memory * tasks_per_node / 1000 ))
[1]	508	fi
	509
	510
[1289]	511	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	512	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	513	then
	514	(( memory = memory * threads_per_task ))
	515	fi
	516
	517
[1289]	518	# CALCULATE NUMBER OF REQUIRED NODES
[1]	519	if (( tasks_per_node != 0 ))
	520	then
[1279]	521	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	522	fi
	523
[1094]	524
[1289]	525	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	526	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	527
[1094]	528
[1289]	529	# CALCULATE NUMBER OF MPI TASKS
[696]	530	(( mpi_tasks = numprocs / threads_per_task ))
[1]	531
[696]	532
[1289]	533	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	534	if [[ "$scp_port" != "" ]]
	535	then
	536	PORTOPT="-P $scp_port"
	537	SSH_PORTOPT="-p $scp_port"
	538	fi
	539
	540
[1289]	541	# HEADER-OUTPUT
[1]	542	if [[ $verify = true ]]
	543	then
	544	printf "\n\n"
	545	printf "#--------------------------------------------------------------# \n"
	546	spalte1=SUBJOB;spalte2=$(date)
[1103]	547	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	548	printf "\| \| \n"
	549	printf "\| values of parameters/options: \| \n"
[1103]	550	spalte1=$(echo local_host$punkte \| cut -c-20)
	551	spalte2=$punkte$local_host
	552	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	553	spalte1=$(echo remote_host$punkte \| cut -c-20)
	554	spalte2=$punkte$remote_host
	555	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	556	spalte1=$(echo queue$punkte \| cut -c-20)
	557	spalte2=$punkte$queue
	558	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	559	spalte1=$(echo memory$punkte \| cut -c-20)
	560	spalte2="$punkte$memory mb"
	561	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	562	spalte1=$(echo cputime$punkte \| cut -c-20)
	563	spalte2="$punkte$cputime sec"
	564	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	565	spalte1=$(echo job_name$punkte \| cut -c-20)
	566	spalte2="$punkte$job_name"
	567	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	568	printf "#--------------------------------------------------------------# \n\n"
	569
	570
[1289]	571	# QUERY CHECK
[1]	572	antwort="dummy"
	573	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	574	do
	575	read antwort?" >>> continue (y/n) ? "
	576	done
	577	if [[ $antwort = n \|\| $antwort = N ]]
	578	then
	579	locat=verify; exit
	580	fi
	581	printf "\n"
	582	fi
	583
	584
	585
[1289]	586	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	587	identifier=$RANDOM
	588	job_on_remhost=${job_name}_${identifier}_$local_host
	589	job_to_send=job_to_send_$identifier
[1]	590	if [[ $delete_dayfile = false ]]
	591	then
[1289]	592	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	593	local_dayfile=${remote_host}_${job_name}
	594	else
	595	remote_dayfile=/dev/null
	596	fi
	597
	598
	599
[1289]	600	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	601	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	602	then
	603
[1289]	604	# GENERAL LOADLEVELER SETTINGS
[622]	605	execute_in_shell="#!/bin/ksh"
	606	use_shell="# @ shell = /bin/ksh"
	607	consumable_memory="ConsumableMemory($memory mb)"
	608	class="# @ class = $queue"
	609	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	610	network_to_use="# @ network.mpi = sn_all,shared,us"
	611	data_limit="# @ data_limit = 1.76gb"
	612	image_size="# @ image_size = 50"
[693]	613	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	614
[693]	615	if [[ $email_notification = none ]]
	616	then
	617	notify_user=""
	618	else
	619	notify_user="# @ notify_user = $email_notification"
	620	if [[ $delete_dayfile = true ]]
	621	then
	622	notification='# @ notification = never'
	623	fi
	624	fi
[622]	625
[312]	626	if [[ $remote_host = ibmh ]]
[1]	627	then
[312]	628	data_limit=""
	629	network_to_use=""
[1184]	630	class="# @ class = $queue"
[312]	631	environment=""
[814]	632	rset="# @ rset = RSET_MCM_AFFINITY"
	633	task_affinity="# @ task_affinity = core(1)"
[693]	634	elif [[ $remote_host = ibmkisti ]]
	635	then
	636	network_to_use="# @ network.MPI = sn_all,shared,US"
