Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1468

Last change on this file since 1468 was 1468, checked in by maronga, 10 years ago
New flag files allow to force unscheduled termination/restarts of batch jobs, progress output is made for batch runs, small adjustments for lxce6 and lccrayh/lccrayb
Property svn:keywords set to `Id Rev`
File size: 52.1 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	# ------------------
[1468]	25	# Typo removed (addres->address)
	26	# Adjustments for lcxe6
[1351]	27	#
[1046]	28	# Former revisions:
	29	# -----------------
[169]	30	# $Id: subjob 1468 2014-09-24 14:06:57Z maronga $
[1200]	31	#
[1453]	32	# 1452 2014-08-22 09:41:06Z heinze
	33	# local hosts for blizzard added
	34	#
[1451]	35	# 1450 2014-08-21 07:31:51Z heinze
	36	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
	37	#
[1443]	38	# 1442 2014-07-28 07:09:10Z raasch
	39	# HLRN-III (lccrayb/lccrayh) queues adjusted
	40	#
[1379]	41	# 1378 2014-04-28 06:04:58Z raasch
	42	# -et option added for lctit
	43	#
[1351]	44	# 1350 2014-04-04 13:01:30Z maronga
	45	# location of qsub updated for lcxe6
	46	#
[1290]	47	# 1289 2014-03-04 07:12:34Z raasch
	48	# German comments translated to English
	49	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
	50	#
[1280]	51	# 1279 2014-01-28 12:10:14Z raasch
	52	# node calculation modified due to changes in mrun (tasks_per_node must not be
	53	# an integral divisor of numprocs any more)
	54	#
[1275]	55	# 1274 2014-01-09 13:14:54Z heinze
	56	# adjustments for lccrayh
	57	#
[1267]	58	# 1266 2013-12-11 12:07:34Z heinze
	59	# further adjustments for lccrayb (use msub instead of qsub)
	60	#
[1265]	61	# 1264 2013-12-09 12:46:09Z fricke
	62	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	63	#
[1263]	64	# 1262 2013-12-09 10:57:20Z fricke
	65	# further adjustments for lccrayb
	66	#
[1261]	67	# 1260 2013-12-04 12:48:04Z raasch
	68	# jaboticaba admitted
	69	#
[1256]	70	# 1255 2013-11-07 14:43:35Z raasch
	71	# further adjustments for lccrayb
	72	#
[1225]	73	# 1224 2013-09-16 07:27:23Z raasch
	74	# first adjustments for lccrayb
	75	#
[1203]	76	# 1202 2013-07-10 16:22:07Z witha
	77	# adjustments for Forwind cluster (lcflow)
	78	#
[1200]	79	# 1199 2013-07-05 14:52:22Z raasch
	80	# adjustments for CSC Helsinki (lccrayf)
	81	#
[1185]	82	# use of cluster/express queue enabled (ibmh)
	83	# vinessa added (imuk)
[1047]	84	#
[1104]	85	# 1103 2013-02-20 02:15:53Z raasch
	86	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	87	# instead typeset -L/R),
	88	# further adjustments for lckyuh
	89	#
[1100]	90	# 2013-02-10 01:47:43Z raasch
	91	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	92	# and for Forwind cluster (lcflow)
	93	#
[1096]	94	# 1094 2013-02-03 01:52:12Z raasch
	95	# new option -P for explicit setting of ssh/scp port,
	96	# decalpha parts (yonsei) removed
	97	#
[1091]	98	# 2013-02-02 07:06:13Z raasch
[1099]	99	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	100	# old changelog messages removed
	101	#
[1047]	102	# 1046 2012-11-09 14:38:45Z maronga
	103	# code put under GPL (PALM 3.9)
	104	#
[1090]	105	# 08/07/94 - Siggi - first version finished
	106	# 29/06/94 - Siggi - script development started
	107	#--------------------------------------------------------------------------------#
	108	# subjob - script for automatic generation and submission of batch-job files
	109	# for various batch queuing systems
	110	#--------------------------------------------------------------------------------#
[1]	111
	112
[1289]	113	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	114	delete_dayfile=false
[799]	115	email_notification=none
[122]	116	group_number=none
[1]	117	locat=normal
	118	no_default_queue=none
	119	no_submit=false
	120	job_catalog="~/job_queue"
	121	job_name=none
	122	local_user=$LOGNAME
	123	node_usage=shared
[475]	124	numprocs=0
[1]	125	punkte="..........................................................."
