Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1567

Last change on this file since 1567 was 1548, checked in by witha, 10 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 52.3 KB

Rev	Line
[1]	1	#!/bin/ksh
[1090]	2
	3	# subjob - script for automatic generation and submission of batch-job files
	4	# for various batch queuing systems
	5
[1046]	6	#--------------------------------------------------------------------------------#
	7	# This file is part of PALM.
	8	#
	9	# PALM is free software: you can redistribute it and/or modify it under the terms
	10	# of the GNU General Public License as published by the Free Software Foundation,
	11	# either version 3 of the License, or (at your option) any later version.
	12	#
	13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
	14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
	15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
	16	#
	17	# You should have received a copy of the GNU General Public License along with
	18	# PALM. If not, see <http://www.gnu.org/licenses/>.
	19	#
[1310]	20	# Copyright 1997-2014 Leibniz Universitaet Hannover
[1046]	21	#--------------------------------------------------------------------------------#
	22	#
	23	# Current revisions:
[1090]	24	# ------------------
[1351]	25	#
[1548]	26	#
[1046]	27	# Former revisions:
	28	# -----------------
[169]	29	# $Id: subjob 1548 2015-01-29 15:11:41Z suehring $
[1200]	30	#
[1548]	31	# 1547 2015-01-29 15:09:12Z witha
	32	# adjustments for ForWind computing cluster (lcflow)
	33	#
[1546]	34	# 1545 2015-01-29 06:52:23Z heinze
	35	# local host name for blizzard further specified
	36	#
[1481]	37	# 1480 2014-10-17 14:41:49Z raasch
	38	# adjustments for 2nd stage of HLRNIII
	39	#
[1469]	40	# 1468 2014-09-24 14:06:57Z maronga
	41	# Typo removed (addres->address)
	42	# Adjustments for lcxe6
	43	#
[1453]	44	# 1452 2014-08-22 09:41:06Z heinze
	45	# local hosts for blizzard added
	46	#
[1451]	47	# 1450 2014-08-21 07:31:51Z heinze
	48	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
	49	#
[1443]	50	# 1442 2014-07-28 07:09:10Z raasch
	51	# HLRN-III (lccrayb/lccrayh) queues adjusted
	52	#
[1379]	53	# 1378 2014-04-28 06:04:58Z raasch
	54	# -et option added for lctit
	55	#
[1351]	56	# 1350 2014-04-04 13:01:30Z maronga
	57	# location of qsub updated for lcxe6
	58	#
[1290]	59	# 1289 2014-03-04 07:12:34Z raasch
	60	# German comments translated to English
	61	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
	62	#
[1280]	63	# 1279 2014-01-28 12:10:14Z raasch
	64	# node calculation modified due to changes in mrun (tasks_per_node must not be
	65	# an integral divisor of numprocs any more)
	66	#
[1275]	67	# 1274 2014-01-09 13:14:54Z heinze
	68	# adjustments for lccrayh
	69	#
[1267]	70	# 1266 2013-12-11 12:07:34Z heinze
	71	# further adjustments for lccrayb (use msub instead of qsub)
	72	#
[1265]	73	# 1264 2013-12-09 12:46:09Z fricke
	74	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	75	#
[1263]	76	# 1262 2013-12-09 10:57:20Z fricke
	77	# further adjustments for lccrayb
	78	#
[1261]	79	# 1260 2013-12-04 12:48:04Z raasch
	80	# jaboticaba admitted
	81	#
[1256]	82	# 1255 2013-11-07 14:43:35Z raasch
	83	# further adjustments for lccrayb
	84	#
[1225]	85	# 1224 2013-09-16 07:27:23Z raasch
	86	# first adjustments for lccrayb
	87	#
[1203]	88	# 1202 2013-07-10 16:22:07Z witha
	89	# adjustments for Forwind cluster (lcflow)
	90	#
[1200]	91	# 1199 2013-07-05 14:52:22Z raasch
	92	# adjustments for CSC Helsinki (lccrayf)
	93	#
[1185]	94	# use of cluster/express queue enabled (ibmh)
	95	# vinessa added (imuk)
[1047]	96	#
[1104]	97	# 1103 2013-02-20 02:15:53Z raasch
	98	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	99	# instead typeset -L/R),
	100	# further adjustments for lckyuh
	101	#
[1100]	102	# 2013-02-10 01:47:43Z raasch
	103	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	104	# and for Forwind cluster (lcflow)
	105	#
[1096]	106	# 1094 2013-02-03 01:52:12Z raasch
	107	# new option -P for explicit setting of ssh/scp port,
	108	# decalpha parts (yonsei) removed
	109	#
[1091]	110	# 2013-02-02 07:06:13Z raasch
[1099]	111	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	112	# old changelog messages removed
	113	#
[1047]	114	# 1046 2012-11-09 14:38:45Z maronga
	115	# code put under GPL (PALM 3.9)
	116	#
[1090]	117	# 08/07/94 - Siggi - first version finished
	118	# 29/06/94 - Siggi - script development started
	119	#--------------------------------------------------------------------------------#
	120	# subjob - script for automatic generation and submission of batch-job files
	121	# for various batch queuing systems
	122	#--------------------------------------------------------------------------------#
[1]	123
	124
[1289]	125	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	126	delete_dayfile=false
[799]	127	email_notification=none
[122]	128	group_number=none
[1]	129	locat=normal
	130	no_default_queue=none
	131	no_submit=false
	132	job_catalog="~/job_queue"
	133	job_name=none
	134	local_user=$LOGNAME
	135	node_usage=shared
[475]	136	numprocs=0
[1]	137	punkte="..........................................................."