	637	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	638	if [[ $threads_per_task = 1 ]]
	639	then
	640	rset="# @ rset = RSET_MCM_AFFINITY"
	641	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	642	fi
[693]	643	environment=""
	644	use_shell=""
	645	data_limit=""
	646	image_size=""
[622]	647	elif [[ $remote_host = ibmku ]]
	648	then
	649	execute_in_shell="#!/usr/bin/ksh"
	650	use_shell="# @ shell = /usr/bin/ksh"
	651	consumable_memory=""
	652	environment=""
	653	network_to_use="# @ network.mpi = sn_all,shared,us"
	654	data_limit=""
	655	image_size=""
	656	elif [[ $remote_host = ibms ]]
	657	then
	658	network_to_use="# @ network.mpi = csss,shared,us"
[1]	659	fi
	660
	661	cat > $job_to_send << %%END%%
[622]	662	$execute_in_shell
	663	$use_shell
[1]	664
	665	# @ job_type = parallel
[693]	666	# @ job_name = $job_name
[1]	667	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	668	# @ output = $remote_dayfile
	669	# @ error = $remote_dayfile
[693]	670	$wall_clock_limit
[622]	671	$image_size
[312]	672	$class
	673	$environment
[1]	674	$network_to_use
	675	$data_limit
[693]	676	$rset
	677	$mcm_affinity_options
[814]	678	$task_affinity
[1]	679	$notification
[693]	680	$notify_user
[1]	681
	682	%%END%%
	683
	684	if (( nodes > 0 ))
	685	then
	686
[693]	687	if [[ $remote_host != ibmkisti ]]
	688	then
	689
	690	cat >> $job_to_send << %%END%%
[1]	691	# @ node = $nodes
[622]	692	# @ tasks_per_node = $processes_per_node
[1]	693	# @ node_usage = $node_usage
	694	# @ queue
	695
	696	%%END%%
	697
[693]	698	else
	699
	700	cat >> $job_to_send << %%END%%
[696]	701	# @ total_tasks = $mpi_tasks
[693]	702	# @ blocking = unlimited
	703	# @ queue
	704
	705	%%END%%
	706
	707	fi
	708
[1]	709	else
	710
[1289]	711	cat >> $job_to_send << %%END%%
[1]	712	# @ blocking = unlimited
	713	# @ total_tasks = $numprocs
	714	# @ node_usage = $node_usage
	715	# @ queue
	716
	717	%%END%%
	718
	719	fi
	720
[1289]	721	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	722	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	723	then
	724	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	725	fi
	726
[1]	727	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	728	then
	729
	730	cat > $job_to_send << %%END%%
	731	#!/bin/ksh
	732
	733	# @ job_type = serial
	734	# @ node_usage = $node_usage
	735	# @ job_name = palm
	736	# @ wall_clock_limit = ${timestring},$timestring
	737	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	738	# @ output = $remote_dayfile
	739	# @ error = $remote_dayfile
[312]	740	$class
[1]	741	$notification
	742
	743	# @ queue
	744
	745	%%END%%
	746
[1274]	747	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	748	then
	749
	750	if [[ $numprocs != 0 ]]
	751	then
	752	cat > $job_to_send << %%END%%
[1255]	753	#!/bin/bash
[1224]	754	#PBS -N $job_name
	755	#PBS -l walltime=$timestring
[1264]	756	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	757	#PBS -o $remote_dayfile
	758	#PBS -j oe
	759	#PBS -q $queue
	760
	761	$module_calls
	762
	763	%%END%%
	764
	765	else
	766
	767	continue
	768
	769	fi
	770
[1197]	771	elif [[ $remote_host = lccrayf ]]
	772	then
	773
	774	if [[ $numprocs != 0 ]]
	775	then
	776	cat > $job_to_send << %%END%%
	777	#!/bin/bash -l
	778	#SBATCH -J $job_name
	779	#SBATCH -t $timestring
	780	#SBATCH -N $nodes
	781	#SBATCH --ntasks-per-node=$processes_per_node
	782	#SBATCH -p $queue
	783	#SBATCH -o $remote_dayfile
	784	#SBATCH -e $remote_dayfile
	785
	786	$init_cmds
	787	$module_calls
	788
	789	%%END%%
	790
	791	else
	792	cat > $job_to_send << %%END%%
	793	#!/bin/bash -l
	794	#SBATCH -J $job_name
	795	#SBATCH -t $timestring
	796	#SBATCH -l ncpus=1
	797	#SBATCH -l pmem=${memory}mb
	798	#SBATCH -m abe
	799	#SBATCH -o $remote_dayfile
	800	#SBATCH -e $remote_dayfile
	801
	802	$init_cmds
	803	$module_calls
	804
	805	%%END%%
	806
	807	fi
	808
[1099]	809	elif [[ $remote_host = lcflow ]]
	810	then
[1202]	811	if [ $memory -gt 1800 ]; then
[1099]	812	use_himem=""
	813	else
	814	use_himem="#"
	815	fi
	816	if [[ $numprocs != 0 ]]
	817	then
[1202]	818	pe_set="#$ -pe impi41 $numprocs"