	126	submcom=qsub
	127	queue=default
	128	remote_host=none
	129	remote_user=""
	130	verify=true
	131
	132	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	133	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	134
	135
	136
[1289]	137	# ERROR HANDLING
	138	# IN CASE OF EXIT:
[1]	139	trap 'if [[ $locat != normal ]]
	140	then
	141	case $locat in
	142	(option) printf "\n --> available optios can be displayed"
	143	printf " by typing:"
	144	printf "\n \"subjob ?\" \n";;
	145	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	146	(*) printf "\n +++ unknown error"
	147	printf "\n please inform S. Raasch!\n"
	148	esac
	149	[[ -f $job_to_send ]] && rm $job_to_send
	150	printf "\n\n+++ SUBJOB killed \n\n"
	151	fi' exit
	152
	153
[1289]	154	# IN CASE OF TERMINAL-BREAK:
[1]	155	trap '[[ -f $job_to_send ]] && rm $job_to_send
	156	printf "\n\n+++ SUBJOB killed \n\n"
	157	exit
	158	' 2
	159
	160
	161
	162
[1289]	163	# DETERMINE NAME OF LOCAL HOST
[1]	164	local_host=$(hostname)
	165
	166
	167
[1289]	168	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	169	# IS ADMITTED AT ALL)
	170	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	171	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	172	case $local_host in
[1468]	173	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
	174	(atmos) local_address=172.20.25.35; local_host=lcide;;
	175	(austru) local_address=130.75.105.128; local_host=lcmuk;;
	176	(autan) local_address=130.75.105.57; local_host=lcmuk;;
	177	(bora) local_address=130.75.105.103; local_host=lcmuk;;
	178	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
	179	(blizzard1\|p*) local_address=136.172.40.15; local_host=ibmh;;
	180	(blizzard2\|p*) local_address=136.172.40.16; local_host=ibmh;;
	181	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
	182	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
	183	(breva) local_address=130.75.105.98; local_host=lcmuk;;
	184	(buran) local_address=130.75.105.58; local_host=lcmuk;;
	185	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
	186	(climate*) local_address=165.132.26.68; local_host=lcyon;;
	187	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
	188	(cs*) local_address=136.172.44.131; local_host=nech;;
	189	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
	190	(flow01) local_address=10.141.255.71; local_host=lcflow;;
	191	(flow02) local_address=10.141.255.72; local_host=lcflow;;
	192	(node*) local_address=165.132.26.61 local_host=lck;;
	193	# (node*) local_address=210.219.61.8 local_host=lckordi;;
	194	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
	195	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
	196	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
	197	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
	198	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
	199	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
	200	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
	201	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
	202	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
	203	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
	204	(sno) local_address=130.75.105.113; local_host=lcmuk;;
	205	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
	206	(login*) local_address=118.128.66.223; local_host=lckiaps;;
	207	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
	208	(meller) local_address=134.106.74.155; local_host=lcfor;;
	209	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
	210	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
	211	(nobel*) local_address=150.183.5.101; local_host=ibms;;
	212	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
	213	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
	214	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
	215	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
	216	(pingui) local_address=134.106.74.118; local_host=lcfor;;
	217	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
	218	(rte*) local_address=133.5.185.60; local_host=lcrte;;
	219	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
	220	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
	221	(solano) local_address=130.75.105.110; local_host=lcmuk;;
	222	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
	223	(t2a*) local_address=10.1.6.165; local_host=lctit;;
	224	(urban*) local_address=147.46.30.151 local_host=lcsb;;
	225	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
	226	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
	227	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
[1]	228	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	229	printf "\n please contact the PALM group at IMUK";
[1]	230	locat=parameter; exit;;
	231	esac
	232
	233
	234
[1289]	235	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	236	remote_host=$local_host
	237
	238
	239
	240
[1289]	241	# READ THE SHELLSCRIPT-OPTIONS
[1094]	242	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	243	do
	244	case $option in
	245	(c) job_catalog=$OPTARG;;
	246	(d) delete_dayfile=true;;
	247	(D) no_submit=true;;
[352]	248	(e) email_notification=$OPTARG;;
[125]	249	(g) group_number=$OPTARG;;
[1]	250	(h) remote_host=$OPTARG;;
	251	(m) memory=$OPTARG;;
	252	(n) job_name=$OPTARG;;
	253	(N) node_usage=$OPTARG;;
	254	(O) threads_per_task=$OPTARG;;
[1094]	255	(P) scp_port=$OPTARG;;
[1]	256	(q) no_default_queue=$OPTARG;;
	257	(t) cputime=$OPTARG;;
	258	(T) tasks_per_node=$OPTARG;;
	259	(u) remote_user=$OPTARG;;
	260	(v) verify=false;;
	261	(X) numprocs=$OPTARG;;
	262	(\?) printf "\n +++ Option $OPTARG unknown \n";
	263	locat=option; exit;;
	264	esac
	265	done
	266
	267
[1289]	268	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	269	(( to_shift = $OPTIND - 1 ))
	270	shift $to_shift; file_to_send=$1
[1]	271
	272
[1289]	273	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	274	if [ "$1" = "?" ]
	275	then
	276	(printf "\n *** subjob can be called as follows:\n"
	277	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	278	printf "\n Description of available options:\n"
	279	printf "\n Option Description Default-Value"
	280	printf "\n -c job-input- and output-catalog ~/job_queue"
	281	printf "\n -d no job-protocol will be created ---"
	282	printf "\n -D only the job-file will be created ---"
	283	printf "\n -h execution host, available hosts: $remote_host"
[1289]	284	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	285	printf "\n lckiaps, lctit, nech"
[1]	286	printf "\n -m memory demand per process in MByte ---"
	287	printf "\n -n jobname <jobdatei>"
	288	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	289	printf "\n -P ssh/scp port default port"
[1]	290	printf "\n -q job-queue to be used default"
	291	printf "\n -t allowed cpu-time in seconds ---"
	292	printf "\n -T tasks per node (on parallel hosts) ---"
	293	printf "\n -u username on execution host from .netrc"
	294	printf "\n -v no prompt for confirmation ---"
	295	printf "\n -X # of processors (on parallel hosts) 1"
	296	printf "\n "
	297	printf "\n The only possible positional parameter is <jobfile>:"
	298	printf "\n The complete NQS-job must be provided here."