	138	submcom=qsub
	139	queue=default
	140	remote_host=none
	141	remote_user=""
	142	verify=true
	143
	144	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	145	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	146
	147
	148
[1289]	149	# ERROR HANDLING
	150	# IN CASE OF EXIT:
[1]	151	trap 'if [[ $locat != normal ]]
	152	then
	153	case $locat in
	154	(option) printf "\n --> available optios can be displayed"
	155	printf " by typing:"
	156	printf "\n \"subjob ?\" \n";;
	157	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	158	(*) printf "\n +++ unknown error"
	159	printf "\n please inform S. Raasch!\n"
	160	esac
	161	[[ -f $job_to_send ]] && rm $job_to_send
	162	printf "\n\n+++ SUBJOB killed \n\n"
	163	fi' exit
	164
	165
[1289]	166	# IN CASE OF TERMINAL-BREAK:
[1]	167	trap '[[ -f $job_to_send ]] && rm $job_to_send
	168	printf "\n\n+++ SUBJOB killed \n\n"
	169	exit
	170	' 2
	171
	172
	173
	174
[1289]	175	# DETERMINE NAME OF LOCAL HOST
[1]	176	local_host=$(hostname)
	177
	178
	179
[1289]	180	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	181	# IS ADMITTED AT ALL)
	182	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	183	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	184	case $local_host in
[1468]	185	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
	186	(atmos) local_address=172.20.25.35; local_host=lcide;;
	187	(austru) local_address=130.75.105.128; local_host=lcmuk;;
	188	(autan) local_address=130.75.105.57; local_host=lcmuk;;
	189	(bora) local_address=130.75.105.103; local_host=lcmuk;;
	190	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
[1545]	191	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
	192	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
[1468]	193	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
	194	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
	195	(breva) local_address=130.75.105.98; local_host=lcmuk;;
	196	(buran) local_address=130.75.105.58; local_host=lcmuk;;
	197	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
	198	(climate*) local_address=165.132.26.68; local_host=lcyon;;
	199	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
	200	(cs*) local_address=136.172.44.131; local_host=nech;;
	201	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
	202	(flow01) local_address=10.141.255.71; local_host=lcflow;;
	203	(flow02) local_address=10.141.255.72; local_host=lcflow;;
	204	(node*) local_address=165.132.26.61 local_host=lck;;
	205	# (node*) local_address=210.219.61.8 local_host=lckordi;;
	206	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
	207	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
	208	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
	209	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
	210	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
	211	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
	212	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
	213	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
	214	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
	215	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
	216	(sno) local_address=130.75.105.113; local_host=lcmuk;;
	217	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
	218	(login*) local_address=118.128.66.223; local_host=lckiaps;;
	219	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
	220	(meller) local_address=134.106.74.155; local_host=lcfor;;
	221	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
	222	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
	223	(nobel*) local_address=150.183.5.101; local_host=ibms;;
	224	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
	225	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
	226	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
	227	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
	228	(pingui) local_address=134.106.74.118; local_host=lcfor;;
	229	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
	230	(rte*) local_address=133.5.185.60; local_host=lcrte;;
	231	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
	232	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
	233	(solano) local_address=130.75.105.110; local_host=lcmuk;;
	234	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
	235	(t2a*) local_address=10.1.6.165; local_host=lctit;;
	236	(urban*) local_address=147.46.30.151 local_host=lcsb;;
	237	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
	238	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
	239	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
[1]	240	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	241	printf "\n please contact the PALM group at IMUK";
[1]	242	locat=parameter; exit;;
	243	esac
	244
	245
	246
[1289]	247	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	248	remote_host=$local_host
	249
	250
	251
	252
[1289]	253	# READ THE SHELLSCRIPT-OPTIONS
[1094]	254	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	255	do
	256	case $option in
	257	(c) job_catalog=$OPTARG;;
	258	(d) delete_dayfile=true;;
	259	(D) no_submit=true;;
[352]	260	(e) email_notification=$OPTARG;;
[125]	261	(g) group_number=$OPTARG;;
[1]	262	(h) remote_host=$OPTARG;;
	263	(m) memory=$OPTARG;;
	264	(n) job_name=$OPTARG;;
	265	(N) node_usage=$OPTARG;;
	266	(O) threads_per_task=$OPTARG;;
[1094]	267	(P) scp_port=$OPTARG;;
[1]	268	(q) no_default_queue=$OPTARG;;
	269	(t) cputime=$OPTARG;;
	270	(T) tasks_per_node=$OPTARG;;
	271	(u) remote_user=$OPTARG;;
	272	(v) verify=false;;
	273	(X) numprocs=$OPTARG;;
	274	(\?) printf "\n +++ Option $OPTARG unknown \n";
	275	locat=option; exit;;
	276	esac
	277	done
	278
	279
[1289]	280	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	281	(( to_shift = $OPTIND - 1 ))
	282	shift $to_shift; file_to_send=$1
[1]	283
	284
[1289]	285	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	286	if [ "$1" = "?" ]
	287	then
	288	(printf "\n *** subjob can be called as follows:\n"
	289	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	290	printf "\n Description of available options:\n"
	291	printf "\n Option Description Default-Value"
	292	printf "\n -c job-input- and output-catalog ~/job_queue"
	293	printf "\n -d no job-protocol will be created ---"
	294	printf "\n -D only the job-file will be created ---"
	295	printf "\n -h execution host, available hosts: $remote_host"
[1289]	296	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	297	printf "\n lckiaps, lctit, nech"
[1]	298	printf "\n -m memory demand per process in MByte ---"
	299	printf "\n -n jobname <jobdatei>"
	300	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	301	printf "\n -P ssh/scp port default port"
[1]	302	printf "\n -q job-queue to be used default"
	303	printf "\n -t allowed cpu-time in seconds ---"
	304	printf "\n -T tasks per node (on parallel hosts) ---"
	305	printf "\n -u username on execution host from .netrc"
	306	printf "\n -v no prompt for confirmation ---"
	307	printf "\n -X # of processors (on parallel hosts) 1"
	308	printf "\n "
	309	printf "\n The only possible positional parameter is <jobfile>:"
	310	printf "\n The complete NQS-job must be provided here."