[1099]	819	else
[1202]	820	pe_set="#$ -pe impi41 1"
[1099]	821	fi
	822	if [[ $queue = default ]]
	823	then
	824	queue_set=""
	825	else
	826	queue_set="#$ -q $queue"
	827	fi
	828	[[ "$disc_space" = "" ]] && disc_space=50
	829
	830	cat > $job_to_send << %%END%%
	831	#!/bin/bash
	832	#$ -S /bin/bash
	833	#$ -N $job_name
	834	#$ -cwd
	835	#$ -l h_rt=$timestring
	836	#$ -l h_vmem=${memory}M
	837	#$ -o $remote_dayfile
	838	#$ -j y
	839	$pe_set
	840	#$ -R y
	841	${use_himem}#$ -l highmem=true
	842	#$ -l h_fsize=${disc_space}G
	843	$queue_set
	844
	845	%%END%%
	846
[1021]	847	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	848	then
	849
	850	if [[ $numprocs != 0 ]]
	851	then
	852	cat > $job_to_send << %%END%%
	853	#!/bin/ksh
	854	#PBS -N $job_name
	855	#PBS -l walltime=$timestring
	856	#PBS -l ncpus=$numprocs
	857	#PBS -l pmem=${memory}mb
	858	#PBS -o $remote_dayfile
[1021]	859	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	860	#PBS -j oe
	861
	862	mpd &
	863
	864	%%END%%
	865
	866	else
	867	cat > $job_to_send << %%END%%
	868	#!/bin/ksh
	869	#PBS -N $job_name
	870	#PBS -l walltime=$timestring
	871	#PBS -l ncpus=1
	872	#PBS -l pmem=${memory}mb
	873	#PBS -o $remote_dayfile
	874	#PBS -j oe
	875
	876	%%END%%
	877
	878	fi
	879
[1040]	880	elif [[ $remote_host = lckiaps ]]
	881	then
	882
	883	if [[ $numprocs != 0 ]]
	884	then
	885	cat > $job_to_send << %%END%%
	886	#!/bin/ksh
	887	#PBS -N $job_name
	888	#PBS -l walltime=$timestring
	889	#PBS -l select=1:ncpus=$numprocs
	890	#PBS -l pmem=${memory}mb
	891	#PBS -q $queue
	892	#PBS -o $remote_dayfile
	893	#PBS -j oe
	894	#PBS -V
	895
	896	%%END%%
	897
	898	else
	899	cat > $job_to_send << %%END%%
	900	#!/bin/ksh
	901	#PBS -N $job_name
	902	#PBS -l walltime=$timestring
	903	#PBS -l ncpus=1
	904	#PBS -l pmem=${memory}mb
	905	#PBS -o $remote_dayfile
	906	#PBS -j oe
	907
	908	%%END%%
	909
	910	fi
	911
[693]	912	elif [[ $remote_host = lcyon ]]
	913	then
	914
	915	if [[ $numprocs != 0 ]]
	916	then
	917	cat > $job_to_send << %%END%%
	918	#!/bin/ksh
	919	#PBS -N $job_name
	920	#PBS -l walltime=$timestring
	921	#PBS -l ncpus=$numprocs
	922	#PBS -l pmem=${memory}mb
	923	#PBS -o $remote_dayfile
	924	#PBS -j oe
	925
	926	%%END%%
	927
	928	else
	929	cat > $job_to_send << %%END%%
	930	#!/bin/ksh
	931	#PBS -N $job_name
	932	#PBS -l walltime=$timestring
	933	#PBS -l ncpus=1
	934	#PBS -l pmem=${memory}mb
	935	#PBS -o $remote_dayfile
	936	#PBS -j oe
	937
	938	%%END%%
	939
	940	fi
	941
[892]	942	elif [[ $remote_host = lcxe6 ]]
[164]	943	then
	944
	945	if [[ $numprocs != 0 ]]
	946	then
	947	cat > $job_to_send << %%END%%
[799]	948	#!/bin/ksh
[164]	949	#PBS -S /bin/ksh
	950	#PBS -N $job_name
[552]	951	#PBS -A $project_account
[206]	952	#PBS -j oe
[164]	953	#PBS -l walltime=$timestring
	954	#PBS -l mppwidth=${numprocs}
[622]	955	#PBS -l mppnppn=${processes_per_node}
[164]	956	#PBS -m abe
	957	#PBS -o $remote_dayfile
[492]	958	$email_directive
[164]	959
[892]	960	$init_cmds
[493]	961	$module_calls
[343]	962
[164]	963	%%END%%
	964
[108]	965	else
	966	cat > $job_to_send << %%END%%
	967	#!/bin/ksh
[168]	968	#PBS -S /bin/ksh
[108]	969	#PBS -N $job_name
[552]	970	#PBS -A $project_account
[206]	971	#PBS -j oe
[108]	972	#PBS -l walltime=$timestring
	973	#PBS -l ncpus=1
	974	#PBS -l pmem=${memory}mb
	975	#PBS -m abe
[492]	976	$email_directive
[108]	977	#PBS -o $remote_dayfile
	978
[892]	979	$init_cmds
[493]	980	$module_calls
[343]	981
[108]	982	%%END%%
	983
	984	fi
	985
[440]	986	elif [[ $remote_host = lckyoto ]]
[437]	987	then
	988
[440]	989	cat > $job_to_send << %%END%%
[799]	990	#!/bin/ksh
[440]	991	# @\$-o $remote_dayfile
	992	# @\$-eo -oi
	993	# @\$-lP 16
[799]	994	# @\$-lp 1
[440]	995	# @\$-lm 28gb -llm unlimited -ls unlimited
	996	# @\$-q $queue
	997	# @\$-Pvn abs_pack
	998	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	999	#. /thin/local/etc/setprofile/intel-11.0.sh
	1000	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1001	. ~/.myprofile
	1002	#. /home2/t/t51254/palm/current_version/myprofile
	1003	#. /thin/apps/pgi/mpi.sh
	1004	#
	1005	env
	1006	#
	1007	set -x
	1008
	1009	%%END%%