	299	printf "\n <jobfile>=? creates this outline\n\n") \| more
	300	exit
	301	fi
	302
	303
	304
[1289]	305	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	306	if [[ "$file_to_send" = "" ]]
	307	then
	308	printf "\n +++ job-file missing"
	309	locat=parameter; exit
	310	else
	311	if [[ -f $file_to_send ]]
	312	then
	313	true
	314	else
	315	printf "\n +++ job-file: "
	316	printf "\n $file_to_send"
	317	printf "\n does not exist"
	318	locat=parameter; exit
	319	fi
	320	fi
	321
	322
	323
[1289]	324	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	325	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	326	if [[ $job_name = none ]]
	327	then
	328	job_name=$file_to_send
	329	fi
	330	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	331	then
	332	printf "\n +++ job-file name: "
	333	printf "\n $job_name"
	334	printf "\n must not contain \"/\"-characters"
	335	locat=parameter; exit
	336	fi
	337
	338
	339
	340
[1289]	341	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	342	# OR IF NO HOST HAS BEEN GIVEN
[1]	343	if [[ $remote_host = none ]]
	344	then
	345	printf "\n +++ host missing"
	346	locat=option; exit
	347	else
	348	case $remote_host in
[1468]	349	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	350	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	351	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	352	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
	353	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	354	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
	355	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/7.2.6/bin/msub;;
	356	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
	357	(lcflow) remote_address=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
	358	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
	359	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
	360	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
	361	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
	362	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
	363	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
	364	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
	365	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
	366	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
	367	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
	368	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
	369	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
[251]	370	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	371	locat=parameter; exit;;
[1]	372	esac
	373	fi
	374
	375
[1289]	376	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	377	if [[ $no_default_queue != none ]]
	378	then
	379	error=false
	380	ndq=$no_default_queue
	381	case $remote_host in
	382	(ibm) case $ndq in
	383	(p690_express\|p690_standard\|p690_long) error=false;;
	384	(*) error=true;;
	385	esac;;
	386	(ibmh) case $ndq in
[1184]	387	(cluster\|express) error=false;;
[1]	388	(*) error=true;;
	389	esac;;
[693]	390	(ibmkisti) case $ndq in
	391	(class.32plus\|class.1-2\|class.2-32) error=false;;
	392	(*) error=true;;
	393	esac;;
[622]	394	(ibmku) case $ndq in
	395	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	396	(*) error=true;;
	397	esac;;
[1]	398	(ibms) case $ndq in
	399	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	400	(*) error=true;;
	401	esac;;
[1224]	402	(lccrayb) case $ndq in
[1442]	403	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1224]	404	(*) error=true;;
	405	esac;;
[1274]	406	(lccrayh) case $ndq in
[1442]	407	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
[1274]	408	(*) error=true;;
	409	esac;;
[1197]	410	(lccrayf) case $ndq in
	411	(usup\|test*\|small\|large) error=false;;
	412	(*) error=true;;
	413	esac;;
[1099]	414	(lcflow) case $ndq in
	415	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
	416	(*) error=true;;
	417	esac;;
[1040]	418	(lckiaps) case $ndq in
	419	(express\|normal) error=false;;
	420	(*) error=true;;
	421	esac;;
[440]	422	(lckyoto) case $ndq in
	423	(eh\|ph) error=false;;
	424	(*) error=true;;
	425	esac;;
[1099]	426	(lckyuh) case $ndq in
	427	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	428	(*) error=true;;
	429	esac;;
[1090]	430	(lckyut) case $ndq in
	431	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	432	(*) error=true;;
	433	esac;;
[1]	434	(lctit) case $ndq in
[635]	435	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	436	(*) error=true;;
	437	esac;;
	438	(t3eb) case $ndq in
	439	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	440	(*) error=true;;
	441	esac;;
	442	(t3eh) case $ndq in
	443	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	444	(*) error=true;;
	445	esac;;
	446	(t3ej2\|t3ej5) case $ndq in
	447	(low\|normal\|high) error=false;;
	448	(*) error=true;;
	449	esac;;
	450	(t3es) case $ndq in
	451	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	452	(*) error=true;;
	453	esac;;
	454	esac
	455	if [[ $error = true ]]
	456	then
	457	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	458	locat=parameter; exit
	459	else
	460	queue=$no_default_queue
	461	fi
	462	fi
	463
	464
	465
[1289]	466	# CHECK THE CPU-TIME
	467	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	468	done=false
	469	while [[ $done = false ]]
	470	do
	471	if (( $cputime <= 0 ))
	472	then
	473	printf "\n +++ wrong cpu-time or cpu-time missing"
	474	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	475	printf "\n >>> "
	476	read cputime 1>/dev/null 2>&1
	477	else
	478	done=true
	479	fi
	480	done
	481	if [[ $remote_host = nech ]]
	482	then
	483	if (( tasks_per_node != 0 ))
	484	then
	485	(( cputime = cputime * tasks_per_node ))
	486	elif [[ $numprocs != 0 ]]
	487	then
	488	(( cputime = cputime * numprocs ))
	489	fi
	490	fi
	491	(( stunden = cputime / 3600 ))
	492	(( resttime = cputime - stunden * 3600 ))
	493	(( minuten = resttime / 60 ))
	494	(( sekunden = resttime - minuten * 60 ))