	311	printf "\n <jobfile>=? creates this outline\n\n") \| more
	312	exit
	313	fi
	314
	315
	316
[1289]	317	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	318	if [[ "$file_to_send" = "" ]]
	319	then
	320	printf "\n +++ job-file missing"
	321	locat=parameter; exit
	322	else
	323	if [[ -f $file_to_send ]]
	324	then
	325	true
	326	else
	327	printf "\n +++ job-file: "
	328	printf "\n $file_to_send"
	329	printf "\n does not exist"
	330	locat=parameter; exit
	331	fi
	332	fi
	333
	334
	335
[1289]	336	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	337	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	338	if [[ $job_name = none ]]
	339	then
	340	job_name=$file_to_send
	341	fi
	342	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	343	then
	344	printf "\n +++ job-file name: "
	345	printf "\n $job_name"
	346	printf "\n must not contain \"/\"-characters"
	347	locat=parameter; exit
	348	fi
	349
	350
	351
	352
[1289]	353	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	354	# OR IF NO HOST HAS BEEN GIVEN
[1]	355	if [[ $remote_host = none ]]
	356	then
	357	printf "\n +++ host missing"
	358	locat=option; exit
	359	else
	360	case $remote_host in
[1468]	361	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	362	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	363	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	364	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
	365	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1480]	366	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
	367	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
[1468]	368	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
[1547]	369	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
[1468]	370	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
	371	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
	372	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
	373	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
	374	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
	375	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
	376	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
	377	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
	378	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
	379	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
	380	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
	381	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
[251]	382	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	383	locat=parameter; exit;;
[1]	384	esac
	385	fi
	386
	387
[1289]	388	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	389	if [[ $no_default_queue != none ]]
	390	then
	391	error=false
	392	ndq=$no_default_queue
	393	case $remote_host in
	394	(ibm) case $ndq in
	395	(p690_express\|p690_standard\|p690_long) error=false;;
	396	(*) error=true;;
	397	esac;;
	398	(ibmh) case $ndq in
[1184]	399	(cluster\|express) error=false;;
[1]	400	(*) error=true;;
	401	esac;;
[693]	402	(ibmkisti) case $ndq in
	403	(class.32plus\|class.1-2\|class.2-32) error=false;;
	404	(*) error=true;;
	405	esac;;
[622]	406	(ibmku) case $ndq in
	407	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	408	(*) error=true;;
	409	esac;;
[1]	410	(ibms) case $ndq in
	411	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	412	(*) error=true;;
	413	esac;;
[1224]	414	(lccrayb) case $ndq in
[1480]	415	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1224]	416	(*) error=true;;
	417	esac;;
[1274]	418	(lccrayh) case $ndq in
[1442]	419	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1274]	420	(*) error=true;;
	421	esac;;
[1197]	422	(lccrayf) case $ndq in
	423	(usup\|test*\|small\|large) error=false;;
	424	(*) error=true;;
	425	esac;;
[1099]	426	(lcflow) case $ndq in
[1547]	427	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q\|cfd_ivy_shrt.q) error=false;;
[1099]	428	(*) error=true;;
	429	esac;;
[1040]	430	(lckiaps) case $ndq in
	431	(express\|normal) error=false;;
	432	(*) error=true;;
	433	esac;;
[440]	434	(lckyoto) case $ndq in
	435	(eh\|ph) error=false;;
	436	(*) error=true;;
	437	esac;;
[1099]	438	(lckyuh) case $ndq in
	439	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	440	(*) error=true;;
	441	esac;;
[1090]	442	(lckyut) case $ndq in
	443	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	444	(*) error=true;;
	445	esac;;
[1]	446	(lctit) case $ndq in
[635]	447	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	448	(*) error=true;;
	449	esac;;
	450	(t3eb) case $ndq in
	451	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	452	(*) error=true;;
	453	esac;;
	454	(t3eh) case $ndq in
	455	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	456	(*) error=true;;
	457	esac;;
	458	(t3ej2\|t3ej5) case $ndq in
	459	(low\|normal\|high) error=false;;
	460	(*) error=true;;
	461	esac;;
	462	(t3es) case $ndq in
	463	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	464	(*) error=true;;
	465	esac;;
	466	esac
	467	if [[ $error = true ]]
	468	then
	469	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	470	locat=parameter; exit
	471	else
	472	queue=$no_default_queue
	473	fi
	474	fi
	475
	476
	477
[1289]	478	# CHECK THE CPU-TIME
	479	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	480	done=false
	481	while [[ $done = false ]]
	482	do
	483	if (( $cputime <= 0 ))
	484	then
	485	printf "\n +++ wrong cpu-time or cpu-time missing"
	486	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	487	printf "\n >>> "
	488	read cputime 1>/dev/null 2>&1
	489	else
	490	done=true
	491	fi
	492	done
	493	if [[ $remote_host = nech ]]
	494	then
	495	if (( tasks_per_node != 0 ))
	496	then
	497	(( cputime = cputime * tasks_per_node ))
	498	elif [[ $numprocs != 0 ]]
	499	then
	500	(( cputime = cputime * numprocs ))
	501	fi
	502	fi
	503	(( stunden = cputime / 3600 ))
	504	(( resttime = cputime - stunden * 3600 ))
	505	(( minuten = resttime / 60 ))