	1010
	1011	elif [[ $remote_host = lcxt5m ]]
	1012	then
	1013
[437]	1014	if [[ $numprocs != 0 ]]
	1015	then
	1016	cat > $job_to_send << %%END%%
[799]	1017	#!/bin/ksh
[437]	1018	#PBS -S /bin/ksh
	1019	#PBS -N $job_name
	1020	#PBS -j oe
	1021	#PBS -l walltime=$timestring
	1022	#PBS -l mppwidth=${numprocs}
[622]	1023	#PBS -l mppnppn=${processes_per_node}
[437]	1024	#PBS -m abe
	1025	#PBS -o $remote_dayfile
	1026
[892]	1027	$init_cmds
[493]	1028	$module_calls
[437]	1029
	1030	%%END%%
	1031
	1032	else
	1033	cat > $job_to_send << %%END%%
	1034	#!/bin/ksh
	1035	#PBS -S /bin/ksh
	1036	#PBS -N $job_name
	1037	#PBS -j oe
	1038	#PBS -l walltime=$timestring
	1039	#PBS -l ncpus=1
	1040	#PBS -l pmem=${memory}mb
	1041	#PBS -m abe
	1042	#PBS -o $remote_dayfile
	1043
[892]	1044	$init_cmds
[493]	1045	$module_calls
[437]	1046
	1047	%%END%%
	1048
	1049	fi
	1050
[1099]	1051	elif [[ $remote_host = lckyuh ]]
	1052	then
	1053	cat > $job_to_send << %%END%%
	1054	#!/bin/bash
	1055	#PJM -L "rscgrp=$queue"
	1056	#PJM -L "node=$nodes"
	1057	#PJM --mpi "proc=$numprocs"
	1058	#PJM -L "elapse=$timestring"
	1059	#PJM -o $remote_dayfile
	1060	#PJM -j
	1061	#PJM -X
	1062	#PJM --no-stging
	1063
	1064	export LANG=en_US.UTF-8
	1065	%%END%%
	1066
[1090]	1067	elif [[ $remote_host = lckyut ]]
	1068	then
	1069	cat > $job_to_send << %%END%%
	1070	#!/bin/bash
[1099]	1071	#PJM -L "rscgrp=$queue"
[1090]	1072	#PJM -L "vnode=$numprocs"
	1073	#PJM -L "vnode-core=1"
	1074	#PJM -L "elapse=$timestring"
	1075	#PJM --mpi proc=$numprocs
	1076	#PJM -o $remote_dayfile
	1077	#PJM -j
[1099]	1078	#PJM -X
	1079	#PJM --no-stging
[1090]	1080
	1081	export LANG=en_US.UTF-8
	1082	%%END%%
	1083
[1]	1084	elif [[ $remote_host = nech ]]
	1085	then
	1086
	1087	if (( nodes > 1 ))
	1088	then
	1089	cat > $job_to_send << %%END%%
	1090	#!/bin/ksh
[622]	1091	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1092	#PBS -l ${qsubmem}=${Memory}gb
	1093	#PBS -b $nodes
	1094	#PBS -o $remote_dayfile
	1095	#PBS -N palm
	1096	#PBS -j o
	1097	#PBS -T mpisx
	1098
	1099	%%END%%
	1100
	1101	elif [[ $numprocs != 0 ]]
	1102	then
	1103	cat > $job_to_send << %%END%%
	1104	#!/bin/ksh
[622]	1105	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1106	#PBS -l ${qsubmem}=${Memory}gb
	1107	#PBS -o $remote_dayfile
	1108	#PBS -N palm
	1109	#PBS -j o
	1110
	1111	%%END%%
	1112
	1113	else
	1114	cat > $job_to_send << %%END%%
	1115	#!/bin/ksh
	1116	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1117	#PBS -o $remote_dayfile
	1118	#PBS -j o
	1119
	1120	%%END%%
	1121
	1122	fi
	1123
	1124	elif [[ $remote_host = lctit ]]
	1125	then
	1126	cat > $job_to_send << %%END%%
[635]	1127	#!/bin/ksh
[892]	1128	$init_cmds
[678]	1129	$module_calls
	1130
[1]	1131	%%END%%
	1132
[1289]	1133	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1134	if [[ $tasks_per_node != $processes_per_node ]]
	1135	then
	1136	submcom="$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 $queue "
	1137	else
	1138	submcom="$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 "
	1139	fi
[1]	1140
	1141	else
	1142
	1143	cat > $job_to_send << %%END%%
	1144	# @\$-q ${queue}
	1145	# @\$-l${qsubtime} $timestring
	1146	# @\$-l${qsubmem} ${memory}mb
	1147	# @\$-o $remote_dayfile
	1148	# @\$-eo
	1149
	1150	%%END%%
	1151
	1152	fi
	1153
	1154
[1289]	1155	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1156	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1157	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1158	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1159	then
	1160	echo "set +vx" >> $job_to_send
	1161	echo "trap '" >> $job_to_send
	1162	echo "set +vx" >> $job_to_send
[1289]	1163	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]	1164	then
[622]	1165	if [[ $remote_host = ibmh ]]
[1]	1166	then
	1167	return_queue=c1
[693]	1168	elif [[ $remote_host = ibmkisti ]]
	1169	then
	1170	return_queue=class.1-2
[622]	1171	elif [[ $remote_host = ibmku ]]
	1172	then
	1173	return_queue=sdbg2
[1]	1174	elif [[ $remote_host = ibms ]]
	1175	then
	1176	return_queue=p_normal
[1274]	1177	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1178	then