	495	timestring=${stunden}:${minuten}:${sekunden}
	496
	497
	498
[1289]	499	# CHECK THE MEMORY DEMAND
[1]	500	done=false
	501	while [[ $done = false ]]
	502	do
	503	if (( memory <= 0 ))
	504	then
	505	printf "\n +++ wrong memory demand or memory demand missing"
	506	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	507	printf "\n >>> "
	508	read memory 1>/dev/null 2>&1
	509	else
	510	done=true
	511	fi
	512	done
	513
[1289]	514	if [[ $remote_host = nech ]]
[1]	515	then
	516	if (( tasks_per_node != 0 ))
	517	then
	518	(( Memory = memory * tasks_per_node / 1000 ))
	519	elif [[ $numprocs != 0 ]]
	520	then
	521	(( Memory = memory * numprocs / 1000 ))
	522	else
	523	(( Memory = memory / 1000 ))
	524	fi
[635]	525	elif [[ $remote_host = lctit ]]
	526	then
	527	(( Memory = memory * tasks_per_node / 1000 ))
[1]	528	fi
	529
	530
[1289]	531	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	532	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	533	then
	534	(( memory = memory * threads_per_task ))
	535	fi
	536
	537
[1289]	538	# CALCULATE NUMBER OF REQUIRED NODES
[1]	539	if (( tasks_per_node != 0 ))
	540	then
[1279]	541	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	542	fi
	543
[1094]	544
[1289]	545	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	546	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	547
[1094]	548
[1289]	549	# CALCULATE NUMBER OF MPI TASKS
[696]	550	(( mpi_tasks = numprocs / threads_per_task ))
[1]	551
[696]	552
[1289]	553	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	554	if [[ "$scp_port" != "" ]]
	555	then
	556	PORTOPT="-P $scp_port"
	557	SSH_PORTOPT="-p $scp_port"
	558	fi
	559
	560
[1289]	561	# HEADER-OUTPUT
[1]	562	if [[ $verify = true ]]
	563	then
	564	printf "\n\n"
	565	printf "#--------------------------------------------------------------# \n"
	566	spalte1=SUBJOB;spalte2=$(date)
[1103]	567	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	568	printf "\| \| \n"
	569	printf "\| values of parameters/options: \| \n"
[1103]	570	spalte1=$(echo local_host$punkte \| cut -c-20)
	571	spalte2=$punkte$local_host
	572	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	573	spalte1=$(echo remote_host$punkte \| cut -c-20)
	574	spalte2=$punkte$remote_host
	575	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	576	spalte1=$(echo queue$punkte \| cut -c-20)
	577	spalte2=$punkte$queue
	578	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	579	spalte1=$(echo memory$punkte \| cut -c-20)
	580	spalte2="$punkte$memory mb"
	581	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	582	spalte1=$(echo cputime$punkte \| cut -c-20)
	583	spalte2="$punkte$cputime sec"
	584	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	585	spalte1=$(echo job_name$punkte \| cut -c-20)
	586	spalte2="$punkte$job_name"
	587	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	588	printf "#--------------------------------------------------------------# \n\n"
	589
	590
[1289]	591	# QUERY CHECK
[1]	592	antwort="dummy"
	593	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	594	do
	595	read antwort?" >>> continue (y/n) ? "
	596	done
	597	if [[ $antwort = n \|\| $antwort = N ]]
	598	then
	599	locat=verify; exit
	600	fi
	601	printf "\n"
	602	fi
	603
	604
	605
[1289]	606	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	607	identifier=$RANDOM
	608	job_on_remhost=${job_name}_${identifier}_$local_host
	609	job_to_send=job_to_send_$identifier
[1]	610	if [[ $delete_dayfile = false ]]
	611	then
[1289]	612	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	613	local_dayfile=${remote_host}_${job_name}
	614	else
	615	remote_dayfile=/dev/null
	616	fi
	617
	618
	619
[1289]	620	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	621	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	622	then
	623
[1289]	624	# GENERAL LOADLEVELER SETTINGS
[622]	625	execute_in_shell="#!/bin/ksh"
	626	use_shell="# @ shell = /bin/ksh"
	627	consumable_memory="ConsumableMemory($memory mb)"
	628	class="# @ class = $queue"
	629	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	630	network_to_use="# @ network.mpi = sn_all,shared,us"
	631	data_limit="# @ data_limit = 1.76gb"
	632	image_size="# @ image_size = 50"
[693]	633	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	634
[693]	635	if [[ $email_notification = none ]]
	636	then
	637	notify_user=""
	638	else
	639	notify_user="# @ notify_user = $email_notification"
	640	if [[ $delete_dayfile = true ]]
	641	then
	642	notification='# @ notification = never'
	643	fi
	644	fi
[622]	645
[312]	646	if [[ $remote_host = ibmh ]]
[1]	647	then
[312]	648	data_limit=""
	649	network_to_use=""
[1184]	650	class="# @ class = $queue"
[312]	651	environment=""
[814]	652	rset="# @ rset = RSET_MCM_AFFINITY"
	653	task_affinity="# @ task_affinity = core(1)"
[693]	654	elif [[ $remote_host = ibmkisti ]]
	655	then
	656	network_to_use="# @ network.MPI = sn_all,shared,US"
	657	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	658	if [[ $threads_per_task = 1 ]]
	659	then
	660	rset="# @ rset = RSET_MCM_AFFINITY"
	661	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	662	fi
[693]	663	environment=""
	664	use_shell=""
	665	data_limit=""
	666	image_size=""
[622]	667	elif [[ $remote_host = ibmku ]]
	668	then
	669	execute_in_shell="#!/usr/bin/ksh"
	670	use_shell="# @ shell = /usr/bin/ksh"
	671	consumable_memory=""
	672	environment=""
	673	network_to_use="# @ network.mpi = sn_all,shared,us"
	674	data_limit=""
	675	image_size=""
	676	elif [[ $remote_host = ibms ]]
	677	then
	678	network_to_use="# @ network.mpi = csss,shared,us"
[1]	679	fi
	680
	681	cat > $job_to_send << %%END%%
[622]	682	$execute_in_shell
	683	$use_shell
[1]	684
	685	# @ job_type = parallel
[693]	686	# @ job_name = $job_name
[1]	687	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	688	# @ output = $remote_dayfile
	689	# @ error = $remote_dayfile
[693]	690	$wall_clock_limit
[622]	691	$image_size
[312]	692	$class
