Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1586

Last change on this file since 1586 was 1576, checked in by raasch, 10 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 52.4 KB

Rev	Line
[1]	1	#!/bin/ksh
[1090]	2
	3	# subjob - script for automatic generation and submission of batch-job files
	4	# for various batch queuing systems
	5
[1046]	6	#--------------------------------------------------------------------------------#
	7	# This file is part of PALM.
	8	#
	9	# PALM is free software: you can redistribute it and/or modify it under the terms
	10	# of the GNU General Public License as published by the Free Software Foundation,
	11	# either version 3 of the License, or (at your option) any later version.
	12	#
	13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
	14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
	15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
	16	#
	17	# You should have received a copy of the GNU General Public License along with
	18	# PALM. If not, see <http://www.gnu.org/licenses/>.
	19	#
[1310]	20	# Copyright 1997-2014 Leibniz Universitaet Hannover
[1046]	21	#--------------------------------------------------------------------------------#
	22	#
	23	# Current revisions:
[1090]	24	# ------------------
[1351]	25	#
[1576]	26	#
[1046]	27	# Former revisions:
	28	# -----------------
[169]	29	# $Id: subjob 1576 2015-03-27 10:23:30Z maronga $
[1200]	30	#
[1576]	31	# 1575 2015-03-27 09:56:27Z raasch
	32	# mpp2-queues added to lccrayh
	33	#
[1548]	34	# 1547 2015-01-29 15:09:12Z witha
	35	# adjustments for ForWind computing cluster (lcflow)
	36	#
[1546]	37	# 1545 2015-01-29 06:52:23Z heinze
	38	# local host name for blizzard further specified
	39	#
[1481]	40	# 1480 2014-10-17 14:41:49Z raasch
	41	# adjustments for 2nd stage of HLRNIII
	42	#
[1469]	43	# 1468 2014-09-24 14:06:57Z maronga
	44	# Typo removed (addres->address)
	45	# Adjustments for lcxe6
	46	#
[1453]	47	# 1452 2014-08-22 09:41:06Z heinze
	48	# local hosts for blizzard added
	49	#
[1451]	50	# 1450 2014-08-21 07:31:51Z heinze
	51	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
	52	#
[1443]	53	# 1442 2014-07-28 07:09:10Z raasch
	54	# HLRN-III (lccrayb/lccrayh) queues adjusted
	55	#
[1379]	56	# 1378 2014-04-28 06:04:58Z raasch
	57	# -et option added for lctit
	58	#
[1351]	59	# 1350 2014-04-04 13:01:30Z maronga
	60	# location of qsub updated for lcxe6
	61	#
[1290]	62	# 1289 2014-03-04 07:12:34Z raasch
	63	# German comments translated to English
	64	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
	65	#
[1280]	66	# 1279 2014-01-28 12:10:14Z raasch
	67	# node calculation modified due to changes in mrun (tasks_per_node must not be
	68	# an integral divisor of numprocs any more)
	69	#
[1275]	70	# 1274 2014-01-09 13:14:54Z heinze
	71	# adjustments for lccrayh
	72	#
[1267]	73	# 1266 2013-12-11 12:07:34Z heinze
	74	# further adjustments for lccrayb (use msub instead of qsub)
	75	#
[1265]	76	# 1264 2013-12-09 12:46:09Z fricke
	77	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	78	#
[1263]	79	# 1262 2013-12-09 10:57:20Z fricke
	80	# further adjustments for lccrayb
	81	#
[1261]	82	# 1260 2013-12-04 12:48:04Z raasch
	83	# jaboticaba admitted
	84	#
[1256]	85	# 1255 2013-11-07 14:43:35Z raasch
	86	# further adjustments for lccrayb
	87	#
[1225]	88	# 1224 2013-09-16 07:27:23Z raasch
	89	# first adjustments for lccrayb
	90	#
[1203]	91	# 1202 2013-07-10 16:22:07Z witha
	92	# adjustments for Forwind cluster (lcflow)
	93	#
[1200]	94	# 1199 2013-07-05 14:52:22Z raasch
	95	# adjustments for CSC Helsinki (lccrayf)
	96	#
[1185]	97	# use of cluster/express queue enabled (ibmh)
	98	# vinessa added (imuk)
[1047]	99	#
[1104]	100	# 1103 2013-02-20 02:15:53Z raasch
	101	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	102	# instead typeset -L/R),
	103	# further adjustments for lckyuh
	104	#
[1100]	105	# 2013-02-10 01:47:43Z raasch
	106	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	107	# and for Forwind cluster (lcflow)
	108	#
[1096]	109	# 1094 2013-02-03 01:52:12Z raasch
	110	# new option -P for explicit setting of ssh/scp port,
	111	# decalpha parts (yonsei) removed
	112	#
[1091]	113	# 2013-02-02 07:06:13Z raasch
[1099]	114	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	115	# old changelog messages removed
	116	#
[1047]	117	# 1046 2012-11-09 14:38:45Z maronga
	118	# code put under GPL (PALM 3.9)
	119	#
[1090]	120	# 08/07/94 - Siggi - first version finished
	121	# 29/06/94 - Siggi - script development started
	122	#--------------------------------------------------------------------------------#
	123	# subjob - script for automatic generation and submission of batch-job files
	124	# for various batch queuing systems
	125	#--------------------------------------------------------------------------------#
[1]	126
	127
[1289]	128	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	129	delete_dayfile=false
[799]	130	email_notification=none
[122]	131	group_number=none
[1]	132	locat=normal
	133	no_default_queue=none
	134	no_submit=false
	135	job_catalog="~/job_queue"
	136	job_name=none
	137	local_user=$LOGNAME
	138	node_usage=shared
[475]	139	numprocs=0
[1]	140	punkte="..........................................................."