	1179	return_queue=dataq
[1040]	1180	elif [[ $remote_host = lckiaps ]]
	1181	then
	1182	return_queue=express
[1099]	1183	elif [[ $remote_host = lckyuh ]]
	1184	then
	1185	return_queue=cx-single
[1090]	1186	elif [[ $remote_host = lckyut ]]
	1187	then
	1188	return_queue=cx-single
[1]	1189	else
	1190	return_queue=unknown
	1191	fi
	1192
	1193	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1194	then
	1195
[622]	1196	if [[ $remote_host = ibmku ]]
	1197	then
[1289]	1198	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1199	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1200	else
[1289]	1201	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1202	fi
[1289]	1203	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1204	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1205	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1206	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1207	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1208	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1209	if [[ $host != "ibmh" ]]
	1210	then
[1289]	1211	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1212	fi
[1289]	1213	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1214	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1215
[1289]	1216	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1217	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1218
[1289]	1219	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
	1220	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]	1221	if [[ $remote_host = ibmku ]]
	1222	then
[1289]	1223	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1224	fi
[1289]	1225	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1226
	1227	elif [[ $remote_host = nech ]]
	1228	then
	1229	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1230	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1231	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1232	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1233	echo "#PBS -j o" >> $job_to_send
	1234	echo " " >> $job_to_send
	1235	echo "set -x" >> $job_to_send
	1236	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1094]	1237	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1238	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1239	echo "%%END%%" >> $job_to_send
	1240
[1099]	1241	elif [[ $remote_host = lckyuh ]]
	1242	then
[1289]	1243	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1244	echo "#!/bin/bash" >> $job_to_send
	1245	echo "#PJM -L \"node=1\"" >> $job_to_send
	1246	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1247	echo "#PJM --no-stging" >> $job_to_send
	1248	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1249	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1250	echo "#PJM -j" >> $job_to_send
	1251	echo " " >> $job_to_send
	1252	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1253	echo "set -x" >> $job_to_send
	1254	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1255	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1256	echo "%%END%%" >> $job_to_send
	1257
[1090]	1258	elif [[ $remote_host = lckyut ]]
	1259	then
[1289]	1260	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1261	echo "#!/bin/bash" >> $job_to_send
	1262	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1263	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1264	echo "#PJM --no-stging" >> $job_to_send
	1265	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1266	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1267	echo "#PJM -j" >> $job_to_send
	1268	echo " " >> $job_to_send
	1269	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1270	echo "set -x" >> $job_to_send
[1094]	1271	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1272	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1273	echo "%%END%%" >> $job_to_send
	1274
[1274]	1275	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1276	then
[1289]	1277	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1278	echo "#!/bin/bash" >> $job_to_send