	693	$environment
[1]	694	$network_to_use
	695	$data_limit
[693]	696	$rset
	697	$mcm_affinity_options
[814]	698	$task_affinity
[1]	699	$notification
[693]	700	$notify_user
[1]	701
	702	%%END%%
	703
	704	if (( nodes > 0 ))
	705	then
	706
[693]	707	if [[ $remote_host != ibmkisti ]]
	708	then
	709
	710	cat >> $job_to_send << %%END%%
[1]	711	# @ node = $nodes
[622]	712	# @ tasks_per_node = $processes_per_node
[1]	713	# @ node_usage = $node_usage
	714	# @ queue
	715
	716	%%END%%
	717
[693]	718	else
	719
	720	cat >> $job_to_send << %%END%%
[696]	721	# @ total_tasks = $mpi_tasks
[693]	722	# @ blocking = unlimited
	723	# @ queue
	724
	725	%%END%%
	726
	727	fi
	728
[1]	729	else
	730
[1289]	731	cat >> $job_to_send << %%END%%
[1]	732	# @ blocking = unlimited
	733	# @ total_tasks = $numprocs
	734	# @ node_usage = $node_usage
	735	# @ queue
	736
	737	%%END%%
	738
	739	fi
	740
[1289]	741	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	742	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	743	then
	744	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	745	fi
	746
[1]	747	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	748	then
	749
	750	cat > $job_to_send << %%END%%
	751	#!/bin/ksh
	752
	753	# @ job_type = serial
	754	# @ node_usage = $node_usage
	755	# @ job_name = palm
	756	# @ wall_clock_limit = ${timestring},$timestring
	757	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	758	# @ output = $remote_dayfile
	759	# @ error = $remote_dayfile
[312]	760	$class
[1]	761	$notification
	762
	763	# @ queue
	764
	765	%%END%%
	766
[1274]	767	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	768	then
	769
	770	if [[ $numprocs != 0 ]]
	771	then
	772	cat > $job_to_send << %%END%%
[1255]	773	#!/bin/bash
[1224]	774	#PBS -N $job_name
	775	#PBS -l walltime=$timestring
[1264]	776	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	777	#PBS -o $remote_dayfile
	778	#PBS -j oe
	779	#PBS -q $queue
	780
	781	$module_calls
	782
	783	%%END%%
	784
	785	else
	786
	787	continue
	788
	789	fi
	790
[1197]	791	elif [[ $remote_host = lccrayf ]]
	792	then
	793
	794	if [[ $numprocs != 0 ]]
	795	then
	796	cat > $job_to_send << %%END%%
	797	#!/bin/bash -l
	798	#SBATCH -J $job_name
	799	#SBATCH -t $timestring
	800	#SBATCH -N $nodes
	801	#SBATCH --ntasks-per-node=$processes_per_node
	802	#SBATCH -p $queue
	803	#SBATCH -o $remote_dayfile
	804	#SBATCH -e $remote_dayfile
	805
	806	$init_cmds
	807	$module_calls
	808
	809	%%END%%
	810
	811	else
	812	cat > $job_to_send << %%END%%
	813	#!/bin/bash -l
	814	#SBATCH -J $job_name
	815	#SBATCH -t $timestring
	816	#SBATCH -l ncpus=1
	817	#SBATCH -l pmem=${memory}mb
	818	#SBATCH -m abe
	819	#SBATCH -o $remote_dayfile
	820	#SBATCH -e $remote_dayfile
	821
	822	$init_cmds
	823	$module_calls
	824
	825	%%END%%
	826
	827	fi
	828
[1099]	829	elif [[ $remote_host = lcflow ]]
	830	then
[1202]	831	if [ $memory -gt 1800 ]; then
[1099]	832	use_himem=""
	833	else
	834	use_himem="#"
	835	fi
	836	if [[ $numprocs != 0 ]]
	837	then
[1202]	838	pe_set="#$ -pe impi41 $numprocs"
[1099]	839	else
[1202]	840	pe_set="#$ -pe impi41 1"
[1099]	841	fi
	842	if [[ $queue = default ]]
	843	then
	844	queue_set=""
	845	else
	846	queue_set="#$ -q $queue"
	847	fi
	848	[[ "$disc_space" = "" ]] && disc_space=50
	849
	850	cat > $job_to_send << %%END%%
	851	#!/bin/bash
	852	#$ -S /bin/bash
	853	#$ -N $job_name
	854	#$ -cwd
	855	#$ -l h_rt=$timestring
	856	#$ -l h_vmem=${memory}M
	857	#$ -o $remote_dayfile
	858	#$ -j y
	859	$pe_set
	860	#$ -R y
	861	${use_himem}#$ -l highmem=true
	862	#$ -l h_fsize=${disc_space}G
	863	$queue_set
	864
	865	%%END%%
	866
[1021]	867	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	868	then
	869
	870	if [[ $numprocs != 0 ]]
	871	then
	872	cat > $job_to_send << %%END%%
	873	#!/bin/ksh
	874	#PBS -N $job_name
	875	#PBS -l walltime=$timestring
	876	#PBS -l ncpus=$numprocs
	877	#PBS -l pmem=${memory}mb
	878	#PBS -o $remote_dayfile
[1021]	879	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	880	#PBS -j oe
	881
	882	mpd &
	883
	884	%%END%%
	885
	886	else
	887	cat > $job_to_send << %%END%%
	888	#!/bin/ksh
	889	#PBS -N $job_name
	890	#PBS -l walltime=$timestring
	891	#PBS -l ncpus=1
	892	#PBS -l pmem=${memory}mb
	893	#PBS -o $remote_dayfile
	894	#PBS -j oe
	895
	896	%%END%%
	897
	898	fi
	899
[1040]	900	elif [[ $remote_host = lckiaps ]]
	901	then
	902
	903	if [[ $numprocs != 0 ]]
	904	then
	905	cat > $job_to_send << %%END%%
	906	#!/bin/ksh
	907	#PBS -N $job_name
	908	#PBS -l walltime=$timestring
	909	#PBS -l select=1:ncpus=$numprocs
	910	#PBS -l pmem=${memory}mb
	911	#PBS -q $queue
	912	#PBS -o $remote_dayfile
	913	#PBS -j oe
	914	#PBS -V
	915
	916	%%END%%
	917
	918	else
	919	cat > $job_to_send << %%END%%
	920	#!/bin/ksh
	921	#PBS -N $job_name
	922	#PBS -l walltime=$timestring
	923	#PBS -l ncpus=1
	924	#PBS -l pmem=${memory}mb
	925	#PBS -o $remote_dayfile
	926	#PBS -j oe
	927
	928	%%END%%
	929
	930	fi
	931
[693]	932	elif [[ $remote_host = lcyon ]]
	933	then
	934
	935	if [[ $numprocs != 0 ]]
	936	then
	937	cat > $job_to_send << %%END%%
	938	#!/bin/ksh
	939	#PBS -N $job_name
	940	#PBS -l walltime=$timestring
	941	#PBS -l ncpus=$numprocs
	942	#PBS -l pmem=${memory}mb
	943	#PBS -o $remote_dayfile
	944	#PBS -j oe
	945
	946	%%END%%
	947
	948	else
	949	cat > $job_to_send << %%END%%
	950	#!/bin/ksh
	951	#PBS -N $job_name
	952	#PBS -l walltime=$timestring
	953	#PBS -l ncpus=1
	954	#PBS -l pmem=${memory}mb
	955	#PBS -o $remote_dayfile
	956	#PBS -j oe
	957
	958	%%END%%
	959
	960	fi
	961
[892]	962	elif [[ $remote_host = lcxe6 ]]
[164]	963	then
	964
	965	if [[ $numprocs != 0 ]]
	966	then
	967	cat > $job_to_send << %%END%%
[799]	968	#!/bin/ksh
[164]	969	#PBS -S /bin/ksh
	970	#PBS -N $job_name
[552]	971	#PBS -A $project_account
[206]	972	#PBS -j oe
[164]	973	#PBS -l walltime=$timestring
	974	#PBS -l mppwidth=${numprocs}
[622]	975	#PBS -l mppnppn=${processes_per_node}
[164]	976	#PBS -m abe
	977	#PBS -o $remote_dayfile