	141	submcom=qsub
	142	queue=default
	143	remote_host=none
	144	remote_user=""
	145	verify=true
	146
	147	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	148	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	149
	150
	151
[1289]	152	# ERROR HANDLING
	153	# IN CASE OF EXIT:
[1]	154	trap 'if [[ $locat != normal ]]
	155	then
	156	case $locat in
	157	(option) printf "\n --> available optios can be displayed"
	158	printf " by typing:"
	159	printf "\n \"subjob ?\" \n";;
	160	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	161	(*) printf "\n +++ unknown error"
	162	printf "\n please inform S. Raasch!\n"
	163	esac
	164	[[ -f $job_to_send ]] && rm $job_to_send
	165	printf "\n\n+++ SUBJOB killed \n\n"
	166	fi' exit
	167
	168
[1289]	169	# IN CASE OF TERMINAL-BREAK:
[1]	170	trap '[[ -f $job_to_send ]] && rm $job_to_send
	171	printf "\n\n+++ SUBJOB killed \n\n"
	172	exit
	173	' 2
	174
	175
	176
	177
[1289]	178	# DETERMINE NAME OF LOCAL HOST
[1]	179	local_host=$(hostname)
	180
	181
	182
[1289]	183	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	184	# IS ADMITTED AT ALL)
	185	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	186	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	187	case $local_host in
[1468]	188	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
	189	(atmos) local_address=172.20.25.35; local_host=lcide;;
	190	(austru) local_address=130.75.105.128; local_host=lcmuk;;
	191	(autan) local_address=130.75.105.57; local_host=lcmuk;;
	192	(bora) local_address=130.75.105.103; local_host=lcmuk;;
	193	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
[1545]	194	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
	195	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
[1468]	196	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
	197	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
	198	(breva) local_address=130.75.105.98; local_host=lcmuk;;
	199	(buran) local_address=130.75.105.58; local_host=lcmuk;;
	200	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
	201	(climate*) local_address=165.132.26.68; local_host=lcyon;;
	202	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
	203	(cs*) local_address=136.172.44.131; local_host=nech;;
	204	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
	205	(flow01) local_address=10.141.255.71; local_host=lcflow;;
	206	(flow02) local_address=10.141.255.72; local_host=lcflow;;
	207	(node*) local_address=165.132.26.61 local_host=lck;;
	208	# (node*) local_address=210.219.61.8 local_host=lckordi;;
	209	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
	210	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
	211	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
	212	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
	213	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
	214	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
	215	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
	216	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
	217	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
	218	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
	219	(sno) local_address=130.75.105.113; local_host=lcmuk;;
	220	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
	221	(login*) local_address=118.128.66.223; local_host=lckiaps;;
	222	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
	223	(meller) local_address=134.106.74.155; local_host=lcfor;;
	224	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
	225	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
	226	(nobel*) local_address=150.183.5.101; local_host=ibms;;
	227	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
	228	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
	229	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
	230	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
	231	(pingui) local_address=134.106.74.118; local_host=lcfor;;
	232	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
	233	(rte*) local_address=133.5.185.60; local_host=lcrte;;
	234	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
	235	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
	236	(solano) local_address=130.75.105.110; local_host=lcmuk;;
	237	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
	238	(t2a*) local_address=10.1.6.165; local_host=lctit;;
	239	(urban*) local_address=147.46.30.151 local_host=lcsb;;
	240	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
	241	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
	242	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
[1]	243	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	244	printf "\n please contact the PALM group at IMUK";
[1]	245	locat=parameter; exit;;
	246	esac
	247
	248
	249
[1289]	250	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	251	remote_host=$local_host
	252
	253
	254
	255
[1289]	256	# READ THE SHELLSCRIPT-OPTIONS
[1094]	257	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	258	do
	259	case $option in
	260	(c) job_catalog=$OPTARG;;
	261	(d) delete_dayfile=true;;
	262	(D) no_submit=true;;
[352]	263	(e) email_notification=$OPTARG;;
[125]	264	(g) group_number=$OPTARG;;
[1]	265	(h) remote_host=$OPTARG;;
	266	(m) memory=$OPTARG;;
	267	(n) job_name=$OPTARG;;
	268	(N) node_usage=$OPTARG;;
	269	(O) threads_per_task=$OPTARG;;
[1094]	270	(P) scp_port=$OPTARG;;
[1]	271	(q) no_default_queue=$OPTARG;;
	272	(t) cputime=$OPTARG;;
	273	(T) tasks_per_node=$OPTARG;;
	274	(u) remote_user=$OPTARG;;
	275	(v) verify=false;;
	276	(X) numprocs=$OPTARG;;
	277	(\?) printf "\n +++ Option $OPTARG unknown \n";
	278	locat=option; exit;;
	279	esac
	280	done
	281
	282
[1289]	283	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	284	(( to_shift = $OPTIND - 1 ))
	285	shift $to_shift; file_to_send=$1
[1]	286
	287
[1289]	288	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	289	if [ "$1" = "?" ]
	290	then
	291	(printf "\n *** subjob can be called as follows:\n"
	292	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	293	printf "\n Description of available options:\n"
	294	printf "\n Option Description Default-Value"
	295	printf "\n -c job-input- and output-catalog ~/job_queue"
	296	printf "\n -d no job-protocol will be created ---"
	297	printf "\n -D only the job-file will be created ---"
	298	printf "\n -h execution host, available hosts: $remote_host"
[1289]	299	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	300	printf "\n lckiaps, lctit, nech"
[1]	301	printf "\n -m memory demand per process in MByte ---"
	302	printf "\n -n jobname <jobdatei>"
	303	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	304	printf "\n -P ssh/scp port default port"
[1]	305	printf "\n -q job-queue to be used default"
	306	printf "\n -t allowed cpu-time in seconds ---"
	307	printf "\n -T tasks per node (on parallel hosts) ---"
	308	printf "\n -u username on execution host from .netrc"
	309	printf "\n -v no prompt for confirmation ---"
	310	printf "\n -X # of processors (on parallel hosts) 1"
	311	printf "\n "
	312	printf "\n The only possible positional parameter is <jobfile>:"
	313	printf "\n The complete NQS-job must be provided here."