	506	(( sekunden = resttime - minuten * 60 ))
	507	timestring=${stunden}:${minuten}:${sekunden}
	508
	509
	510
[1289]	511	# CHECK THE MEMORY DEMAND
[1]	512	done=false
	513	while [[ $done = false ]]
	514	do
	515	if (( memory <= 0 ))
	516	then
	517	printf "\n +++ wrong memory demand or memory demand missing"
	518	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	519	printf "\n >>> "
	520	read memory 1>/dev/null 2>&1
	521	else
	522	done=true
	523	fi
	524	done
	525
[1289]	526	if [[ $remote_host = nech ]]
[1]	527	then
	528	if (( tasks_per_node != 0 ))
	529	then
	530	(( Memory = memory * tasks_per_node / 1000 ))
	531	elif [[ $numprocs != 0 ]]
	532	then
	533	(( Memory = memory * numprocs / 1000 ))
	534	else
	535	(( Memory = memory / 1000 ))
	536	fi
[635]	537	elif [[ $remote_host = lctit ]]
	538	then
	539	(( Memory = memory * tasks_per_node / 1000 ))
[1]	540	fi
	541
	542
[1289]	543	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	544	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	545	then
	546	(( memory = memory * threads_per_task ))
	547	fi
	548
	549
[1289]	550	# CALCULATE NUMBER OF REQUIRED NODES
[1]	551	if (( tasks_per_node != 0 ))
	552	then
[1279]	553	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	554	fi
	555
[1094]	556
[1289]	557	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	558	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	559
[1094]	560
[1289]	561	# CALCULATE NUMBER OF MPI TASKS
[696]	562	(( mpi_tasks = numprocs / threads_per_task ))
[1]	563
[696]	564
[1289]	565	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	566	if [[ "$scp_port" != "" ]]
	567	then
	568	PORTOPT="-P $scp_port"
	569	SSH_PORTOPT="-p $scp_port"
	570	fi
	571
	572
[1289]	573	# HEADER-OUTPUT
[1]	574	if [[ $verify = true ]]
	575	then
	576	printf "\n\n"
	577	printf "#--------------------------------------------------------------# \n"
	578	spalte1=SUBJOB;spalte2=$(date)
[1103]	579	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	580	printf "\| \| \n"
	581	printf "\| values of parameters/options: \| \n"
[1103]	582	spalte1=$(echo local_host$punkte \| cut -c-20)
	583	spalte2=$punkte$local_host
	584	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	585	spalte1=$(echo remote_host$punkte \| cut -c-20)
	586	spalte2=$punkte$remote_host
	587	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	588	spalte1=$(echo queue$punkte \| cut -c-20)
	589	spalte2=$punkte$queue
	590	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	591	spalte1=$(echo memory$punkte \| cut -c-20)
	592	spalte2="$punkte$memory mb"
	593	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	594	spalte1=$(echo cputime$punkte \| cut -c-20)
	595	spalte2="$punkte$cputime sec"
	596	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	597	spalte1=$(echo job_name$punkte \| cut -c-20)
	598	spalte2="$punkte$job_name"
	599	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	600	printf "#--------------------------------------------------------------# \n\n"
	601
	602
[1289]	603	# QUERY CHECK
[1]	604	antwort="dummy"
	605	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	606	do
	607	read antwort?" >>> continue (y/n) ? "
	608	done
	609	if [[ $antwort = n \|\| $antwort = N ]]
	610	then
	611	locat=verify; exit
	612	fi
	613	printf "\n"
	614	fi
	615
	616
	617
[1289]	618	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	619	identifier=$RANDOM
	620	job_on_remhost=${job_name}_${identifier}_$local_host
	621	job_to_send=job_to_send_$identifier
[1]	622	if [[ $delete_dayfile = false ]]
	623	then
[1289]	624	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	625	local_dayfile=${remote_host}_${job_name}
	626	else
	627	remote_dayfile=/dev/null
	628	fi
	629
	630
	631
[1289]	632	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	633	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	634	then
	635
[1289]	636	# GENERAL LOADLEVELER SETTINGS
[622]	637	execute_in_shell="#!/bin/ksh"
	638	use_shell="# @ shell = /bin/ksh"
	639	consumable_memory="ConsumableMemory($memory mb)"
	640	class="# @ class = $queue"
	641	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	642	network_to_use="# @ network.mpi = sn_all,shared,us"
	643	data_limit="# @ data_limit = 1.76gb"
	644	image_size="# @ image_size = 50"
[693]	645	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	646
[693]	647	if [[ $email_notification = none ]]
	648	then
	649	notify_user=""
	650	else
	651	notify_user="# @ notify_user = $email_notification"
	652	if [[ $delete_dayfile = true ]]
	653	then
	654	notification='# @ notification = never'
	655	fi
	656	fi
[622]	657
[312]	658	if [[ $remote_host = ibmh ]]
[1]	659	then
[312]	660	data_limit=""
	661	network_to_use=""
[1184]	662	class="# @ class = $queue"
[312]	663	environment=""
[814]	664	rset="# @ rset = RSET_MCM_AFFINITY"
	665	task_affinity="# @ task_affinity = core(1)"
[693]	666	elif [[ $remote_host = ibmkisti ]]
	667	then
	668	network_to_use="# @ network.MPI = sn_all,shared,US"
	669	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	670	if [[ $threads_per_task = 1 ]]
	671	then
	672	rset="# @ rset = RSET_MCM_AFFINITY"
	673	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	674	fi
[693]	675	environment=""
	676	use_shell=""
	677	data_limit=""
	678	image_size=""
[622]	679	elif [[ $remote_host = ibmku ]]
	680	then
	681	execute_in_shell="#!/usr/bin/ksh"
	682	use_shell="# @ shell = /usr/bin/ksh"
	683	consumable_memory=""
	684	environment=""
	685	network_to_use="# @ network.mpi = sn_all,shared,us"
	686	data_limit=""
	687	image_size=""
	688	elif [[ $remote_host = ibms ]]
	689	then
	690	network_to_use="# @ network.mpi = csss,shared,us"
[1]	691	fi
	692
	693	cat > $job_to_send << %%END%%
[622]	694	$execute_in_shell
	695	$use_shell
[1]	696
	697	# @ job_type = parallel
[693]	698	# @ job_name = $job_name
[1]	699	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	700	# @ output = $remote_dayfile