[492]	978	$email_directive
[164]	979
[892]	980	$init_cmds
[493]	981	$module_calls
[343]	982
[164]	983	%%END%%
	984
[108]	985	else
	986	cat > $job_to_send << %%END%%
	987	#!/bin/ksh
[168]	988	#PBS -S /bin/ksh
[108]	989	#PBS -N $job_name
[552]	990	#PBS -A $project_account
[206]	991	#PBS -j oe
[108]	992	#PBS -l walltime=$timestring
	993	#PBS -l ncpus=1
	994	#PBS -l pmem=${memory}mb
	995	#PBS -m abe
[492]	996	$email_directive
[108]	997	#PBS -o $remote_dayfile
	998
[892]	999	$init_cmds
[493]	1000	$module_calls
[343]	1001
[108]	1002	%%END%%
	1003
	1004	fi
	1005
[440]	1006	elif [[ $remote_host = lckyoto ]]
[437]	1007	then
	1008
[440]	1009	cat > $job_to_send << %%END%%
[799]	1010	#!/bin/ksh
[440]	1011	# @\$-o $remote_dayfile
	1012	# @\$-eo -oi
	1013	# @\$-lP 16
[799]	1014	# @\$-lp 1
[440]	1015	# @\$-lm 28gb -llm unlimited -ls unlimited
	1016	# @\$-q $queue
	1017	# @\$-Pvn abs_pack
	1018	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1019	#. /thin/local/etc/setprofile/intel-11.0.sh
	1020	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1021	. ~/.myprofile
	1022	#. /home2/t/t51254/palm/current_version/myprofile
	1023	#. /thin/apps/pgi/mpi.sh
	1024	#
	1025	env
	1026	#
	1027	set -x
	1028
	1029	%%END%%
	1030
	1031	elif [[ $remote_host = lcxt5m ]]
	1032	then
	1033
[437]	1034	if [[ $numprocs != 0 ]]
	1035	then
	1036	cat > $job_to_send << %%END%%
[799]	1037	#!/bin/ksh
[437]	1038	#PBS -S /bin/ksh
	1039	#PBS -N $job_name
	1040	#PBS -j oe
	1041	#PBS -l walltime=$timestring
	1042	#PBS -l mppwidth=${numprocs}
[622]	1043	#PBS -l mppnppn=${processes_per_node}
[437]	1044	#PBS -m abe
	1045	#PBS -o $remote_dayfile
	1046
[892]	1047	$init_cmds
[493]	1048	$module_calls
[437]	1049
	1050	%%END%%
	1051
	1052	else
	1053	cat > $job_to_send << %%END%%
	1054	#!/bin/ksh
	1055	#PBS -S /bin/ksh
	1056	#PBS -N $job_name
	1057	#PBS -j oe
	1058	#PBS -l walltime=$timestring
	1059	#PBS -l ncpus=1
	1060	#PBS -l pmem=${memory}mb
	1061	#PBS -m abe
	1062	#PBS -o $remote_dayfile
	1063
[892]	1064	$init_cmds
[493]	1065	$module_calls
[437]	1066
	1067	%%END%%
	1068
	1069	fi
	1070
[1099]	1071	elif [[ $remote_host = lckyuh ]]
	1072	then
	1073	cat > $job_to_send << %%END%%
	1074	#!/bin/bash
	1075	#PJM -L "rscgrp=$queue"
	1076	#PJM -L "node=$nodes"
	1077	#PJM --mpi "proc=$numprocs"
	1078	#PJM -L "elapse=$timestring"
	1079	#PJM -o $remote_dayfile
	1080	#PJM -j
	1081	#PJM -X
	1082	#PJM --no-stging
	1083
	1084	export LANG=en_US.UTF-8
	1085	%%END%%
	1086
[1090]	1087	elif [[ $remote_host = lckyut ]]
	1088	then
	1089	cat > $job_to_send << %%END%%
	1090	#!/bin/bash
[1099]	1091	#PJM -L "rscgrp=$queue"
[1090]	1092	#PJM -L "vnode=$numprocs"
	1093	#PJM -L "vnode-core=1"
	1094	#PJM -L "elapse=$timestring"
	1095	#PJM --mpi proc=$numprocs
	1096	#PJM -o $remote_dayfile
	1097	#PJM -j
[1099]	1098	#PJM -X
	1099	#PJM --no-stging
[1090]	1100
	1101	export LANG=en_US.UTF-8
	1102	%%END%%
	1103
[1]	1104	elif [[ $remote_host = nech ]]
	1105	then
	1106
	1107	if (( nodes > 1 ))
	1108	then
	1109	cat > $job_to_send << %%END%%
	1110	#!/bin/ksh
[622]	1111	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1112	#PBS -l ${qsubmem}=${Memory}gb
	1113	#PBS -b $nodes
	1114	#PBS -o $remote_dayfile
	1115	#PBS -N palm
	1116	#PBS -j o
	1117	#PBS -T mpisx
	1118
	1119	%%END%%
	1120
	1121	elif [[ $numprocs != 0 ]]
	1122	then
	1123	cat > $job_to_send << %%END%%
	1124	#!/bin/ksh
[622]	1125	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1126	#PBS -l ${qsubmem}=${Memory}gb
	1127	#PBS -o $remote_dayfile
	1128	#PBS -N palm
	1129	#PBS -j o
	1130
	1131	%%END%%
	1132
	1133	else
	1134	cat > $job_to_send << %%END%%
	1135	#!/bin/ksh
	1136	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1137	#PBS -o $remote_dayfile
	1138	#PBS -j o
	1139
	1140	%%END%%
	1141
	1142	fi
	1143
	1144	elif [[ $remote_host = lctit ]]
	1145	then
	1146	cat > $job_to_send << %%END%%
[635]	1147	#!/bin/ksh
[892]	1148	$init_cmds
[678]	1149	$module_calls
	1150
[1]	1151	%%END%%
	1152
[1289]	1153	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1154	if [[ $tasks_per_node != $processes_per_node ]]
	1155	then
[1378]	1156	submcom="$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 $queue "
[678]	1157	else
[1378]	1158	submcom="$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]	1159	fi
[1]	1160
	1161	else
	1162
	1163	cat > $job_to_send << %%END%%
	1164	# @\$-q ${queue}
	1165	# @\$-l${qsubtime} $timestring
	1166	# @\$-l${qsubmem} ${memory}mb
	1167	# @\$-o $remote_dayfile
	1168	# @\$-eo
	1169
	1170	%%END%%
	1171
	1172	fi
	1173
	1174
[1289]	1175	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1176	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1177	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1178	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1179	then
	1180	echo "set +vx" >> $job_to_send
	1181	echo "trap '" >> $job_to_send
	1182	echo "set +vx" >> $job_to_send
[1468]	1183	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]	1184	then
[622]	1185	if [[ $remote_host = ibmh ]]
[1]	1186	then
	1187	return_queue=c1
[693]	1188	elif [[ $remote_host = ibmkisti ]]
	1189	then
	1190	return_queue=class.1-2
[622]	1191	elif [[ $remote_host = ibmku ]]
	1192	then
	1193	return_queue=sdbg2
[1]	1194	elif [[ $remote_host = ibms ]]
	1195	then
	1196	return_queue=p_normal
[1274]	1197	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1198	then
	1199	return_queue=dataq
[1468]	1200	elif [[ $remote_host = lcxe6 ]]
	1201	then
	1202	return_queue=debug
[1040]	1203	elif [[ $remote_host = lckiaps ]]
	1204	then
	1205	return_queue=express
[1099]	1206	elif [[ $remote_host = lckyuh ]]
	1207	then
	1208	return_queue=cx-single
[1090]	1209	elif [[ $remote_host = lckyut ]]
	1210	then
	1211	return_queue=cx-single
[1]	1212	else
	1213	return_queue=unknown
	1214	fi
	1215
	1216	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1217	then
	1218
[622]	1219	if [[ $remote_host = ibmku ]]
	1220	then