	1279	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1280	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1281	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1282	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1283	echo "#PBS -j oe" >> $job_to_send
	1284	echo " " >> $job_to_send
	1285	echo "set -x" >> $job_to_send
	1286	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1287	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1288	echo "%%END%%" >> $job_to_send
	1289
[1099]	1290	elif [[ $remote_host = lcflow ]]
	1291	then
[1289]	1292	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1293	echo "#!/bin/bash" >> $job_to_send
	1294	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1295	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1296	echo "SGEPREFIX -cwd" >> $job_to_send
	1297	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1298	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
	1299	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
[1099]	1300	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1301	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1302	echo " " >> $job_to_send
	1303	echo "set -x" >> $job_to_send
	1304	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1305	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1306	echo "" >> $job_to_send
	1307	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1308	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1309	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1310	echo "%%END%%" >> $job_to_send
[1289]	1311	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1312	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1099]	1313
[1]	1314	else
	1315
[1289]	1316	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1317	echo "# @\\\$-q $return_queue" >> $job_to_send
	1318	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1319	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1320	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1321	then
	1322	echo "# @\$-l mpp_p=0" >> $job_to_send
	1323	fi
	1324	echo '# @\$-lF 10mb' >> $job_to_send
	1325	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1326	echo '# @\\\$-eo' >> $job_to_send
	1327	echo " " >> $job_to_send
	1328	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1329	then
	1330	echo "set +vx" >> $job_to_send
	1331	echo ". .profile" >> $job_to_send
	1332	fi
	1333	echo "set -x" >> $job_to_send
[1094]	1334	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]	1335	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1336	echo "%%END%%" >> $job_to_send
	1337	fi
	1338
	1339	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1340	then
[1289]	1341	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1342	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1343	then
[1289]	1344	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1345	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1346	then
[1289]	1347	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1348	elif [[ $remote_host = t3es ]]
	1349	then
[1289]	1350	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1351	elif [[ $remote_host = lckiaps ]]
	1352	then
[1289]	1353	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1354	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
	1355	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1356	elif [[ $remote_host = lckyu* ]]
[1090]	1357	then
[1289]	1358	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
	1359	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1360	elif [[ $remote_host = lcflow ]]
	1361	then
[1289]	1362	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1363	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1364	else
[1289]	1365	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1366	fi
[1043]	1367	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1368	then
[1289]	1369	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1370	fi
[1]	1371	if [[ $remote_host = nech ]]
	1372	then
	1373	echo "cd -" >> $job_to_send
	1374	fi
	1375	else
	1376	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1377	# ??? funktioniert das ÃŒberhaupt noch ???