	701	# @ error = $remote_dayfile
[693]	702	$wall_clock_limit
[622]	703	$image_size
[312]	704	$class
	705	$environment
[1]	706	$network_to_use
	707	$data_limit
[693]	708	$rset
	709	$mcm_affinity_options
[814]	710	$task_affinity
[1]	711	$notification
[693]	712	$notify_user
[1]	713
	714	%%END%%
	715
	716	if (( nodes > 0 ))
	717	then
	718
[693]	719	if [[ $remote_host != ibmkisti ]]
	720	then
	721
	722	cat >> $job_to_send << %%END%%
[1]	723	# @ node = $nodes
[622]	724	# @ tasks_per_node = $processes_per_node
[1]	725	# @ node_usage = $node_usage
	726	# @ queue
	727
	728	%%END%%
	729
[693]	730	else
	731
	732	cat >> $job_to_send << %%END%%
[696]	733	# @ total_tasks = $mpi_tasks
[693]	734	# @ blocking = unlimited
	735	# @ queue
	736
	737	%%END%%
	738
	739	fi
	740
[1]	741	else
	742
[1289]	743	cat >> $job_to_send << %%END%%
[1]	744	# @ blocking = unlimited
	745	# @ total_tasks = $numprocs
	746	# @ node_usage = $node_usage
	747	# @ queue
	748
	749	%%END%%
	750
	751	fi
	752
[1289]	753	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	754	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	755	then
	756	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	757	fi
	758
[1]	759	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	760	then
	761
	762	cat > $job_to_send << %%END%%
	763	#!/bin/ksh
	764
	765	# @ job_type = serial
	766	# @ node_usage = $node_usage
	767	# @ job_name = palm
	768	# @ wall_clock_limit = ${timestring},$timestring
	769	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	770	# @ output = $remote_dayfile
	771	# @ error = $remote_dayfile
[312]	772	$class
[1]	773	$notification
	774
	775	# @ queue
	776
	777	%%END%%
	778
[1274]	779	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	780	then
	781
[1480]	782	if [[ "$feature" != "" ]]
	783	then
	784	featuredir="#PBS -l feature=$feature"
	785	fi
	786
[1224]	787	if [[ $numprocs != 0 ]]
	788	then
	789	cat > $job_to_send << %%END%%
[1255]	790	#!/bin/bash
[1224]	791	#PBS -N $job_name
	792	#PBS -l walltime=$timestring
[1264]	793	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	794	#PBS -o $remote_dayfile
	795	#PBS -j oe
	796	#PBS -q $queue
[1480]	797	$featuredir
[1224]	798
	799	$module_calls
	800
	801	%%END%%
	802
	803	else
	804
	805	continue
	806
	807	fi
	808
[1197]	809	elif [[ $remote_host = lccrayf ]]
	810	then
	811
	812	if [[ $numprocs != 0 ]]
	813	then
	814	cat > $job_to_send << %%END%%
	815	#!/bin/bash -l
	816	#SBATCH -J $job_name
	817	#SBATCH -t $timestring
	818	#SBATCH -N $nodes
	819	#SBATCH --ntasks-per-node=$processes_per_node
	820	#SBATCH -p $queue
	821	#SBATCH -o $remote_dayfile
	822	#SBATCH -e $remote_dayfile
	823
	824	$init_cmds
	825	$module_calls
	826
	827	%%END%%
	828
	829	else
	830	cat > $job_to_send << %%END%%
	831	#!/bin/bash -l
	832	#SBATCH -J $job_name
	833	#SBATCH -t $timestring
	834	#SBATCH -l ncpus=1
	835	#SBATCH -l pmem=${memory}mb
	836	#SBATCH -m abe
	837	#SBATCH -o $remote_dayfile
	838	#SBATCH -e $remote_dayfile
	839
	840	$init_cmds
	841	$module_calls
	842
	843	%%END%%
	844
	845	fi
	846
[1099]	847	elif [[ $remote_host = lcflow ]]
	848	then
	849	if [[ $numprocs != 0 ]]
	850	then
[1547]	851	pe_set="#$ -pe impi $numprocs"
[1099]	852	else
[1547]	853	pe_set="#$ -pe impi 1"
[1099]	854	fi
	855	if [[ $queue = default ]]
	856	then
	857	queue_set=""
	858	else
	859	queue_set="#$ -q $queue"
	860	fi
	861	[[ "$disc_space" = "" ]] && disc_space=50
	862
	863	cat > $job_to_send << %%END%%
	864	#!/bin/bash
	865	#$ -S /bin/bash
	866	#$ -N $job_name
	867	#$ -cwd
	868	#$ -l h_rt=$timestring
	869	#$ -l h_vmem=${memory}M
	870	#$ -o $remote_dayfile
	871	#$ -j y
	872	$pe_set
	873	#$ -R y
	874	#$ -l h_fsize=${disc_space}G
	875	$queue_set
	876
	877	%%END%%
	878
[1021]	879	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	880	then
	881
	882	if [[ $numprocs != 0 ]]
	883	then
	884	cat > $job_to_send << %%END%%
	885	#!/bin/ksh
	886	#PBS -N $job_name
	887	#PBS -l walltime=$timestring
	888	#PBS -l ncpus=$numprocs
	889	#PBS -l pmem=${memory}mb
	890	#PBS -o $remote_dayfile
[1021]	891	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	892	#PBS -j oe
	893
	894	mpd &
	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
[1040]	912	elif [[ $remote_host = lckiaps ]]
	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 select=1:ncpus=$numprocs
	922	#PBS -l pmem=${memory}mb
	923	#PBS -q $queue
	924	#PBS -o $remote_dayfile
	925	#PBS -j oe
	926	#PBS -V
	927
	928	%%END%%
	929
	930	else
	931	cat > $job_to_send << %%END%%
	932	#!/bin/ksh
	933	#PBS -N $job_name
	934	#PBS -l walltime=$timestring
	935	#PBS -l ncpus=1
	936	#PBS -l pmem=${memory}mb
	937	#PBS -o $remote_dayfile
	938	#PBS -j oe
	939
	940	%%END%%
	941
	942	fi
	943
[693]	944	elif [[ $remote_host = lcyon ]]
	945	then
	946
	947	if [[ $numprocs != 0 ]]
	948	then
	949	cat > $job_to_send << %%END%%
	950	#!/bin/ksh
	951	#PBS -N $job_name
	952	#PBS -l walltime=$timestring
	953	#PBS -l ncpus=$numprocs
	954	#PBS -l pmem=${memory}mb
	955	#PBS -o $remote_dayfile
	956	#PBS -j oe
	957
	958	%%END%%
	959
	960	else
	961	cat > $job_to_send << %%END%%
	962	#!/bin/ksh
	963	#PBS -N $job_name
	964	#PBS -l walltime=$timestring
	965	#PBS -l ncpus=1
	966	#PBS -l pmem=${memory}mb
	967	#PBS -o $remote_dayfile
	968	#PBS -j oe
	969
	970	%%END%%
	971
	972	fi
	973
[892]	974	elif [[ $remote_host = lcxe6 ]]
[164]	975	then
	976
	977	if [[ $numprocs != 0 ]]
	978	then
	979	cat > $job_to_send << %%END%%
[799]	980	#!/bin/ksh
[164]	981	#PBS -S /bin/ksh
	982	#PBS -N $job_name
[552]	983	#PBS -A $project_account
[206]	984	#PBS -j oe
[164]	985	#PBS -l walltime=$timestring
	986	#PBS -l mppwidth=${numprocs}
[622]	987	#PBS -l mppnppn=${processes_per_node}
[164]	988	#PBS -m abe
	989	#PBS -o $remote_dayfile
[492]	990	$email_directive
[164]	991
[892]	992	$init_cmds
[493]	993	$module_calls
[343]	994
[164]	995	%%END%%
	996
[108]	997	else
	998	cat > $job_to_send << %%END%%
	999	#!/bin/ksh