	314	printf "\n <jobfile>=? creates this outline\n\n") \| more
	315	exit
	316	fi
	317
	318
	319
[1289]	320	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	321	if [[ "$file_to_send" = "" ]]
	322	then
	323	printf "\n +++ job-file missing"
	324	locat=parameter; exit
	325	else
	326	if [[ -f $file_to_send ]]
	327	then
	328	true
	329	else
	330	printf "\n +++ job-file: "
	331	printf "\n $file_to_send"
	332	printf "\n does not exist"
	333	locat=parameter; exit
	334	fi
	335	fi
	336
	337
	338
[1289]	339	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	340	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	341	if [[ $job_name = none ]]
	342	then
	343	job_name=$file_to_send
	344	fi
	345	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	346	then
	347	printf "\n +++ job-file name: "
	348	printf "\n $job_name"
	349	printf "\n must not contain \"/\"-characters"
	350	locat=parameter; exit
	351	fi
	352
	353
	354
	355
[1289]	356	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	357	# OR IF NO HOST HAS BEEN GIVEN
[1]	358	if [[ $remote_host = none ]]
	359	then
	360	printf "\n +++ host missing"
	361	locat=option; exit
	362	else
	363	case $remote_host in
[1468]	364	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	365	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	366	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	367	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
	368	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1480]	369	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
	370	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
[1468]	371	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
[1547]	372	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
[1468]	373	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
	374	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
	375	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
	376	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
	377	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
	378	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
	379	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
	380	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
	381	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
	382	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
	383	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
	384	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
[251]	385	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	386	locat=parameter; exit;;
[1]	387	esac
	388	fi
	389
	390
[1289]	391	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	392	if [[ $no_default_queue != none ]]
	393	then
	394	error=false
	395	ndq=$no_default_queue
	396	case $remote_host in
	397	(ibm) case $ndq in
	398	(p690_express\|p690_standard\|p690_long) error=false;;
	399	(*) error=true;;
	400	esac;;
	401	(ibmh) case $ndq in
[1184]	402	(cluster\|express) error=false;;
[1]	403	(*) error=true;;
	404	esac;;
[693]	405	(ibmkisti) case $ndq in
	406	(class.32plus\|class.1-2\|class.2-32) error=false;;
	407	(*) error=true;;
	408	esac;;
[622]	409	(ibmku) case $ndq in
	410	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	411	(*) error=true;;
	412	esac;;
[1]	413	(ibms) case $ndq in
	414	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	415	(*) error=true;;
	416	esac;;
[1224]	417	(lccrayb) case $ndq in
[1480]	418	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1224]	419	(*) error=true;;
	420	esac;;
[1274]	421	(lccrayh) case $ndq in
[1575]	422	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1274]	423	(*) error=true;;
	424	esac;;
[1197]	425	(lccrayf) case $ndq in
	426	(usup\|test*\|small\|large) error=false;;
	427	(*) error=true;;
	428	esac;;
[1099]	429	(lcflow) case $ndq in
[1547]	430	(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]	431	(*) error=true;;
	432	esac;;
[1040]	433	(lckiaps) case $ndq in
	434	(express\|normal) error=false;;
	435	(*) error=true;;
	436	esac;;
[440]	437	(lckyoto) case $ndq in
	438	(eh\|ph) error=false;;
	439	(*) error=true;;
	440	esac;;
[1099]	441	(lckyuh) case $ndq in
	442	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	443	(*) error=true;;
	444	esac;;
[1090]	445	(lckyut) case $ndq in
	446	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	447	(*) error=true;;
	448	esac;;
[1]	449	(lctit) case $ndq in
[635]	450	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	451	(*) error=true;;
	452	esac;;
	453	(t3eb) case $ndq in
	454	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	455	(*) error=true;;
	456	esac;;
	457	(t3eh) case $ndq in
	458	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	459	(*) error=true;;
	460	esac;;
	461	(t3ej2\|t3ej5) case $ndq in
	462	(low\|normal\|high) error=false;;
	463	(*) error=true;;
	464	esac;;
	465	(t3es) case $ndq in
	466	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	467	(*) error=true;;
	468	esac;;
	469	esac
	470	if [[ $error = true ]]
	471	then
	472	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	473	locat=parameter; exit
	474	else
	475	queue=$no_default_queue
	476	fi
	477	fi
	478
	479
	480
[1289]	481	# CHECK THE CPU-TIME
	482	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	483	done=false
	484	while [[ $done = false ]]
	485	do
	486	if (( $cputime <= 0 ))
	487	then
	488	printf "\n +++ wrong cpu-time or cpu-time missing"
	489	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	490	printf "\n >>> "
	491	read cputime 1>/dev/null 2>&1
	492	else
	493	done=true
	494	fi
	495	done
	496	if [[ $remote_host = nech ]]
	497	then
	498	if (( tasks_per_node != 0 ))
	499	then
	500	(( cputime = cputime * tasks_per_node ))
	501	elif [[ $numprocs != 0 ]]
	502	then
	503	(( cputime = cputime * numprocs ))
	504	fi
	505	fi
	506	(( stunden = cputime / 3600 ))
	507	(( resttime = cputime - stunden * 3600 ))
	508	(( minuten = resttime / 60 ))
	509	(( sekunden = resttime - minuten * 60 ))