[1289]	1221	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1222	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1223	else
[1289]	1224	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1225	fi
[1289]	1226	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1227	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1228	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1229	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1230	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1231	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1232	if [[ $host != "ibmh" ]]
	1233	then
[1289]	1234	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1235	fi
[1289]	1236	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1237	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1238
[1289]	1239	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1240	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1241
[1289]	1242	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
[1468]	1243	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]	1244	if [[ $remote_host = ibmku ]]
	1245	then
[1289]	1246	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1247	fi
[1289]	1248	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1249
	1250	elif [[ $remote_host = nech ]]
	1251	then
	1252	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1253	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1254	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1255	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1256	echo "#PBS -j o" >> $job_to_send
	1257	echo " " >> $job_to_send
	1258	echo "set -x" >> $job_to_send
	1259	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1468]	1260	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1261	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1262	echo "%%END%%" >> $job_to_send
	1263
[1099]	1264	elif [[ $remote_host = lckyuh ]]
	1265	then
[1289]	1266	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1267	echo "#!/bin/bash" >> $job_to_send
	1268	echo "#PJM -L \"node=1\"" >> $job_to_send
	1269	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1270	echo "#PJM --no-stging" >> $job_to_send
	1271	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1272	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1273	echo "#PJM -j" >> $job_to_send
	1274	echo " " >> $job_to_send
	1275	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1276	echo "set -x" >> $job_to_send
[1468]	1277	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1099]	1278	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1279	echo "%%END%%" >> $job_to_send
	1280
[1090]	1281	elif [[ $remote_host = lckyut ]]
	1282	then
[1289]	1283	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1284	echo "#!/bin/bash" >> $job_to_send
	1285	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1286	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1287	echo "#PJM --no-stging" >> $job_to_send
	1288	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1289	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1290	echo "#PJM -j" >> $job_to_send
	1291	echo " " >> $job_to_send
	1292	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1293	echo "set -x" >> $job_to_send
[1468]	1294	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1295	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1296	echo "%%END%%" >> $job_to_send
	1297
[1274]	1298	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1299	then
[1289]	1300	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1301	echo "#!/bin/bash" >> $job_to_send
	1302	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1303	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1304	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1305	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1306	echo "#PBS -j oe" >> $job_to_send
	1307	echo " " >> $job_to_send
	1308	echo "set -x" >> $job_to_send
[1468]	1309	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1255]	1310	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1311	echo "%%END%%" >> $job_to_send
	1312
[1099]	1313	elif [[ $remote_host = lcflow ]]
	1314	then
[1289]	1315	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1316	echo "#!/bin/bash" >> $job_to_send
	1317	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1318	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1319	echo "SGEPREFIX -cwd" >> $job_to_send
	1320	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1321	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
	1322	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
[1099]	1323	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1324	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1325	echo " " >> $job_to_send
	1326	echo "set -x" >> $job_to_send
	1327	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1328	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1329	echo "" >> $job_to_send
[1468]	1330	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1099]	1331	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1332	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1333	echo "%%END%%" >> $job_to_send
[1289]	1334	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1335	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1468]	1336	elif [[ $remote_host = lcxe6 ]]
	1337	then
	1338	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
	1339	echo "#!/bin/ksh" >> $job_to_send
	1340	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1341	echo "#PBS -l walltime=00:30:00" >> $job_to_send
	1342	echo "#PBS -A $project_account" >> $job_to_send
	1343	echo "#PBS -l mppwidth=1" >> $job_to_send
	1344	echo "#PBS -l mppnppn=1" >> $job_to_send
	1345	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1346	echo "#PBS -j oe" >> $job_to_send
	1347	echo " " >> $job_to_send
	1348	echo "set -x" >> $job_to_send
	1349	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1350	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1351	echo "%%END%%" >> $job_to_send