[168]	1000	#PBS -S /bin/ksh
[108]	1001	#PBS -N $job_name
[552]	1002	#PBS -A $project_account
[206]	1003	#PBS -j oe
[108]	1004	#PBS -l walltime=$timestring
	1005	#PBS -l ncpus=1
	1006	#PBS -l pmem=${memory}mb
	1007	#PBS -m abe
[492]	1008	$email_directive
[108]	1009	#PBS -o $remote_dayfile
	1010
[892]	1011	$init_cmds
[493]	1012	$module_calls
[343]	1013
[108]	1014	%%END%%
	1015
	1016	fi
	1017
[440]	1018	elif [[ $remote_host = lckyoto ]]
[437]	1019	then
	1020
[440]	1021	cat > $job_to_send << %%END%%
[799]	1022	#!/bin/ksh
[440]	1023	# @\$-o $remote_dayfile
	1024	# @\$-eo -oi
	1025	# @\$-lP 16
[799]	1026	# @\$-lp 1
[440]	1027	# @\$-lm 28gb -llm unlimited -ls unlimited
	1028	# @\$-q $queue
	1029	# @\$-Pvn abs_pack
	1030	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1031	#. /thin/local/etc/setprofile/intel-11.0.sh
	1032	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1033	. ~/.myprofile
	1034	#. /home2/t/t51254/palm/current_version/myprofile
	1035	#. /thin/apps/pgi/mpi.sh
	1036	#
	1037	env
	1038	#
	1039	set -x
	1040
	1041	%%END%%
	1042
	1043	elif [[ $remote_host = lcxt5m ]]
	1044	then
	1045
[437]	1046	if [[ $numprocs != 0 ]]
	1047	then
	1048	cat > $job_to_send << %%END%%
[799]	1049	#!/bin/ksh
[437]	1050	#PBS -S /bin/ksh
	1051	#PBS -N $job_name
	1052	#PBS -j oe
	1053	#PBS -l walltime=$timestring
	1054	#PBS -l mppwidth=${numprocs}
[622]	1055	#PBS -l mppnppn=${processes_per_node}
[437]	1056	#PBS -m abe
	1057	#PBS -o $remote_dayfile
	1058
[892]	1059	$init_cmds
[493]	1060	$module_calls
[437]	1061
	1062	%%END%%
	1063
	1064	else
	1065	cat > $job_to_send << %%END%%
	1066	#!/bin/ksh
	1067	#PBS -S /bin/ksh
	1068	#PBS -N $job_name
	1069	#PBS -j oe
	1070	#PBS -l walltime=$timestring
	1071	#PBS -l ncpus=1
	1072	#PBS -l pmem=${memory}mb
	1073	#PBS -m abe
	1074	#PBS -o $remote_dayfile
	1075
[892]	1076	$init_cmds
[493]	1077	$module_calls
[437]	1078
	1079	%%END%%
	1080
	1081	fi
	1082
[1099]	1083	elif [[ $remote_host = lckyuh ]]
	1084	then
	1085	cat > $job_to_send << %%END%%
	1086	#!/bin/bash
	1087	#PJM -L "rscgrp=$queue"
	1088	#PJM -L "node=$nodes"
	1089	#PJM --mpi "proc=$numprocs"
	1090	#PJM -L "elapse=$timestring"
	1091	#PJM -o $remote_dayfile
	1092	#PJM -j
	1093	#PJM -X
	1094	#PJM --no-stging
	1095
	1096	export LANG=en_US.UTF-8
	1097	%%END%%
	1098
[1090]	1099	elif [[ $remote_host = lckyut ]]
	1100	then
	1101	cat > $job_to_send << %%END%%
	1102	#!/bin/bash
[1099]	1103	#PJM -L "rscgrp=$queue"
[1090]	1104	#PJM -L "vnode=$numprocs"
	1105	#PJM -L "vnode-core=1"
	1106	#PJM -L "elapse=$timestring"
	1107	#PJM --mpi proc=$numprocs
	1108	#PJM -o $remote_dayfile
	1109	#PJM -j
[1099]	1110	#PJM -X
	1111	#PJM --no-stging
[1090]	1112
	1113	export LANG=en_US.UTF-8
	1114	%%END%%
	1115
[1]	1116	elif [[ $remote_host = nech ]]
	1117	then
	1118
	1119	if (( nodes > 1 ))
	1120	then
	1121	cat > $job_to_send << %%END%%
	1122	#!/bin/ksh
[622]	1123	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1124	#PBS -l ${qsubmem}=${Memory}gb
	1125	#PBS -b $nodes
	1126	#PBS -o $remote_dayfile
	1127	#PBS -N palm
	1128	#PBS -j o
	1129	#PBS -T mpisx
	1130
	1131	%%END%%
	1132
	1133	elif [[ $numprocs != 0 ]]
	1134	then
	1135	cat > $job_to_send << %%END%%
	1136	#!/bin/ksh
[622]	1137	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1138	#PBS -l ${qsubmem}=${Memory}gb
	1139	#PBS -o $remote_dayfile
	1140	#PBS -N palm
	1141	#PBS -j o
	1142
	1143	%%END%%
	1144
	1145	else
	1146	cat > $job_to_send << %%END%%
	1147	#!/bin/ksh
	1148	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1149	#PBS -o $remote_dayfile
	1150	#PBS -j o
	1151
	1152	%%END%%
	1153
	1154	fi
	1155
	1156	elif [[ $remote_host = lctit ]]
	1157	then
	1158	cat > $job_to_send << %%END%%
[635]	1159	#!/bin/ksh
[892]	1160	$init_cmds
[678]	1161	$module_calls
	1162
[1]	1163	%%END%%
	1164
[1289]	1165	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1166	if [[ $tasks_per_node != $processes_per_node ]]
	1167	then
[1378]	1168	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -o $remote_dayfile -j oe -et 1 -q $queue "
[678]	1169	else
[1378]	1170	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -et 1 -q $queue "
[678]	1171	fi
[1]	1172
	1173	else
	1174
	1175	cat > $job_to_send << %%END%%
	1176	# @\$-q ${queue}
	1177	# @\$-l${qsubtime} $timestring
	1178	# @\$-l${qsubmem} ${memory}mb
	1179	# @\$-o $remote_dayfile
	1180	# @\$-eo
	1181
	1182	%%END%%
	1183
	1184	fi
	1185
	1186
[1289]	1187	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1188	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1189	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1190	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1191	then
	1192	echo "set +vx" >> $job_to_send
	1193	echo "trap '" >> $job_to_send
	1194	echo "set +vx" >> $job_to_send
[1468]	1195	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* \|\| $remote_host = lcxe6 ]]
[1]	1196	then
[622]	1197	if [[ $remote_host = ibmh ]]
[1]	1198	then
	1199	return_queue=c1
[693]	1200	elif [[ $remote_host = ibmkisti ]]
	1201	then
	1202	return_queue=class.1-2
[622]	1203	elif [[ $remote_host = ibmku ]]
	1204	then
	1205	return_queue=sdbg2
[1]	1206	elif [[ $remote_host = ibms ]]
	1207	then
	1208	return_queue=p_normal
[1274]	1209	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1210	then
	1211	return_queue=dataq
[1468]	1212	elif [[ $remote_host = lcxe6 ]]
	1213	then
	1214	return_queue=debug
[1040]	1215	elif [[ $remote_host = lckiaps ]]
	1216	then
	1217	return_queue=express
[1099]	1218	elif [[ $remote_host = lckyuh ]]
	1219	then
	1220	return_queue=cx-single
[1090]	1221	elif [[ $remote_host = lckyut ]]
	1222	then
	1223	return_queue=cx-single
[1]	1224	else
	1225	return_queue=unknown
	1226	fi
	1227
	1228	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1229	then
	1230
[622]	1231	if [[ $remote_host = ibmku ]]
	1232	then
[1289]	1233	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1234	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1235	else