	510	timestring=${stunden}:${minuten}:${sekunden}
	511
	512
	513
[1289]	514	# CHECK THE MEMORY DEMAND
[1]	515	done=false
	516	while [[ $done = false ]]
	517	do
	518	if (( memory <= 0 ))
	519	then
	520	printf "\n +++ wrong memory demand or memory demand missing"
	521	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	522	printf "\n >>> "
	523	read memory 1>/dev/null 2>&1
	524	else
	525	done=true
	526	fi
	527	done
	528
[1289]	529	if [[ $remote_host = nech ]]
[1]	530	then
	531	if (( tasks_per_node != 0 ))
	532	then
	533	(( Memory = memory * tasks_per_node / 1000 ))
	534	elif [[ $numprocs != 0 ]]
	535	then
	536	(( Memory = memory * numprocs / 1000 ))
	537	else
	538	(( Memory = memory / 1000 ))
	539	fi
[635]	540	elif [[ $remote_host = lctit ]]
	541	then
	542	(( Memory = memory * tasks_per_node / 1000 ))
[1]	543	fi
	544
	545
[1289]	546	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	547	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	548	then
	549	(( memory = memory * threads_per_task ))
	550	fi
	551
	552
[1289]	553	# CALCULATE NUMBER OF REQUIRED NODES
[1]	554	if (( tasks_per_node != 0 ))
	555	then
[1279]	556	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	557	fi
	558
[1094]	559
[1289]	560	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	561	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	562
[1094]	563
[1289]	564	# CALCULATE NUMBER OF MPI TASKS
[696]	565	(( mpi_tasks = numprocs / threads_per_task ))
[1]	566
[696]	567
[1289]	568	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	569	if [[ "$scp_port" != "" ]]
	570	then
	571	PORTOPT="-P $scp_port"
	572	SSH_PORTOPT="-p $scp_port"
	573	fi
	574
	575
[1289]	576	# HEADER-OUTPUT
[1]	577	if [[ $verify = true ]]
	578	then
	579	printf "\n\n"
	580	printf "#--------------------------------------------------------------# \n"
	581	spalte1=SUBJOB;spalte2=$(date)
[1103]	582	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	583	printf "\| \| \n"
	584	printf "\| values of parameters/options: \| \n"
[1103]	585	spalte1=$(echo local_host$punkte \| cut -c-20)
	586	spalte2=$punkte$local_host
	587	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	588	spalte1=$(echo remote_host$punkte \| cut -c-20)
	589	spalte2=$punkte$remote_host
	590	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	591	spalte1=$(echo queue$punkte \| cut -c-20)
	592	spalte2=$punkte$queue
	593	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	594	spalte1=$(echo memory$punkte \| cut -c-20)
	595	spalte2="$punkte$memory mb"
	596	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	597	spalte1=$(echo cputime$punkte \| cut -c-20)
	598	spalte2="$punkte$cputime sec"
	599	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	600	spalte1=$(echo job_name$punkte \| cut -c-20)
	601	spalte2="$punkte$job_name"
	602	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	603	printf "#--------------------------------------------------------------# \n\n"
	604
	605
[1289]	606	# QUERY CHECK
[1]	607	antwort="dummy"
	608	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	609	do
	610	read antwort?" >>> continue (y/n) ? "
	611	done
	612	if [[ $antwort = n \|\| $antwort = N ]]
	613	then
	614	locat=verify; exit
	615	fi
	616	printf "\n"
	617	fi
	618
	619
	620
[1289]	621	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	622	identifier=$RANDOM
	623	job_on_remhost=${job_name}_${identifier}_$local_host
	624	job_to_send=job_to_send_$identifier
[1]	625	if [[ $delete_dayfile = false ]]
	626	then
[1289]	627	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	628	local_dayfile=${remote_host}_${job_name}
	629	else
	630	remote_dayfile=/dev/null
	631	fi
	632
	633
	634
[1289]	635	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	636	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	637	then
	638
[1289]	639	# GENERAL LOADLEVELER SETTINGS
[622]	640	execute_in_shell="#!/bin/ksh"
	641	use_shell="# @ shell = /bin/ksh"
	642	consumable_memory="ConsumableMemory($memory mb)"
	643	class="# @ class = $queue"
	644	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	645	network_to_use="# @ network.mpi = sn_all,shared,us"
	646	data_limit="# @ data_limit = 1.76gb"
	647	image_size="# @ image_size = 50"
[693]	648	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	649
[693]	650	if [[ $email_notification = none ]]
	651	then
	652	notify_user=""
	653	else
	654	notify_user="# @ notify_user = $email_notification"
	655	if [[ $delete_dayfile = true ]]
	656	then
	657	notification='# @ notification = never'
	658	fi
	659	fi
[622]	660
[312]	661	if [[ $remote_host = ibmh ]]
[1]	662	then
[312]	663	data_limit=""
	664	network_to_use=""
[1184]	665	class="# @ class = $queue"
[312]	666	environment=""
[814]	667	rset="# @ rset = RSET_MCM_AFFINITY"
	668	task_affinity="# @ task_affinity = core(1)"
[693]	669	elif [[ $remote_host = ibmkisti ]]
	670	then
	671	network_to_use="# @ network.MPI = sn_all,shared,US"
	672	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	673	if [[ $threads_per_task = 1 ]]
	674	then
	675	rset="# @ rset = RSET_MCM_AFFINITY"
	676	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	677	fi
[693]	678	environment=""
	679	use_shell=""
	680	data_limit=""
	681	image_size=""
[622]	682	elif [[ $remote_host = ibmku ]]
	683	then
	684	execute_in_shell="#!/usr/bin/ksh"
	685	use_shell="# @ shell = /usr/bin/ksh"
	686	consumable_memory=""
	687	environment=""
	688	network_to_use="# @ network.mpi = sn_all,shared,us"
	689	data_limit=""
	690	image_size=""
	691	elif [[ $remote_host = ibms ]]
	692	then
	693	network_to_use="# @ network.mpi = csss,shared,us"
[1]	694	fi
	695
	696	cat > $job_to_send << %%END%%
[622]	697	$execute_in_shell
	698	$use_shell
[1]	699
	700	# @ job_type = parallel
[693]	701	# @ job_name = $job_name
[1]	702	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	703	# @ output = $remote_dayfile
	704	# @ error = $remote_dayfile
[693]	705	$wall_clock_limit
[622]	706	$image_size
[312]	707	$class