[1]	1352	else
	1353
[1289]	1354	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1355	echo "# @\\\$-q $return_queue" >> $job_to_send
	1356	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1357	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1358	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1359	then
	1360	echo "# @\$-l mpp_p=0" >> $job_to_send
	1361	fi
	1362	echo '# @\$-lF 10mb' >> $job_to_send
	1363	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1364	echo '# @\\\$-eo' >> $job_to_send
	1365	echo " " >> $job_to_send
	1366	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1367	then
	1368	echo "set +vx" >> $job_to_send
	1369	echo ". .profile" >> $job_to_send
	1370	fi
	1371	echo "set -x" >> $job_to_send
[1468]	1372	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]	1373	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1374	echo "%%END%%" >> $job_to_send
	1375	fi
	1376
	1377	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1378	then
[1289]	1379	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1380	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1381	then
[1289]	1382	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1383	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1384	then
[1289]	1385	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1386	elif [[ $remote_host = t3es ]]
	1387	then
[1289]	1388	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1389	elif [[ $remote_host = lckiaps ]]
	1390	then
[1289]	1391	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
[1468]	1392	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
[1289]	1393	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1394	elif [[ $remote_host = lckyu* ]]
[1090]	1395	then
[1468]	1396	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
	1397	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1398	elif [[ $remote_host = lcflow ]]
	1399	then
[1289]	1400	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
[1468]	1401	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1402	else
[1289]	1403	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1404	fi
[1043]	1405	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1406	then
[1289]	1407	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1408	fi
[1]	1409	if [[ $remote_host = nech ]]
	1410	then
	1411	echo "cd -" >> $job_to_send
	1412	fi
	1413	else
[1468]	1414	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1415	# ??? funktioniert das ÃŒberhaupt noch ???
[1468]	1416	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
[1]	1417	fi
	1418	echo "set -x" >> $job_to_send
	1419	echo " ' exit" >> $job_to_send
	1420	echo "set -x" >> $job_to_send
	1421	fi
	1422
	1423
	1424
[1289]	1425	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1426	cat $file_to_send >> $job_to_send
[69]	1427
[1]	1428	if [[ $remote_host = ibm ]]
	1429	then
	1430	echo " " >> $job_to_send
	1431	echo "exit" >> $job_to_send
	1432	fi
[635]	1433
[1289]	1434	# REMOVE JOB-FILE
[1099]	1435	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1436	then
	1437	echo " " >> $job_to_send
	1438	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1439	fi
[1]	1440
	1441
[1289]	1442	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1443	if [[ $no_submit = false ]]
	1444	then
	1445	if [[ $remote_host != $local_host ]]
	1446	then
	1447	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1448	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1449	then
[82]	1450	job_catalog_save=$job_catalog
	1451	job_catalog=job_queue
	1452	elif [[ $remote_host = nech ]]
	1453	then
	1454	job_catalog_save=$job_catalog
	1455	job_catalog=/hpf/b/${remote_user}/job_queue
	1456	fi
[1096]	1457	if [[ $remote_host = nech ]]
[82]	1458	then
[1289]	1459	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1460	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1461	else
[1468]	1462	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
[1]	1463	fi
[82]	1464	if [[ $? = 1 ]]
	1465	then
	1466	locat=scp; exit
	1467	fi
	1468	if [[ $remote_host = ibms ]]
	1469	then
	1470	job_catalog=$job_catalog_save
	1471	fi
[1]	1472	[[ $verify = true ]] && printf "\n >>> finished\n"
	1473	else
	1474	eval job_catalog=$job_catalog
	1475	cp $job_to_send ${job_catalog}/$job_on_remhost
	1476	fi
	1477
	1478
	1479
[1289]	1480	# START NQS- / LOADLEVELER-JOB
[1]	1481	if [[ $remote_host != $local_host ]]
	1482	then
	1483	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1484
[1289]	1485	if [[ $remote_host = ibmku ]]
[1]	1486	then
[1468]	1487	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1488	elif [[ $remote_host = lcflow ]]
	1489	then
[1468]	1490	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1491	else
[1468]	1492	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1493	fi
[1]	1494
	1495	[[ $verify = true ]] && printf " >>> o.k.\n"
	1496	else
	1497	cd $job_catalog
[1289]	1498	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1499	then
	1500	eval $submcom $job_on_remhost
[1289]	1501	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1502	then
[635]	1503	chmod u+x $job_on_remhost
[108]	1504	eval $submcom $job_on_remhost
[1]	1505	elif [[ $local_host = nech ]]
	1506	then
	1507	if [[ $queue = default ]]
	1508	then
[799]	1509	eval $submcom $job_on_remhost
[1]	1510	else
[799]	1511	eval $submcom -q $queue $job_on_remhost
[1]	1512	fi
	1513	else
	1514	qsub $job_on_remhost
	1515	fi
[622]	1516
[1289]	1517	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1518	# AT THE END OF THE JOB
[1099]	1519	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1520	then
	1521	rm $job_on_remhost
	1522	fi
[1]	1523	cd - > /dev/null
	1524	fi
	1525	fi
	1526
	1527
	1528
[1289]	1529	# FINAL ACTIONS
[1]	1530	if [[ $no_submit = false ]]
	1531	then
[1099]	1532	rm -f $job_to_send
[1]	1533	fi
[1266]	1534	[[ $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 |