[1289]	1236	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1237	fi
[1289]	1238	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1239	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1240	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1241	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1242	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1243	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1244	if [[ $host != "ibmh" ]]
	1245	then
[1289]	1246	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1247	fi
[1289]	1248	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1249	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1250
[1289]	1251	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1252	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1253
[1289]	1254	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
[1468]	1255	echo "echo \"batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$identifier" >> $job_to_send
[622]	1256	if [[ $remote_host = ibmku ]]
	1257	then
[1289]	1258	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1259	fi
[1289]	1260	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1261
	1262	elif [[ $remote_host = nech ]]
	1263	then
	1264	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1265	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1266	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1267	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1268	echo "#PBS -j o" >> $job_to_send
	1269	echo " " >> $job_to_send
	1270	echo "set -x" >> $job_to_send
	1271	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1468]	1272	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1273	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1274	echo "%%END%%" >> $job_to_send
	1275
[1099]	1276	elif [[ $remote_host = lckyuh ]]
	1277	then
[1289]	1278	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1279	echo "#!/bin/bash" >> $job_to_send
	1280	echo "#PJM -L \"node=1\"" >> $job_to_send
	1281	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1282	echo "#PJM --no-stging" >> $job_to_send
	1283	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1284	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1285	echo "#PJM -j" >> $job_to_send
	1286	echo " " >> $job_to_send
	1287	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1288	echo "set -x" >> $job_to_send
[1468]	1289	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1099]	1290	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1291	echo "%%END%%" >> $job_to_send
	1292
[1090]	1293	elif [[ $remote_host = lckyut ]]
	1294	then
[1289]	1295	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1296	echo "#!/bin/bash" >> $job_to_send
	1297	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1298	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1299	echo "#PJM --no-stging" >> $job_to_send
	1300	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1301	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1302	echo "#PJM -j" >> $job_to_send
	1303	echo " " >> $job_to_send
	1304	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1305	echo "set -x" >> $job_to_send
[1468]	1306	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1307	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1308	echo "%%END%%" >> $job_to_send
	1309
[1274]	1310	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1311	then
[1289]	1312	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1313	echo "#!/bin/bash" >> $job_to_send
	1314	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1315	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1316	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1317	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1318	echo "#PBS -j oe" >> $job_to_send
	1319	echo " " >> $job_to_send
	1320	echo "set -x" >> $job_to_send
[1468]	1321	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1255]	1322	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1323	echo "%%END%%" >> $job_to_send
	1324
[1099]	1325	elif [[ $remote_host = lcflow ]]
	1326	then
[1289]	1327	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1328	echo "#!/bin/bash" >> $job_to_send
	1329	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1330	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1331	echo "SGEPREFIX -cwd" >> $job_to_send
	1332	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1333	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
[1099]	1334	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1335	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1336	echo " " >> $job_to_send
	1337	echo "set -x" >> $job_to_send
	1338	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1339	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1340	echo "" >> $job_to_send
[1468]	1341	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1099]	1342	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1343	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1344	echo "%%END%%" >> $job_to_send
[1289]	1345	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1346	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1468]	1347	elif [[ $remote_host = lcxe6 ]]
	1348	then
	1349	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
	1350	echo "#!/bin/ksh" >> $job_to_send
	1351	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1352	echo "#PBS -l walltime=00:30:00" >> $job_to_send
	1353	echo "#PBS -A $project_account" >> $job_to_send
	1354	echo "#PBS -l mppwidth=1" >> $job_to_send
	1355	echo "#PBS -l mppnppn=1" >> $job_to_send
	1356	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1357	echo "#PBS -j oe" >> $job_to_send
	1358	echo " " >> $job_to_send
	1359	echo "set -x" >> $job_to_send
	1360	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1361	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1362	echo "%%END%%" >> $job_to_send