	708	$environment
[1]	709	$network_to_use
	710	$data_limit
[693]	711	$rset
	712	$mcm_affinity_options
[814]	713	$task_affinity
[1]	714	$notification
[693]	715	$notify_user
[1]	716
	717	%%END%%
	718
	719	if (( nodes > 0 ))
	720	then
	721
[693]	722	if [[ $remote_host != ibmkisti ]]
	723	then
	724
	725	cat >> $job_to_send << %%END%%
[1]	726	# @ node = $nodes
[622]	727	# @ tasks_per_node = $processes_per_node
[1]	728	# @ node_usage = $node_usage
	729	# @ queue
	730
	731	%%END%%
	732
[693]	733	else
	734
	735	cat >> $job_to_send << %%END%%
[696]	736	# @ total_tasks = $mpi_tasks
[693]	737	# @ blocking = unlimited
	738	# @ queue
	739
	740	%%END%%
	741
	742	fi
	743
[1]	744	else
	745
[1289]	746	cat >> $job_to_send << %%END%%
[1]	747	# @ blocking = unlimited
	748	# @ total_tasks = $numprocs
	749	# @ node_usage = $node_usage
	750	# @ queue
	751
	752	%%END%%
	753
	754	fi
	755
[1289]	756	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	757	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	758	then
	759	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	760	fi
	761
[1]	762	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	763	then
	764
	765	cat > $job_to_send << %%END%%
	766	#!/bin/ksh
	767
	768	# @ job_type = serial
	769	# @ node_usage = $node_usage
	770	# @ job_name = palm
	771	# @ wall_clock_limit = ${timestring},$timestring
	772	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	773	# @ output = $remote_dayfile
	774	# @ error = $remote_dayfile
[312]	775	$class
[1]	776	$notification
	777
	778	# @ queue
	779
	780	%%END%%
	781
[1274]	782	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	783	then
	784
[1480]	785	if [[ "$feature" != "" ]]
	786	then
	787	featuredir="#PBS -l feature=$feature"
	788	fi
	789
[1224]	790	if [[ $numprocs != 0 ]]
	791	then
	792	cat > $job_to_send << %%END%%
[1255]	793	#!/bin/bash
[1224]	794	#PBS -N $job_name
	795	#PBS -l walltime=$timestring
[1264]	796	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	797	#PBS -o $remote_dayfile
	798	#PBS -j oe
	799	#PBS -q $queue
[1480]	800	$featuredir
[1224]	801
	802	$module_calls
	803
	804	%%END%%
	805
	806	else
	807
	808	continue
	809
	810	fi
	811
[1197]	812	elif [[ $remote_host = lccrayf ]]
	813	then
	814
	815	if [[ $numprocs != 0 ]]
	816	then
	817	cat > $job_to_send << %%END%%
	818	#!/bin/bash -l
	819	#SBATCH -J $job_name
	820	#SBATCH -t $timestring
	821	#SBATCH -N $nodes
	822	#SBATCH --ntasks-per-node=$processes_per_node
	823	#SBATCH -p $queue
	824	#SBATCH -o $remote_dayfile
	825	#SBATCH -e $remote_dayfile
	826
	827	$init_cmds
	828	$module_calls
	829
	830	%%END%%
	831
	832	else
	833	cat > $job_to_send << %%END%%
	834	#!/bin/bash -l
	835	#SBATCH -J $job_name
	836	#SBATCH -t $timestring
	837	#SBATCH -l ncpus=1
	838	#SBATCH -l pmem=${memory}mb
	839	#SBATCH -m abe
	840	#SBATCH -o $remote_dayfile
	841	#SBATCH -e $remote_dayfile
	842
	843	$init_cmds
	844	$module_calls
	845
	846	%%END%%
	847
	848	fi
	849
[1099]	850	elif [[ $remote_host = lcflow ]]
	851	then
	852	if [[ $numprocs != 0 ]]
	853	then
[1547]	854	pe_set="#$ -pe impi $numprocs"
[1099]	855	else
[1547]	856	pe_set="#$ -pe impi 1"
[1099]	857	fi
	858	if [[ $queue = default ]]
	859	then
	860	queue_set=""
	861	else
	862	queue_set="#$ -q $queue"
	863	fi
	864	[[ "$disc_space" = "" ]] && disc_space=50
	865
	866	cat > $job_to_send << %%END%%
	867	#!/bin/bash
	868	#$ -S /bin/bash
	869	#$ -N $job_name
	870	#$ -cwd
	871	#$ -l h_rt=$timestring
	872	#$ -l h_vmem=${memory}M
	873	#$ -o $remote_dayfile
	874	#$ -j y
	875	$pe_set
	876	#$ -R y
	877	#$ -l h_fsize=${disc_space}G
	878	$queue_set
	879
	880	%%END%%
	881
[1021]	882	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	883	then
	884
	885	if [[ $numprocs != 0 ]]
	886	then
	887	cat > $job_to_send << %%END%%
	888	#!/bin/ksh
	889	#PBS -N $job_name
	890	#PBS -l walltime=$timestring
	891	#PBS -l ncpus=$numprocs
	892	#PBS -l pmem=${memory}mb
	893	#PBS -o $remote_dayfile
[1021]	894	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	895	#PBS -j oe
	896
	897	mpd &
	898
	899	%%END%%
	900
	901	else
	902	cat > $job_to_send << %%END%%
	903	#!/bin/ksh
	904	#PBS -N $job_name
	905	#PBS -l walltime=$timestring
	906	#PBS -l ncpus=1
	907	#PBS -l pmem=${memory}mb
	908	#PBS -o $remote_dayfile
	909	#PBS -j oe
	910
	911	%%END%%
	912
	913	fi
	914
[1040]	915	elif [[ $remote_host = lckiaps ]]
	916	then
	917
	918	if [[ $numprocs != 0 ]]
	919	then
	920	cat > $job_to_send << %%END%%
	921	#!/bin/ksh
	922	#PBS -N $job_name
	923	#PBS -l walltime=$timestring
	924	#PBS -l select=1:ncpus=$numprocs
	925	#PBS -l pmem=${memory}mb
	926	#PBS -q $queue
	927	#PBS -o $remote_dayfile
	928	#PBS -j oe
	929	#PBS -V
	930
	931	%%END%%
	932
	933	else
	934	cat > $job_to_send << %%END%%
	935	#!/bin/ksh
	936	#PBS -N $job_name
	937	#PBS -l walltime=$timestring
	938	#PBS -l ncpus=1
	939	#PBS -l pmem=${memory}mb
	940	#PBS -o $remote_dayfile
	941	#PBS -j oe
	942
	943	%%END%%
	944
	945	fi
	946
[693]	947	elif [[ $remote_host = lcyon ]]
	948	then
	949
	950	if [[ $numprocs != 0 ]]
	951	then
	952	cat > $job_to_send << %%END%%
	953	#!/bin/ksh
	954	#PBS -N $job_name
	955	#PBS -l walltime=$timestring
	956	#PBS -l ncpus=$numprocs
	957	#PBS -l pmem=${memory}mb
	958	#PBS -o $remote_dayfile
	959	#PBS -j oe
	960
	961	%%END%%
	962
	963	else
	964	cat > $job_to_send << %%END%%
	965	#!/bin/ksh
	966	#PBS -N $job_name
	967	#PBS -l walltime=$timestring
	968	#PBS -l ncpus=1
	969	#PBS -l pmem=${memory}mb
	970	#PBS -o $remote_dayfile
	971	#PBS -j oe
	972
	973	%%END%%
	974
	975	fi
	976
[892]	977	elif [[ $remote_host = lcxe6 ]]
[164]	978	then
	979
	980	if [[ $numprocs != 0 ]]
	981	then
	982	cat > $job_to_send << %%END%%
[799]	983	#!/bin/ksh
[164]	984	#PBS -S /bin/ksh
	985	#PBS -N $job_name
[552]	986	#PBS -A $project_account
[206]	987	#PBS -j oe
[164]	988	#PBS -l walltime=$timestring
	989	#PBS -l mppwidth=${numprocs}
[622]	990	#PBS -l mppnppn=${processes_per_node}
[164]	991	#PBS -m abe