[1]	1378	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1379	fi
	1380	echo "set -x" >> $job_to_send
	1381	echo " ' exit" >> $job_to_send
	1382	echo "set -x" >> $job_to_send
	1383	fi
	1384
	1385
	1386
[1289]	1387	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1388	cat $file_to_send >> $job_to_send
[69]	1389
[1]	1390	if [[ $remote_host = ibm ]]
	1391	then
	1392	echo " " >> $job_to_send
	1393	echo "exit" >> $job_to_send
	1394	fi
[635]	1395
[1289]	1396	# REMOVE JOB-FILE
[1099]	1397	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1398	then
	1399	echo " " >> $job_to_send
	1400	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1401	fi
[1]	1402
	1403
[1289]	1404	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1405	if [[ $no_submit = false ]]
	1406	then
	1407	if [[ $remote_host != $local_host ]]
	1408	then
	1409	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1410	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1411	then
[82]	1412	job_catalog_save=$job_catalog
	1413	job_catalog=job_queue
	1414	elif [[ $remote_host = nech ]]
	1415	then
	1416	job_catalog_save=$job_catalog
	1417	job_catalog=/hpf/b/${remote_user}/job_queue
	1418	fi
[1096]	1419	if [[ $remote_host = nech ]]
[82]	1420	then
[1289]	1421	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1422	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1423	else
[1094]	1424	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1425	fi
[82]	1426	if [[ $? = 1 ]]
	1427	then
	1428	locat=scp; exit
	1429	fi
	1430	if [[ $remote_host = ibms ]]
	1431	then
	1432	job_catalog=$job_catalog_save
	1433	fi
[1]	1434	[[ $verify = true ]] && printf "\n >>> finished\n"
	1435	else
	1436	eval job_catalog=$job_catalog
	1437	cp $job_to_send ${job_catalog}/$job_on_remhost
	1438	fi
	1439
	1440
	1441
[1289]	1442	# START NQS- / LOADLEVELER-JOB
[1]	1443	if [[ $remote_host != $local_host ]]
	1444	then
	1445	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1446
[1289]	1447	if [[ $remote_host = ibmku ]]
[1]	1448	then
[1094]	1449	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1450	elif [[ $remote_host = lcflow ]]
	1451	then
[1202]	1452	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1453	else
[1094]	1454	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1455	fi
[1]	1456
	1457	[[ $verify = true ]] && printf " >>> o.k.\n"
	1458	else
	1459	cd $job_catalog
[1289]	1460	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1461	then
	1462	eval $submcom $job_on_remhost
[1289]	1463	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1464	then
[635]	1465	chmod u+x $job_on_remhost
[108]	1466	eval $submcom $job_on_remhost
[1]	1467	elif [[ $local_host = nech ]]
	1468	then
	1469	if [[ $queue = default ]]
	1470	then
[799]	1471	eval $submcom $job_on_remhost
[1]	1472	else
[799]	1473	eval $submcom -q $queue $job_on_remhost
[1]	1474	fi
	1475	else
	1476	qsub $job_on_remhost
	1477	fi
[622]	1478
[1289]	1479	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1480	# AT THE END OF THE JOB
[1099]	1481	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1482	then
	1483	rm $job_on_remhost
	1484	fi
[1]	1485	cd - > /dev/null
	1486	fi
	1487	fi
	1488
	1489
	1490
[1289]	1491	# FINAL ACTIONS
[1]	1492	if [[ $no_submit = false ]]
	1493	then
[1099]	1494	rm -f $job_to_send
[1]	1495	fi
[1266]	1496	[[ $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 |