[1]	1363	else
	1364
[1289]	1365	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1366	echo "# @\\\$-q $return_queue" >> $job_to_send
	1367	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1368	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1369	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1370	then
	1371	echo "# @\$-l mpp_p=0" >> $job_to_send
	1372	fi
	1373	echo '# @\$-lF 10mb' >> $job_to_send
	1374	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1375	echo '# @\\\$-eo' >> $job_to_send
	1376	echo " " >> $job_to_send
	1377	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1378	then
	1379	echo "set +vx" >> $job_to_send
	1380	echo ". .profile" >> $job_to_send
	1381	fi
	1382	echo "set -x" >> $job_to_send
[1468]	1383	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
[1]	1384	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1385	echo "%%END%%" >> $job_to_send
	1386	fi
	1387
	1388	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1389	then
[1289]	1390	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1391	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1392	then
[1289]	1393	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1394	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1395	then
[1289]	1396	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1397	elif [[ $remote_host = t3es ]]
	1398	then
[1289]	1399	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1400	elif [[ $remote_host = lckiaps ]]
	1401	then
[1289]	1402	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
[1468]	1403	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
[1289]	1404	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1405	elif [[ $remote_host = lckyu* ]]
[1090]	1406	then
[1468]	1407	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
	1408	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1409	elif [[ $remote_host = lcflow ]]
	1410	then
[1289]	1411	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
[1468]	1412	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1413	else
[1289]	1414	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1415	fi
[1043]	1416	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1417	then
[1289]	1418	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1419	fi
[1]	1420	if [[ $remote_host = nech ]]
	1421	then
	1422	echo "cd -" >> $job_to_send
	1423	fi
	1424	else
[1468]	1425	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1426	# ??? funktioniert das ÃŒberhaupt noch ???
[1468]	1427	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
[1]	1428	fi
	1429	echo "set -x" >> $job_to_send
	1430	echo " ' exit" >> $job_to_send
	1431	echo "set -x" >> $job_to_send
	1432	fi
	1433
	1434
	1435
[1289]	1436	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1437	cat $file_to_send >> $job_to_send
[69]	1438
[1]	1439	if [[ $remote_host = ibm ]]
	1440	then
	1441	echo " " >> $job_to_send
	1442	echo "exit" >> $job_to_send
	1443	fi
[635]	1444
[1289]	1445	# REMOVE JOB-FILE
[1099]	1446	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1447	then
	1448	echo " " >> $job_to_send
	1449	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1450	fi
[1]	1451
	1452
[1289]	1453	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1454	if [[ $no_submit = false ]]
	1455	then
	1456	if [[ $remote_host != $local_host ]]
	1457	then
	1458	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1459	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1460	then
[82]	1461	job_catalog_save=$job_catalog
	1462	job_catalog=job_queue
	1463	elif [[ $remote_host = nech ]]
	1464	then
	1465	job_catalog_save=$job_catalog
	1466	job_catalog=/hpf/b/${remote_user}/job_queue
	1467	fi
[1096]	1468	if [[ $remote_host = nech ]]
[82]	1469	then
[1289]	1470	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1471	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1472	else
[1468]	1473	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
[1]	1474	fi
[82]	1475	if [[ $? = 1 ]]
	1476	then
	1477	locat=scp; exit
	1478	fi
	1479	if [[ $remote_host = ibms ]]
	1480	then
	1481	job_catalog=$job_catalog_save
	1482	fi
[1]	1483	[[ $verify = true ]] && printf "\n >>> finished\n"
	1484	else
	1485	eval job_catalog=$job_catalog
	1486	cp $job_to_send ${job_catalog}/$job_on_remhost
	1487	fi
	1488
	1489
	1490
[1289]	1491	# START NQS- / LOADLEVELER-JOB
[1]	1492	if [[ $remote_host != $local_host ]]
	1493	then
	1494	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1495
[1289]	1496	if [[ $remote_host = ibmku ]]
[1]	1497	then
[1468]	1498	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1499	elif [[ $remote_host = lcflow ]]
	1500	then
[1468]	1501	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1502	else
[1468]	1503	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1504	fi
[1]	1505
	1506	[[ $verify = true ]] && printf " >>> o.k.\n"
	1507	else
	1508	cd $job_catalog
[1289]	1509	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1510	then
	1511	eval $submcom $job_on_remhost
[1289]	1512	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1513	then
[635]	1514	chmod u+x $job_on_remhost
[108]	1515	eval $submcom $job_on_remhost
[1]	1516	elif [[ $local_host = nech ]]
	1517	then
	1518	if [[ $queue = default ]]
	1519	then
[799]	1520	eval $submcom $job_on_remhost
[1]	1521	else
[799]	1522	eval $submcom -q $queue $job_on_remhost
[1]	1523	fi
	1524	else
	1525	qsub $job_on_remhost
	1526	fi
[622]	1527
[1289]	1528	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1529	# AT THE END OF THE JOB
[1099]	1530	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1531	then
	1532	rm $job_on_remhost
	1533	fi
[1]	1534	cd - > /dev/null
	1535	fi
	1536	fi
	1537
	1538
	1539
[1289]	1540	# FINAL ACTIONS
[1]	1541	if [[ $no_submit = false ]]
	1542	then
[1099]	1543	rm -f $job_to_send
[1]	1544	fi
[1266]	1545	[[ $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 |