	992	#PBS -o $remote_dayfile
[492]	993	$email_directive
[164]	994
[892]	995	$init_cmds
[493]	996	$module_calls
[343]	997
[164]	998	%%END%%
	999
[108]	1000	else
	1001	cat > $job_to_send << %%END%%
	1002	#!/bin/ksh
[168]	1003	#PBS -S /bin/ksh
[108]	1004	#PBS -N $job_name
[552]	1005	#PBS -A $project_account
[206]	1006	#PBS -j oe
[108]	1007	#PBS -l walltime=$timestring
	1008	#PBS -l ncpus=1
	1009	#PBS -l pmem=${memory}mb
	1010	#PBS -m abe
[492]	1011	$email_directive
[108]	1012	#PBS -o $remote_dayfile
	1013
[892]	1014	$init_cmds
[493]	1015	$module_calls
[343]	1016
[108]	1017	%%END%%
	1018
	1019	fi
	1020
[440]	1021	elif [[ $remote_host = lckyoto ]]
[437]	1022	then
	1023
[440]	1024	cat > $job_to_send << %%END%%
[799]	1025	#!/bin/ksh
[440]	1026	# @\$-o $remote_dayfile
	1027	# @\$-eo -oi
	1028	# @\$-lP 16
[799]	1029	# @\$-lp 1
[440]	1030	# @\$-lm 28gb -llm unlimited -ls unlimited
	1031	# @\$-q $queue
	1032	# @\$-Pvn abs_pack
	1033	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1034	#. /thin/local/etc/setprofile/intel-11.0.sh
	1035	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1036	. ~/.myprofile
	1037	#. /home2/t/t51254/palm/current_version/myprofile
	1038	#. /thin/apps/pgi/mpi.sh
	1039	#
	1040	env
	1041	#
	1042	set -x
	1043
	1044	%%END%%
	1045
	1046	elif [[ $remote_host = lcxt5m ]]
	1047	then
	1048
[437]	1049	if [[ $numprocs != 0 ]]
	1050	then
	1051	cat > $job_to_send << %%END%%
[799]	1052	#!/bin/ksh
[437]	1053	#PBS -S /bin/ksh
	1054	#PBS -N $job_name
	1055	#PBS -j oe
	1056	#PBS -l walltime=$timestring
	1057	#PBS -l mppwidth=${numprocs}
[622]	1058	#PBS -l mppnppn=${processes_per_node}
[437]	1059	#PBS -m abe
	1060	#PBS -o $remote_dayfile
	1061
[892]	1062	$init_cmds
[493]	1063	$module_calls
[437]	1064
	1065	%%END%%
	1066
	1067	else
	1068	cat > $job_to_send << %%END%%
	1069	#!/bin/ksh
	1070	#PBS -S /bin/ksh
	1071	#PBS -N $job_name
	1072	#PBS -j oe
	1073	#PBS -l walltime=$timestring
	1074	#PBS -l ncpus=1
	1075	#PBS -l pmem=${memory}mb
	1076	#PBS -m abe
	1077	#PBS -o $remote_dayfile
	1078
[892]	1079	$init_cmds
[493]	1080	$module_calls
[437]	1081
	1082	%%END%%
	1083
	1084	fi
	1085
[1099]	1086	elif [[ $remote_host = lckyuh ]]
	1087	then
	1088	cat > $job_to_send << %%END%%
	1089	#!/bin/bash
	1090	#PJM -L "rscgrp=$queue"
	1091	#PJM -L "node=$nodes"
	1092	#PJM --mpi "proc=$numprocs"
	1093	#PJM -L "elapse=$timestring"
	1094	#PJM -o $remote_dayfile
	1095	#PJM -j
	1096	#PJM -X
	1097	#PJM --no-stging
	1098
	1099	export LANG=en_US.UTF-8
	1100	%%END%%
	1101
[1090]	1102	elif [[ $remote_host = lckyut ]]
	1103	then
	1104	cat > $job_to_send << %%END%%
	1105	#!/bin/bash
[1099]	1106	#PJM -L "rscgrp=$queue"
[1090]	1107	#PJM -L "vnode=$numprocs"
	1108	#PJM -L "vnode-core=1"
	1109	#PJM -L "elapse=$timestring"
	1110	#PJM --mpi proc=$numprocs
	1111	#PJM -o $remote_dayfile
	1112	#PJM -j
[1099]	1113	#PJM -X
	1114	#PJM --no-stging
[1090]	1115
	1116	export LANG=en_US.UTF-8
	1117	%%END%%
	1118
[1]	1119	elif [[ $remote_host = nech ]]
	1120	then
	1121
	1122	if (( nodes > 1 ))
	1123	then
	1124	cat > $job_to_send << %%END%%
	1125	#!/bin/ksh
[622]	1126	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1127	#PBS -l ${qsubmem}=${Memory}gb
	1128	#PBS -b $nodes
	1129	#PBS -o $remote_dayfile
	1130	#PBS -N palm
	1131	#PBS -j o
	1132	#PBS -T mpisx
	1133
	1134	%%END%%
	1135
	1136	elif [[ $numprocs != 0 ]]
	1137	then
	1138	cat > $job_to_send << %%END%%
	1139	#!/bin/ksh
[622]	1140	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1141	#PBS -l ${qsubmem}=${Memory}gb
	1142	#PBS -o $remote_dayfile
	1143	#PBS -N palm
	1144	#PBS -j o
	1145
	1146	%%END%%
	1147
	1148	else
	1149	cat > $job_to_send << %%END%%
	1150	#!/bin/ksh
	1151	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1152	#PBS -o $remote_dayfile
	1153	#PBS -j o
	1154
	1155	%%END%%
	1156
	1157	fi
	1158
	1159	elif [[ $remote_host = lctit ]]
	1160	then
	1161	cat > $job_to_send << %%END%%
[635]	1162	#!/bin/ksh
[892]	1163	$init_cmds
[678]	1164	$module_calls
	1165
[1]	1166	%%END%%
	1167
[1289]	1168	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1169	if [[ $tasks_per_node != $processes_per_node ]]
	1170	then
[1378]	1171	submcom="$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 $queue "
[678]	1172	else
[1378]	1173	submcom="$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]	1174	fi
[1]	1175
	1176	else
	1177
	1178	cat > $job_to_send << %%END%%
	1179	# @\$-q ${queue}
	1180	# @\$-l${qsubtime} $timestring
	1181	# @\$-l${qsubmem} ${memory}mb
	1182	# @\$-o $remote_dayfile
	1183	# @\$-eo
	1184
	1185	%%END%%
	1186
	1187	fi
	1188
	1189
[1289]	1190	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1191	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1192	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1193	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1194	then
	1195	echo "set +vx" >> $job_to_send
	1196	echo "trap '" >> $job_to_send
	1197	echo "set +vx" >> $job_to_send
[1468]	1198	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]	1199	then
[622]	1200	if [[ $remote_host = ibmh ]]
[1]	1201	then
	1202	return_queue=c1
[693]	1203	elif [[ $remote_host = ibmkisti ]]
	1204	then
	1205	return_queue=class.1-2
[622]	1206	elif [[ $remote_host = ibmku ]]
	1207	then
	1208	return_queue=sdbg2
[1]	1209	elif [[ $remote_host = ibms ]]
	1210	then
	1211	return_queue=p_normal
[1274]	1212	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1213	then
	1214	return_queue=dataq
[1468]	1215	elif [[ $remote_host = lcxe6 ]]
	1216	then
	1217	return_queue=debug
[1040]	1218	elif [[ $remote_host = lckiaps ]]
	1219	then
	1220	return_queue=express
[1099]	1221	elif [[ $remote_host = lckyuh ]]
	1222	then
	1223	return_queue=cx-single
[1090]	1224	elif [[ $remote_host = lckyut ]]
	1225	then
	1226	return_queue=cx-single
[1]	1227	else
	1228	return_queue=unknown
	1229	fi
	1230
	1231	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1232	then
	1233
[622]	1234	if [[ $remote_host = ibmku ]]
	1235	then
[1289]	1236	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1237	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1238	else
[1289]	1239	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1240	fi
[1289]	1241	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1242	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1243	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1244	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1245	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1246	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1247	if [[ $host != "ibmh" ]]
	1248	then
[1289]	1249	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1250	fi
[1289]	1251	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1252	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1253
[1289]	1254	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1255	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1256
[1289]	1257	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
[1468]	1258	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]	1259	if [[ $remote_host = ibmku ]]
	1260	then
[1289]	1261	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1262	fi
[1289]	1263	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1264
	1265	elif [[ $remote_host = nech ]]
	1266	then
	1267	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1268	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1269	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1270	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1271	echo "#PBS -j o" >> $job_to_send
	1272	echo " " >> $job_to_send
	1273	echo "set -x" >> $job_to_send
	1274	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1468]	1275	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1276	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1277	echo "%%END%%" >> $job_to_send
	1278
[1099]	1279	elif [[ $remote_host = lckyuh ]]
	1280	then
[1289]	1281	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1282	echo "#!/bin/bash" >> $job_to_send
	1283	echo "#PJM -L \"node=1\"" >> $job_to_send
	1284	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1285	echo "#PJM --no-stging" >> $job_to_send
	1286	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1287	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1288	echo "#PJM -j" >> $job_to_send
	1289	echo " " >> $job_to_send
	1290	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1291	echo "set -x" >> $job_to_send
[1468]	1292	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1099]	1293	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1294	echo "%%END%%" >> $job_to_send
	1295
[1090]	1296	elif [[ $remote_host = lckyut ]]
	1297	then
[1289]	1298	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1299	echo "#!/bin/bash" >> $job_to_send
	1300	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1301	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1302	echo "#PJM --no-stging" >> $job_to_send
	1303	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1304	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1305	echo "#PJM -j" >> $job_to_send
	1306	echo " " >> $job_to_send
	1307	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1308	echo "set -x" >> $job_to_send
[1468]	1309	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1310	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1311	echo "%%END%%" >> $job_to_send
	1312
[1274]	1313	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1314	then
[1289]	1315	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1316	echo "#!/bin/bash" >> $job_to_send
	1317	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1318	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1319	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1320	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1321	echo "#PBS -j oe" >> $job_to_send
	1322	echo " " >> $job_to_send
	1323	echo "set -x" >> $job_to_send
[1468]	1324	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1255]	1325	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1326	echo "%%END%%" >> $job_to_send
	1327
[1099]	1328	elif [[ $remote_host = lcflow ]]
	1329	then
[1289]	1330	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1331	echo "#!/bin/bash" >> $job_to_send
	1332	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1333	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1334	echo "SGEPREFIX -cwd" >> $job_to_send
	1335	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1336	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
[1099]	1337	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1338	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1339	echo " " >> $job_to_send
	1340	echo "set -x" >> $job_to_send
	1341	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1342	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1343	echo "" >> $job_to_send
[1468]	1344	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1099]	1345	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1346	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1347	echo "%%END%%" >> $job_to_send
[1289]	1348	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1349	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1468]	1350	elif [[ $remote_host = lcxe6 ]]
	1351	then
	1352	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
	1353	echo "#!/bin/ksh" >> $job_to_send
	1354	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1355	echo "#PBS -l walltime=00:30:00" >> $job_to_send
	1356	echo "#PBS -A $project_account" >> $job_to_send
	1357	echo "#PBS -l mppwidth=1" >> $job_to_send
	1358	echo "#PBS -l mppnppn=1" >> $job_to_send
	1359	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1360	echo "#PBS -j oe" >> $job_to_send
	1361	echo " " >> $job_to_send
	1362	echo "set -x" >> $job_to_send
	1363	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1364	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1365	echo "%%END%%" >> $job_to_send
[1]	1366	else
	1367
[1289]	1368	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1369	echo "# @\\\$-q $return_queue" >> $job_to_send
	1370	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1371	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1372	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1373	then
	1374	echo "# @\$-l mpp_p=0" >> $job_to_send
	1375	fi
	1376	echo '# @\$-lF 10mb' >> $job_to_send
	1377	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1378	echo '# @\\\$-eo' >> $job_to_send
	1379	echo " " >> $job_to_send
	1380	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1381	then
	1382	echo "set +vx" >> $job_to_send
	1383	echo ". .profile" >> $job_to_send
	1384	fi
	1385	echo "set -x" >> $job_to_send
[1468]	1386	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]	1387	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1388	echo "%%END%%" >> $job_to_send
	1389	fi
	1390
	1391	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1392	then
[1289]	1393	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1394	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1395	then
[1289]	1396	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1397	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1398	then
[1289]	1399	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1400	elif [[ $remote_host = t3es ]]
	1401	then
[1289]	1402	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1403	elif [[ $remote_host = lckiaps ]]
	1404	then
[1289]	1405	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
[1468]	1406	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
[1289]	1407	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1408	elif [[ $remote_host = lckyu* ]]
[1090]	1409	then
[1468]	1410	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
	1411	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1412	elif [[ $remote_host = lcflow ]]
	1413	then
[1289]	1414	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
[1468]	1415	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1416	else
[1289]	1417	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1418	fi
[1043]	1419	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1420	then
[1289]	1421	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1422	fi
[1]	1423	if [[ $remote_host = nech ]]
	1424	then
	1425	echo "cd -" >> $job_to_send
	1426	fi
	1427	else
[1468]	1428	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1429	# ??? funktioniert das ÃŒberhaupt noch ???
[1468]	1430	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
[1]	1431	fi
	1432	echo "set -x" >> $job_to_send
	1433	echo " ' exit" >> $job_to_send
	1434	echo "set -x" >> $job_to_send
	1435	fi
	1436
	1437
	1438
[1289]	1439	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1440	cat $file_to_send >> $job_to_send
[69]	1441
[1]	1442	if [[ $remote_host = ibm ]]
	1443	then
	1444	echo " " >> $job_to_send
	1445	echo "exit" >> $job_to_send
	1446	fi
[635]	1447
[1289]	1448	# REMOVE JOB-FILE
[1099]	1449	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1450	then
	1451	echo " " >> $job_to_send
	1452	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1453	fi
[1]	1454
	1455
[1289]	1456	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1457	if [[ $no_submit = false ]]
	1458	then
	1459	if [[ $remote_host != $local_host ]]
	1460	then
	1461	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1462	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1463	then
[82]	1464	job_catalog_save=$job_catalog
	1465	job_catalog=job_queue
	1466	elif [[ $remote_host = nech ]]
	1467	then
	1468	job_catalog_save=$job_catalog
	1469	job_catalog=/hpf/b/${remote_user}/job_queue
	1470	fi
[1096]	1471	if [[ $remote_host = nech ]]
[82]	1472	then
[1289]	1473	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1474	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1475	else
[1468]	1476	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
[1]	1477	fi
[82]	1478	if [[ $? = 1 ]]
	1479	then
	1480	locat=scp; exit
	1481	fi
	1482	if [[ $remote_host = ibms ]]
	1483	then
	1484	job_catalog=$job_catalog_save
	1485	fi
[1]	1486	[[ $verify = true ]] && printf "\n >>> finished\n"
	1487	else
	1488	eval job_catalog=$job_catalog
	1489	cp $job_to_send ${job_catalog}/$job_on_remhost
	1490	fi
	1491
	1492
	1493
[1289]	1494	# START NQS- / LOADLEVELER-JOB
[1]	1495	if [[ $remote_host != $local_host ]]
	1496	then
	1497	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1498
[1289]	1499	if [[ $remote_host = ibmku ]]
[1]	1500	then
[1468]	1501	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1502	elif [[ $remote_host = lcflow ]]
	1503	then
[1468]	1504	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1505	else
[1468]	1506	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1507	fi
[1]	1508
	1509	[[ $verify = true ]] && printf " >>> o.k.\n"
	1510	else
	1511	cd $job_catalog
[1289]	1512	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1513	then
	1514	eval $submcom $job_on_remhost
[1289]	1515	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1516	then
[635]	1517	chmod u+x $job_on_remhost
[108]	1518	eval $submcom $job_on_remhost
[1]	1519	elif [[ $local_host = nech ]]
	1520	then
	1521	if [[ $queue = default ]]
	1522	then
[799]	1523	eval $submcom $job_on_remhost
[1]	1524	else
[799]	1525	eval $submcom -q $queue $job_on_remhost
[1]	1526	fi
	1527	else
	1528	qsub $job_on_remhost
	1529	fi
[622]	1530
[1289]	1531	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1532	# AT THE END OF THE JOB
[1099]	1533	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1534	then
	1535	rm $job_on_remhost
	1536	fi
[1]	1537	cd - > /dev/null
	1538	fi
	1539	fi
	1540
	1541
	1542
[1289]	1543	# FINAL ACTIONS
[1]	1544	if [[ $no_submit = false ]]
	1545	then
[1099]	1546	rm -f $job_to_send
[1]	1547	fi
[1266]	1548	[[ $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 |